Everything you wanted to ask, answered.

Montaj is built for founders and managing partners of established professional service firms, typically £1m to £10m in revenue, in sectors like finance, legal, recruitment and healthcare. Our clients are excellent at what they do but they often lose 15 to 25% of billable capacity to manual admin. If you want a private AI Operating System that reclaims those hours, protects margins, and that you fully own and control, you are exactly who we build for.

We design and deploy a private AI Operating System: a firm-specific intelligence layer with role-based copilots for every seat, running on your own tenancy. You own the data, the models and the configuration outright. There's no vendor lock-in: we productise the capability so it becomes a permanent asset of your firm, not a dependency on us.

Your first automation is typically live within 30 days, with payback in 60 to 90 days. Across a 90-day sprint most firms recover 20 to 40% of admin time per employee. We surface quick wins in the first week to build momentum, then layer in the higher-ROI workflows from there.

Engagements start from £3,000 per month depending on scope. Every one is backed by our promise: a 3× return on your investment, or we keep working for free until you get there. We map and score every opportunity before a single build, so you understand the expected return before you commit.

Compliance is built in from the start, not bolted on at the end. The system runs on your firm's own tenancy, enforces your data policies by architecture, and keeps full audit trails. It is designed for SRA, FCA, ICO, and CQC environments.

No. We start with a fixed-scope 90-day implementation that delivers a working system and a measurable return, not an open-ended fee for activity. Many clients then choose an optional retainer so we run, maintain and extend the system, but only once the core system is already paying for itself.

M.A.P. is how we run every engagement: Map, Automate, Productise. We map your workflows and score every automation opportunity, automate the highest-ROI admin first, then productise the capability with training and governance so your team owns it. Diagnosis before prescription, always.

No. We automate the admin, not the expertise. The goal is to free your people from low-value manual work so they spend more time on the judgement, relationships and billable work only humans can do. Our goal is to amplify, not replace, your team.

Map, Automate, Productise. We map how your business runs, automate the work worth automating one step at a time, then productise it into a system your team owns and can extend.

The core sprint runs over about twelve weeks: weeks one to two for the Map, weeks three to four to agree the roadmap, then weeks five to twelve to build, refine and hand over.

Yes. Productise is the whole third stage. You keep the automations, the brain and the system, and your team can run and extend it without us.

No. We connect to the tools you already use rather than ripping them out, and build the automations on top of them. You keep the systems your team already knows.

Before we build anything, we present the priority matrix and the roadmap, and you approve it. It fixes the scope, and it's what our 3x ROI guarantee is built on.

Yes. We start by monetising the assets you already have: dormant leads, old data and cold lists, before scaling ad spend. That's a separate engine, the M.O.N.T.A.J. Method, and you can read about it on our lead-generation page.

With a free AI Readiness Audit. We look at where AI and automation would pay off in your business and point you to the right next step, whether that's the full sprint, consulting or training.

We map where your team's time goes, identify the highest-ROI tasks to automate, and give you a clear picture of what's worth doing first, whether or not we end up working together. It's a working session, not a sales pitch.

It depends on scope: the number of workflows we automate and how deep the build goes. On the call we size the opportunity against the hours and cost it would save, so you can see the ROI before committing to anything. Most engagements pay for themselves within 60 to 90 days.

No. Everything we build lives inside your own accounts and you own it outright: no vendor lock-in and no proprietary platform you depend on us for. We aim to productise ourselves out of the engagement.

The first automation is typically live within 30 days, with payback in 60 to 90 days. You'll surface quick wins in week one so momentum builds early, rather than waiting months for a big-bang launch.

Service-based firms where time is the product: professional services, agencies, healthcare, finance and similar. If your team loses hours to repetitive admin, there's usually a strong case for automation.

If it's a fit, we start by mapping your workflows and delivering a prioritised diagnostic and a 12-month roadmap: the diagnosis before any prescription. If it's not a fit, you still leave with a clear view of what's worth automating first.

We're based in Cuffley, just north of London, and work with clients remotely across the UK and beyond. The audit is a 30-minute video or phone call, so location is no barrier.

We focus on your P&L, not the tech. We don't implement AI for its own sake: we map your workflows to find exactly where margin leaks, build on your own tenancy so you own everything, and measure success in hours returned to your business, not software logins.

Yes. All five sessions are on the Montaj YouTube channel, free to watch in full, in the order we ran them.

AI Week was five live, one-hour AI training sessions for business owners and their teams, held on 11, 13 and 15 May 2026 at Sópers House in Cuffley. More than 250 people came across the three days, from complete beginners to daily AI users, and every session was filmed.

Yes. That's our AI Team Training: the same hands-on approach, run privately for your team and built around your real workflows. It's a full day, in person or remote, with a 30-day plan afterwards.

Tailored. We run a discovery call first and build the exercises around your team's real workflows and the tools you already use, so the day is about your work rather than a generic syllabus.

We handle mixed rooms. At AI Week the same sessions worked for complete beginners and daily AI users, because the afternoon is hands-on and meets people where they are.

Either. We run the day whichever way suits your team.

You get a 30-day implementation roadmap with two check-ins, a fortnight in and at month end, so the skills become habits. If you want ongoing support we offer a fractional head of AI retainer or hourly consulting, and if you'd rather we build the automations for you, that's our AI and Automation Implementation service.

It works well for a team of three or four, up to about fifteen.

Book a free AI Readiness Audit and we'll point you to the right first step, whether that's a quick win, a training day, or a build.

You keep your CRM. We connect to Vincere, Firefish, Loxo and the other systems you already use, rather than migrating you onto something new. We call it connect, don't migrate. Your data stays where it is.

It depends on what we build and how much help you want from our team. We can run a detailed audit that gives you the roadmap, train your team hands-on, or build out the whole system and hand it back to you. Our solutions page has the full breakdown.

No. We build it, document it and hand it over so your team can run it. There's nothing for you to maintain on your own unless you want to.

Yes. We architect for it from the start, on your own systems, with access scoped and controlled. Handling candidate and client data properly is part of the build, not an afterthought.

No. The relationship and judgement stay with your people. We take the repetitive admin off the desk so they spend more time on the work only they can do.

No. We're not a marketing agency and we're not competing with you. We build the AI and automation that runs your back office and delivery, so your team spends more time on the client work only they can do.

You keep your stack. We connect to HubSpot, HighLevel, Notion, Slack, Make, n8n and the other tools you already use. We call it connect, don't migrate. Your data stays where it is.

Yes. We build it in your template and your voice, and keep a human review step before anything reaches a client. The aim is to remove the manual pull, not the judgement.

Yes. We architect for it from the start, on your own systems, with access scoped and controlled and every automated action logged. Handling regulated financial and client data properly is part of the build, not an afterthought.

You keep your software. We connect to Xero, QuickBooks, Sage and your practice management tools rather than migrating you onto something new. We call it connect, don't migrate.

No. The judgement, the advice and the sign-off stay with your qualified people. We take the routine processing off their desks so they spend more time on advisory work.

Yes. Patient data is processed in a compliant environment, isn't used to train models, and every automated action is logged with an audit trail. We architect for it on your own systems from the start.

No. We automate the admin around care: comms, scheduling, documentation and billing. Clinical judgement stays entirely with your clinicians, who review anything the system drafts.

You keep your systems. We connect to your booking, records and comms tools rather than migrating you onto something new. We call it connect, don't migrate.

You keep your tools. We connect to your CRM, quoting and job-management software rather than migrating you onto something new. We call it connect, don't migrate.

No. We build it, document it and hand it over so your team can run it from the site office. There's nothing for you to maintain on your own unless you want to.

No. Quoting judgement, site work and client relationships stay with your people. We take the chasing, scheduling and paperwork off their plates.

That's normal, and it's where we start. Mapping how you actually quote and run jobs is the first thing we do, and it becomes the knowledge base the automation runs on.

Yes. We architect for it from the start, on your own systems, with access scoped and controlled and every action logged. Handling confidential client data properly is part of the build, not an afterthought.

No. Judgement, advice and sign-off stay with your qualified people, who review anything the system drafts. We take the admin and first-pass work off their desks.

We connect to Street.co.uk, Reapit, Alto and the other systems estate agents use, writing bookings and updates straight back. We call it connect, don't migrate: you keep your CRM.

Handled well, it does the opposite. An instant, useful reply beats a next-day callback, and it hands over to your team the moment a human is the right next step. Speed is what wins the viewing.

No. Viewings, valuations and relationships stay with your people. We make sure every lead gets an instant response and nothing slips, so your team spends time with the ones ready to move.

Research by McKinsey consistently shows knowledge workers spend around 28 hours per week on emails and internal communications alone, far more than their actual core job. Our figure of 15 hours on specifically automatable admin tasks is deliberately conservative. It accounts only for structured, repeatable tasks that a system could handle today. The true number for your business is almost certainly higher once you track it honestly.

Any task that follows a consistent pattern and involves moving or transforming information is a strong candidate. Data entry, document generation, email follow-ups, status reporting, invoice processing, appointment scheduling, CRM updates, and internal approvals workflows are the most common. Tasks that require genuine human judgment, creative thinking, or relationship-building should not be automated, and with good process design, you won't need to.

McKinsey's 2023 research on automation potential suggests that between 60% and 80% of current knowledge work activities could be automated with existing technology. We use 70% as a conservative working figure to avoid overpromising. Some businesses we work with recover more. The actual percentage for your business depends on how repeatable your specific workflows are and how cleanly your data is structured.

For most service businesses in the 5 to 50 staff range, the first wave of automation covers its implementation cost within 3 to 6 months. The highest-leverage automations (typically CRM data entry, report generation, and email follow-up sequences) can return several times their cost in the first year. Payback period depends on your hourly cost rate, hours saved per automation, and build complexity. Our free AI Implementation Audit calculates this for your specific workflows before any commitment is required.

Automation and AI are related but different. Most of the admin cost you've calculated can be addressed with workflow automation, no AI required. AI is an additional layer you can add to handle more complex, judgment-based tasks. Neither replaces your team. The goal is to remove the work your team least wants to do (the repetitive, low-judgment tasks), so they can spend more time on the work that actually differentiates your business. Businesses that automate well tend to grow their teams because they can serve more clients without proportional increases in overhead.

The best starting point is a workflow audit, not a technology decision. Before choosing any tool, map your three most time-consuming repeatable processes. For each one, note the trigger (what starts the task), the steps involved, and the output. Once you can see the pattern clearly, the right tools become obvious. Our free 60-minute AI Implementation Audit does exactly that, and delivers a written report with your top automation opportunities ranked by impact within 24 hours.

No. To keep the comparison clean, the calculator uses a flat salary cost across years 1 and 3. In practice, most employees receive salary increases of 2 to 5% annually, which would push the 3-year hire cost higher than shown here. Automation costs do not change unless you upgrade the tooling or expand the scope, so the real advantage of automating typically grows over time.

The calculator uses the employer NI rate of 13.8% on earnings above the secondary threshold (£9,100 per year for 2024/25), plus 3% pension on total salary. For USD, a 22% uplift approximates US payroll taxes and basic benefits. These are estimates rather than precise figures. Consult your accountant or HR team for an exact cost specific to your situation.

That is often the right answer, particularly for a mixed capacity gap. The recommended approach is to automate the admin overhead first, then assess whether a hire is still needed for the remaining delivery work. Many businesses find that automating 40 to 60% of admin means the remaining work can be absorbed by existing staff, or that a part-time hire at lower cost covers what is left.

For a focused suite covering the main admin workflows in a department (reporting, CRM updates, onboarding sequences, invoice processing), expect a build cost of between £3,000 and £8,000. Ongoing tool costs are typically £50 to £200 per month. Against a £35,000 salary, that means the automation pays for itself within two to four months of year one salary cost alone.

If you have a quote from an automation specialist, use that figure. If you are estimating, use £3,000 to £5,000 for a small business covering 2 to 3 core workflows, or £5,000 to £10,000 for a full department. A free AI Implementation Audit from Montaj Digital will give you a scoped cost estimate specific to your business within 48 hours.

That is exactly what these comparisons are for. Each one lays out the honest trade-offs on cost, speed, risk and ownership so you can see where you land. If you would rather not weigh it up alone, the free AI Readiness Audit maps your workflows and tells you which approach fits your situation, with no pitch attached.

You can, and for a handful of simple, low-stakes automations that is often the right call. The limits show up as you scale: brittle flows, no error handling, and a system only one person understands. We are happy to point you to what you can safely build yourself, and to take on the parts where a mistake is expensive.

Great, that usually makes projects faster, not redundant. A technical hire is brilliant at execution but rarely has the time or the cross-industry pattern library to design the whole system. Most firms get the best result when we set the architecture and hand your in-house person a system they can own and extend.

It happens often, and it is fixable. The usual failure is automating a process that was never properly mapped, so the tool amplifies a broken workflow. We can pick up a stalled DIY build, map what should have come first, and rebuild the parts that matter without you starting from scratch.

Both, chosen to fit the job rather than a preference. Where a no-code platform is reliable and cheaper to run we use it; where you need control, compliance or performance it cannot give, we build custom. Either way you own the result outright, with no lock-in to us.

Yes. Every written review, video testimonial and workshop reaction on this page is from a real client or a real attendee of our AI Week workshops. Our Google reviews are public and verifiable on our Google Business Profile.

If we've worked together, we'd love your feedback. Use the “Leave a Review” button at the top of this page: it opens our Google Business Profile so you can leave a rating and a few words in under a minute.

Often, yes. On a discovery call we can usually arrange a short reference conversation with a client in a similar field, so you can hear about the experience first-hand before committing to anything.

It varies by firm, but examples include going from a handful of leads a month to a pre-qualified pipeline, reactivating dormant data into six figures of warm deals, and reclaiming hundreds of admin hours a year. The video case studies on this page tell the specifics in each client's own words.

No. Most of our published stories come from financial services, healthcare and recruitment, but the approach applies to any professional service firm that runs on repeatable, admin-heavy workflows.

Not always for every seat. The roles on this page are the ones we grow into as the business scales, so we're not running open vacancies for all of them at once. If one of them is you, introduce yourself now and we'll come back to you first when it opens.

We're a UK team, remote-first, with a London-area HQ at Unit 105, Cuffley Place for the times when being in a room together helps.

Most of the team are full-time contractors, working remotely on their own kit. We pay properly for the level and review it as you grow, with a real path up as the client base scales. We'd rather be honest about where the business is than pretend to be a big corporate with a benefits brochure.

People who are genuinely brilliant at what they do, sweat the details, and want their work to count for more than a pay cheque. We're lean, so everyone has to be excellent and everyone touches every part of the work.

Introduce yourself through the section on this page. Tell us the role you'd fit and why Montaj, and it comes straight to us. If nothing's open for your role today, we'll keep you on file and come back to you first when it is.

Score each repetitive task on five weighted criteria: revenue impact, time saved, error and quality risk, feasibility, and repetitiveness. Multiply by the weights, total it out of 85, and sort descending. Anything scoring 70 or more is a build-now priority, 40 to 69 is worth planning, and under 40 can wait. The scoring forces you to compare the money and time behind each task rather than automating whatever felt most annoying that week.

Check the process is worth automating and is not broken. Run it through a simple gate first: does it make money, save time or money, or reduce risk? If it does none of those, do not automate it. Then confirm the underlying process actually works, because automating a broken process only makes the mess run faster. Fix the foundation, then score and build.

Usually not yet. Automation copies your current process and runs it faster and cheaper, so if the process is flawed you are simply reproducing the flaw at scale. Sort the process out first, get it to a standard you are happy with, and only then automate it. The exception is a task that is error-prone precisely because it is manual, like moving hundreds of figures by hand, where a machine is genuinely more reliable.

It is a weighted scoring tool for ranking which workflows to automate first. You list your repetitive tasks, score each from 1 to 10 on five criteria (revenue impact, time saved, error and quality risk, feasibility, repetitiveness), apply a weight to each, and total the result out of 85. A red-amber-green band then tells you what to build now, plan, or park. It turns a vague wish list into an ordered plan with a financial case behind each item.

Tasks that rely on human judgement, nuance, or a personal relationship, where handing the work to AI would drop the quality noticeably or damage trust. A good example is chasing invoices for a low-volume, high-value client you have known for years. If automating something takes your output from a 9 to a 5, the quality risk alone should veto it, however much time it would save.

For most service businesses, off-the-shelf tools cover it: Zapier to start, Make as you grow, and n8n when you need real control, plus GoHighLevel for CRM and a model like Claude where genuine language work is involved. You rarely need a custom-built AI model. In the scoring test, the easier a task is to build with these existing tools, the higher it scores on feasibility.

Usually it isn't the ads, it's the system around them. The five most common reasons are: targeting through audience settings instead of creative, sending cold traffic to a homepage rather than a dedicated lead capture, following up too slowly, running a generic offer that blends in with every other firm, and tracking vanity metrics like clicks and cost per lead instead of the full funnel through to revenue. Fix those and the same ad spend performs very differently.

Send it to a dynamic lead capture built around the specific intent of the ad. A homepage builds credibility but tries to do too many things at once, so cold prospects with short attention spans leave. A focused lead capture asks the right qualifying questions, adapts to the prospect's answers, and delivers something concrete like a diagnostic or a custom report. Done well, these convert well above the 2 to 5% industry average for financial services.

In minutes, not hours. High-intent interest decays fast, and waiting 24 to 72 hours is how good leads go cold. A basic follow-up system fires an instant confirmation with clear next steps, runs a short five to seven touch nurture over 7 to 14 days that reframes the cost of inaction and handles objections, then adds a light qualification step before booking. Ads create attention; follow-up is what turns it into a conversation.

Structured clarity. Safe language like trusted, tailored and holistic is indistinguishable from every competitor, so it converts poorly. A strong offer is specific about who it's for, clear on the problem it solves, honest about the process, and framed around outcomes without over-promising, which matters in a regulated market. Diagnostics, benchmarks and assessments work well because they give the prospect something concrete before asking for a commitment.

Track the full funnel, not vanity metrics. Clicks, impressions and cost per lead look good on reports but don't tell you whether leads convert. The numbers that matter are cost per lead, lead-to-appointment rate, appointment show-up rate, appointment-to-close rate, cost per client, and return on ad spend. Together they show exactly where the system is breaking, so every optimisation is informed rather than a guess.

Paid ads tend to make sense for high-ticket financial services firms doing over roughly half a million a year in revenue that want predictable growth without depending entirely on the founder. Below that, or with a weak offer and no follow-up system, ads usually just expose the gaps faster. The sensible first step isn't more ad spend, it's a diagnosis of the current acquisition system to find what to fix first.

An AI opportunity audit is a structured review of how a business spends its week, followed by a map of which tasks AI or automation could do better, cheaper, or faster. It looks at the data you already own, your admin drains, and your workflows, then gives you a prioritised shortlist of what to automate first. The output is a plan, not a sales pitch.

Export the data you already own, your LinkedIn network, inbox, and CRM, and run it through an AI model with a clear description of your ideal client. The model reads far more than a person can and surfaces contacts and threads that have gone cold. In our live audit this dormant-data approach identified around £110k of pipeline the firm had forgotten it was sitting on.

It's identified pipeline, not banked revenue. The AI analysis estimated it could generate roughly £110,000 from opportunities inside the firm's own network, and they've started working through them. We're careful to call it a projection off owned data rather than money in the bank. A separate recruitment firm we ran the same inbox play for signed around £60,000 in deals, and that figure is confirmed.

The highest-value uses are mining your own network and CRM for dormant opportunities, scoring candidates against a defined matrix instead of skim-reading, automating CV formatting, running structured outreach, and re-engaging cold contacts. Used well, AI in recruitment removes admin and hands time back to the human part of the job: building relationships and placing people.

Less than most people expect. In this audit, roughly 80% of a £40,000 job description could be handled for under £400 a month in software. Individual tools typically run from around £8 to £100 a month each. The point isn't the cost of the tools, it's the swap: a fraction of a salary for work that used to need a person.

No. Most CRMs, including the one this firm used, have an API connector that lets your CRM talk to an AI model without migrating anything. You keep the system you know and add a layer on top that can read your data, score it, and answer questions about it. Replacing the CRM is rarely the right first move.

An AI-first company is one where AI is part of how work is designed, measured and improved, rather than something a few people reach for when they remember. Every tool has a clear job, an owner and a success metric, the knowledge is packaged so it plugs into any model, and improvement is tracked against a road map. The payoff is two to five times the revenue per head from the same team, but only for the businesses with proper systems behind it.

An AI-using company uses AI when it remembers. The team is scattered across ChatGPT, Claude and Gemini with no agreed use case, the knowledge lives in personal chats, outputs vary from person to person, and nothing is owned or measured. An AI-first company has decided where AI goes and why, packages its knowledge for reuse, gives every tool a job and an owner, and measures whether any of it's working. The gap isn't the software, both have the same models. It's the system underneath.

Strategy, workflows, tools, knowledge and adoption, in that order. Strategy finds where the business is losing margin, skilled time and speed. Workflows sort each task into cruise, co-pilot or human. Tools are hired for specific jobs rather than collected. Knowledge is packaged into a company brain any model can read. Adoption gets the team using it, led by an owner, with a review cadence and a success metric. Buy tools before you have a strategy and you just collect subscriptions you never use.

A company brain is a single, structured store of everything that makes your business itself: your processes, your clients, your one-page strategic plan, your ideal customer, your tone of voice. Mine is just a set of linked markdown files, and it plugs into any AI tool. It matters because the AI you use is only as good as the brain behind it: without one, a model gives you the average of everything ever written; with one, every output starts from your reality. Build it in a store you control and you're never tied to a single platform.

Because adoption is a management problem, not a technology one. Nobody owns the initiative, so there's no champion driving it. The team doesn't trust the outputs because the knowledge base and scoring are weak or missing. There's no review loop, so the first failure convinces everyone AI is useless. And success is never measured, so nobody can tell whether it's working. Fix those with clear ownership, a review cadence and a metric, and adoption follows.

No. If your first instinct with AI is to cut the team, you have probably got the wrong person in the role to begin with. The aim is to amplify the people you have by putting systems, processes and infrastructure behind them, so a pod that once capped out at twenty clients can carry three to five times more without burning out. That's where the revenue-per-head gain comes from: the tools free your people to do the work only they can do.

Because you can't automate a process you haven't defined, and automating a bad process just makes the mess run faster. AI is only as good as the input you give it, so mapping the workflow first, every step, handoff and decision, is how you make that input consistent. Businesses that map before they build get a far faster return, because the automation has something solid to work with.

A good process is structured, documented, consistent, scalable and constantly improving. That means a defined order of steps, written down outside anyone's head, producing the same output whoever runs it, still working at a hundred clients as well as ten, and getting a little better over time. A bad process is the reverse: unstructured, undocumented, inconsistent, unscalable, untrackable and static.

The five whys is a root-cause technique from Toyota. When something goes wrong, you ask why, then keep asking, usually about five times, until you reach the underlying cause rather than the symptom. A machine stopping might trace back through a blown fuse and poor lubrication to a missing filter. Fixing the symptom leaves the real problem in place; fixing the root cause solves it for good.

Score your processes on time saved, revenue impact, the risk of quality dropping, how repetitive the task is, and whether it's feasible with current tools. High-repetition, low-risk tasks are the best candidates. Anything where quality is likely to fall is usually a job to keep human. A practical shortcut is to start with the process that annoys you the most, because it tends to give the best return on the effort.

They describe how much of a task AI should handle. Cruise is a clear, repeatable task software can run on its own with light oversight. Co-pilot is a complex task where AI drafts and a person reviews before anything goes out. Captain is a consequential or highly complex task the human handles directly with the systems off. Sort each mapped step into one of the three, and always keep a human in the loop.

For the mapping itself, a whiteboard, pen and paper, or an online tool like Miro or Lucidchart. If drawing the boxes is the blocker, record yourself doing the task with Loom, hand the video to Gemini and ask it to write out the process, then turn that into a diagram. Claude can draw diagrams in the chat too. To build the automations once mapped, the platforms to look at are Make and n8n.

Because AI is already here and already doing a share of everyday knowledge work. More than half of recurring hours in a typical service business are automatable with tools everyone can now access, and people who use AI well are producing several times the output of those who do not. Ignoring it does not keep things as they were; it just means being out-competed by people and businesses who have folded it into how they work.

AI is taking tasks, not whole jobs, at least for now. The more useful question is which parts of your work are easy to explain, repeat, check or outsource, because those are the parts AI compresses first. The real risk is not AI itself but a person who uses AI well replacing someone who does not. You protect yourself by leaning into the work AI is bad at: complex, high-context, high-stakes and expert judgement.

Work that is genuinely complex, needs nuanced context from many sources, carries high stakes where mistakes are costly, or is hard to judge without real expertise. AI predicts the most likely answer from what already exists, so it struggles with novel situations, subtle judgement calls and anything where there is no tested best practice. Deep expertise, taste and the ability to prevent serious mistakes are what stay valuable.

It is a way to decide how much of a task to give AI, based on how planes are flown. Cruise is for repeatable, low-variance tasks software can run with light oversight. Co-pilot is for valuable, nuanced work where AI drafts and you review, like take-off and landing. Captain is for high-stakes work you keep in your own hands, such as pricing, negotiation or legal claims. Before using AI, check the human time, the odds of a good result, and the time to review and fix.

Use five habits: apply the flight deck check to decide AI's role in each task; amplify your workflows by building systems with context rather than doing everything by hand; curate taste by studying the best work in your field; learn to tell stories that make people care, using frameworks like and-but-therefore and SCQA; and think for yourself first before reaching for AI, so you keep your own judgement sharp.

Only if it changes your habits for the worse. Plato feared writing would erode memory, yet writing let humanity pass knowledge on. AI is the same: a tool that helps if you keep thinking, and harms if you outsource all your thinking to it. Forming your own view before you ask, making your own decisions, and challenging the machine's answers keeps your judgement strong while still getting the leverage.

There's no single best tool, because it depends on the job. The practical answer is to cover six buckets: an everyday chat model (ChatGPT, Gemini or Claude), a search and synthesis tool (Perplexity, NotebookLM or Poppy), a creative studio, an autonomous agent like Manus or Claude Code, a smart-assets tool such as Google Workspace or Gamma, and an automation tool like Zapier, Make or n8n. Hold one tool per bucket and you've covered most of what a business needs.

They're close on raw capability, so pick by superpower. ChatGPT is the most obedient and follows instructions to the letter. Gemini wins on modality, handling video, audio and large files natively, and it comes bundled with Google Workspace. Claude tends to produce the best first drafts of writing and code, and can build interactive tools inside the chat. Reach for the one that fits the task in front of you.

Fewer than the lists suggest. One tool per bucket, so around six, covers the vast majority of the work, and about ten tools handle roughly 90% of mine. Every new tool should have to beat the one already sitting in its bucket, and if you can't name the job it does, the person who owns it, and what success looks like at 30 days, don't adopt it.

Make it pass four questions before it earns a place: what recurring task does it do, who in the business owns and improves it, what output does it produce, and what does success look like at 30 days? Treat tools like hires with a defined job, not a collection to complete. Most run generous free plans, so you can test one properly before you pay.

No, and trying to is a common mistake. One of my most valuable workflows saves 16 hours a week and contains no AI at all: it just moves data between apps on a schedule. Use AI where genuine language or judgement is needed, and let plain automation handle repeatable, rule-based work. Forcing AI into everything tends to break things that already worked fine.

For quick research and fact-checking, Perplexity, because it pulls from live sources and cites every one, and its site operator lets you restrict a search to a single place like Reddit. For querying your own documents with no risk of made-up answers, NotebookLM, because it only answers from the sources you give it. They do different jobs: one searches the open web, the other stays inside your own material.

Start with small, repeatable admin, not a grand plan. The seven quick wins from the workshop are: record and transcribe every meeting, dictate to your AI instead of typing, connect your email so you can search and draft in seconds, set up reusable projects that hold your context, use pull prompting, turn your knowledge base into assets like SOPs and training, and build a personal assistant that briefs you each morning. Each takes minutes to set up and needs no technical skill.

Pull prompting is asking the AI to interview you before it does the work, instead of pushing it a request and hoping. Rather than "write me a proposal," you say "I need a proposal for a new client, before you start ask me any clarifying questions you need to do this well." It then asks its questions one at a time and builds from your answers, which are far better than its guesses.

Fewer than you think, and most are free. Fathom transcribes video calls and an app called Easy Recorder handles in-person ones. Wispr Flow is strong for dictation, though most everyday models now build it in. For the core work, one everyday model such as Claude, ChatGPT or Gemini does most of the heavy lifting, and NotebookLM turns your documents into something your team can query.

No. Every win here runs through settings menus and plain-English instructions, not code. Connecting your email is a few clicks under settings and integrations. Setting up a project means writing a paragraph about your business and uploading a few documents. Attendees who called themselves complete AI novices had these running the same week, without a developer or a new hire.

The average knowledge worker loses around 15 hours a week to admin, and McKinsey's data says more than half of that recurring work is automatable with tools that already exist. In one real example, a weekly reporting process that took my media buyer 8 to 16 hours went down to about 30 minutes of me reviewing the output once it was automated.

Score each process on five things: how much time it saves, its revenue impact, how often it goes wrong by hand, how rule-based it is, and how feasible it is with your current tools, then weight them so time saved and revenue count for more. A quicker rule of thumb is to start with the repetitive task that annoys you most, but never automate a broken process, because you'll only make the mess faster.

Montaj is built for founders and managing partners of established professional service firms, typically £1m to £10m in revenue, in sectors like finance, legal, recruitment and healthcare. Our clients are excellent at what they do but they often lose 15 to 25% of billable capacity to manual admin. If you want a private AI Operating System that reclaims those hours, protects margins, and that you fully own and control, you are exactly who we build for.

We design and deploy a private AI Operating System: a firm-specific intelligence layer with role-based copilots for every seat, running on your own tenancy. You own the data, the models and the configuration outright. There's no vendor lock-in: we productise the capability so it becomes a permanent asset of your firm, not a dependency on us.

Your first automation is typically live within 30 days, with payback in 60 to 90 days. Across a 90-day sprint most firms recover 20 to 40% of admin time per employee. We surface quick wins in the first week to build momentum, then layer in the higher-ROI workflows from there.

Engagements start from £3,000 per month depending on scope. Every one is backed by our promise: a 3× return on your investment, or we keep working for free until you get there. We map and score every opportunity before a single build, so you understand the expected return before you commit.

Compliance is built in from the start, not bolted on at the end. The system runs on your firm's own tenancy, enforces your data policies by architecture, and keeps full audit trails. It is designed for SRA, FCA, ICO, and CQC environments.

No. We start with a fixed-scope 90-day implementation that delivers a working system and a measurable return, not an open-ended fee for activity. Many clients then choose an optional retainer so we run, maintain and extend the system, but only once the core system is already paying for itself.

M.A.P. is how we run every engagement: Map, Automate, Productise. We map your workflows and score every automation opportunity, automate the highest-ROI admin first, then productise the capability with training and governance so your team owns it. Diagnosis before prescription, always.

No. We automate the admin, not the expertise. The goal is to free your people from low-value manual work so they spend more time on the judgement, relationships and billable work only humans can do. Our goal is to amplify, not replace, your team.

Map, Automate, Productise. We map how your business runs, automate the work worth automating one step at a time, then productise it into a system your team owns and can extend.

The core sprint runs over about twelve weeks: weeks one to two for the Map, weeks three to four to agree the roadmap, then weeks five to twelve to build, refine and hand over.

Yes. Productise is the whole third stage. You keep the automations, the brain and the system, and your team can run and extend it without us.

No. We connect to the tools you already use rather than ripping them out, and build the automations on top of them. You keep the systems your team already knows.

Before we build anything, we present the priority matrix and the roadmap, and you approve it. It fixes the scope, and it's what our 3x ROI guarantee is built on.

Yes. We start by monetising the assets you already have: dormant leads, old data and cold lists, before scaling ad spend. That's a separate engine, the M.O.N.T.A.J. Method, and you can read about it on our lead-generation page.

With a free AI Readiness Audit. We look at where AI and automation would pay off in your business and point you to the right next step, whether that's the full sprint, consulting or training.

We map where your team's time goes, identify the highest-ROI tasks to automate, and give you a clear picture of what's worth doing first, whether or not we end up working together. It's a working session, not a sales pitch.

It depends on scope: the number of workflows we automate and how deep the build goes. On the call we size the opportunity against the hours and cost it would save, so you can see the ROI before committing to anything. Most engagements pay for themselves within 60 to 90 days.

No. Everything we build lives inside your own accounts and you own it outright: no vendor lock-in and no proprietary platform you depend on us for. We aim to productise ourselves out of the engagement.

The first automation is typically live within 30 days, with payback in 60 to 90 days. You'll surface quick wins in week one so momentum builds early, rather than waiting months for a big-bang launch.

Service-based firms where time is the product: professional services, agencies, healthcare, finance and similar. If your team loses hours to repetitive admin, there's usually a strong case for automation.

If it's a fit, we start by mapping your workflows and delivering a prioritised diagnostic and a 12-month roadmap: the diagnosis before any prescription. If it's not a fit, you still leave with a clear view of what's worth automating first.

We're based in Cuffley, just north of London, and work with clients remotely across the UK and beyond. The audit is a 30-minute video or phone call, so location is no barrier.

We focus on your P&L, not the tech. We don't implement AI for its own sake: we map your workflows to find exactly where margin leaks, build on your own tenancy so you own everything, and measure success in hours returned to your business, not software logins.

Yes. All five sessions are on the Montaj YouTube channel, free to watch in full, in the order we ran them.

AI Week was five live, one-hour AI training sessions for business owners and their teams, held on 11, 13 and 15 May 2026 at Sópers House in Cuffley. More than 250 people came across the three days, from complete beginners to daily AI users, and every session was filmed.

Yes. That's our AI Team Training: the same hands-on approach, run privately for your team and built around your real workflows. It's a full day, in person or remote, with a 30-day plan afterwards.

Tailored. We run a discovery call first and build the exercises around your team's real workflows and the tools you already use, so the day is about your work rather than a generic syllabus.

We handle mixed rooms. At AI Week the same sessions worked for complete beginners and daily AI users, because the afternoon is hands-on and meets people where they are.

Either. We run the day whichever way suits your team.

You get a 30-day implementation roadmap with two check-ins, a fortnight in and at month end, so the skills become habits. If you want ongoing support we offer a fractional head of AI retainer or hourly consulting, and if you'd rather we build the automations for you, that's our AI and Automation Implementation service.

It works well for a team of three or four, up to about fifteen.

Book a free AI Readiness Audit and we'll point you to the right first step, whether that's a quick win, a training day, or a build.

You keep your CRM. We connect to Vincere, Firefish, Loxo and the other systems you already use, rather than migrating you onto something new. We call it connect, don't migrate. Your data stays where it is.

Our sprints run over 90 days: we map the business, build the priority automations, and hand over a system you own. You see the first working automations well before the end.

It depends on what we build and how much help you want from our team. We can run a detailed audit that gives you the roadmap, train your team hands-on, or build out the whole system and hand it back to you. Our solutions page has the full breakdown.

No. We build it, document it and hand it over so your team can run it. There's nothing for you to maintain on your own unless you want to.

Yes. We architect for it from the start, on your own systems, with access scoped and controlled. Handling candidate and client data properly is part of the build, not an afterthought.

No. The relationship and judgement stay with your people. We take the repetitive admin off the desk so they spend more time on the work only they can do.

No. We're not a marketing agency and we're not competing with you. We build the AI and automation that runs your back office and delivery, so your team spends more time on the client work only they can do.

You keep your stack. We connect to HubSpot, HighLevel, Notion, Slack, Make, n8n and the other tools you already use. We call it connect, don't migrate. Your data stays where it is.

Our sprints run over 90 days: we map the agency, build the priority automations, and hand over a system you own. You see the first working automations well before the end.

It depends on what we build and how much help you want from our team. We can run a detailed audit that gives you the roadmap, train your team hands-on, or build the whole system and hand it back. Our solutions page has the full breakdown.

No. We build it, document it and hand it over so your team can run it. There's nothing for you to maintain on your own unless you want to.

Yes. We build it in your template and your voice, and keep a human review step before anything reaches a client. The aim is to remove the manual pull, not the judgement.

Yes. We architect for it from the start, on your own systems, with access scoped and controlled and every automated action logged. Handling regulated financial and client data properly is part of the build, not an afterthought.

You keep your software. We connect to Xero, QuickBooks, Sage and your practice management tools rather than migrating you onto something new. We call it connect, don't migrate.

No. The judgement, the advice and the sign-off stay with your qualified people. We take the routine processing off their desks so they spend more time on advisory work.

Our sprints run over 90 days: we map the firm, build the priority automations, and hand over a system you own. You see the first working automations well before the end.

It depends on what we build and how much help you want from our team. We can run a detailed audit that gives you the roadmap, train your team hands-on, or build the whole system and hand it back. Our solutions page has the full breakdown.

No. We build it, document it and hand it over so your team can run it. There's nothing for you to maintain on your own unless you want to.

Yes. Patient data is processed in a compliant environment, isn't used to train models, and every automated action is logged with an audit trail. We architect for it on your own systems from the start.

No. We automate the admin around care: comms, scheduling, documentation and billing. Clinical judgement stays entirely with your clinicians, who review anything the system drafts.

You keep your systems. We connect to your booking, records and comms tools rather than migrating you onto something new. We call it connect, don't migrate.

Our sprints run over 90 days: we map the practice, build the priority automations, and hand over a system you own. You see the first working automations well before the end.

It depends on what we build and how much help you want from our team. We can run a detailed audit that gives you the roadmap, train your team hands-on, or build the whole system and hand it back. Our solutions page has the full breakdown.

No. We build it, document it and hand it over so your team can run it. There's nothing for you to maintain on your own unless you want to.

You keep your tools. We connect to your CRM, quoting and job-management software rather than migrating you onto something new. We call it connect, don't migrate.

No. We build it, document it and hand it over so your team can run it from the site office. There's nothing for you to maintain on your own unless you want to.

No. Quoting judgement, site work and client relationships stay with your people. We take the chasing, scheduling and paperwork off their plates.

That's normal, and it's where we start. Mapping how you actually quote and run jobs is the first thing we do, and it becomes the knowledge base the automation runs on.

Our sprints run over 90 days: we map the business, build the priority automations, and hand over a system you own. You see the first working automations well before the end.

It depends on what we build and how much help you want from our team. We can run a detailed audit that gives you the roadmap, train your team hands-on, or build the whole system and hand it back. Our solutions page has the full breakdown.

Yes. We architect for it from the start, on your own systems, with access scoped and controlled and every action logged. Handling confidential client data properly is part of the build, not an afterthought.

No. Judgement, advice and sign-off stay with your qualified people, who review anything the system drafts. We take the admin and first-pass work off their desks.

You keep your systems. We connect to your practice-management, document and CRM tools rather than migrating you onto something new. We call it connect, don't migrate.

Our sprints run over 90 days: we map the firm, build the priority automations, and hand over a system you own. You see the first working automations well before the end.

It depends on what we build and how much help you want from our team. We can run a detailed audit that gives you the roadmap, train your team hands-on, or build the whole system and hand it back. Our solutions page has the full breakdown.

No. We build it, document it and hand it over so your team can run it. There's nothing for you to maintain on your own unless you want to.

We connect to Street.co.uk, Reapit, Alto and the other systems estate agents use, writing bookings and updates straight back. We call it connect, don't migrate: you keep your CRM.

Handled well, it does the opposite. An instant, useful reply beats a next-day callback, and it hands over to your team the moment a human is the right next step. Speed is what wins the viewing.

No. Viewings, valuations and relationships stay with your people. We make sure every lead gets an instant response and nothing slips, so your team spends time with the ones ready to move.

Our sprints run over 90 days: we map the agency, build the priority automations, and hand over a system you own. You see the first working automations well before the end.

It depends on what we build and how much help you want from our team. We can run a detailed audit that gives you the roadmap, train your team hands-on, or build the whole system and hand it back. Our solutions page has the full breakdown.

No. We build it, document it and hand it over so your team can run it. There's nothing for you to maintain on your own unless you want to.

Research by McKinsey consistently shows knowledge workers spend around 28 hours per week on emails and internal communications alone, far more than their actual core job. Our figure of 15 hours on specifically automatable admin tasks is deliberately conservative. It accounts only for structured, repeatable tasks that a system could handle today. The true number for your business is almost certainly higher once you track it honestly.

Any task that follows a consistent pattern and involves moving or transforming information is a strong candidate. Data entry, document generation, email follow-ups, status reporting, invoice processing, appointment scheduling, CRM updates, and internal approvals workflows are the most common. Tasks that require genuine human judgment, creative thinking, or relationship-building should not be automated, and with good process design, you won't need to.

McKinsey's 2023 research on automation potential suggests that between 60% and 80% of current knowledge work activities could be automated with existing technology. We use 70% as a conservative working figure to avoid overpromising. Some businesses we work with recover more. The actual percentage for your business depends on how repeatable your specific workflows are and how cleanly your data is structured.

For most service businesses in the 5 to 50 staff range, the first wave of automation covers its implementation cost within 3 to 6 months. The highest-leverage automations (typically CRM data entry, report generation, and email follow-up sequences) can return several times their cost in the first year. Payback period depends on your hourly cost rate, hours saved per automation, and build complexity. Our free AI Implementation Audit calculates this for your specific workflows before any commitment is required.

Automation and AI are related but different. Most of the admin cost you've calculated can be addressed with workflow automation, no AI required. AI is an additional layer you can add to handle more complex, judgment-based tasks. Neither replaces your team. The goal is to remove the work your team least wants to do (the repetitive, low-judgment tasks), so they can spend more time on the work that actually differentiates your business. Businesses that automate well tend to grow their teams because they can serve more clients without proportional increases in overhead.

The best starting point is a workflow audit, not a technology decision. Before choosing any tool, map your three most time-consuming repeatable processes. For each one, note the trigger (what starts the task), the steps involved, and the output. Once you can see the pattern clearly, the right tools become obvious. Our free 60-minute AI Implementation Audit does exactly that, and delivers a written report with your top automation opportunities ranked by impact within 24 hours.

No. To keep the comparison clean, the calculator uses a flat salary cost across years 1 and 3. In practice, most employees receive salary increases of 2 to 5% annually, which would push the 3-year hire cost higher than shown here. Automation costs do not change unless you upgrade the tooling or expand the scope, so the real advantage of automating typically grows over time.

The calculator uses the employer NI rate of 13.8% on earnings above the secondary threshold (£9,100 per year for 2024/25), plus 3% pension on total salary. For USD, a 22% uplift approximates US payroll taxes and basic benefits. These are estimates rather than precise figures. Consult your accountant or HR team for an exact cost specific to your situation.

That is often the right answer, particularly for a mixed capacity gap. The recommended approach is to automate the admin overhead first, then assess whether a hire is still needed for the remaining delivery work. Many businesses find that automating 40 to 60% of admin means the remaining work can be absorbed by existing staff, or that a part-time hire at lower cost covers what is left.

For a focused suite covering the main admin workflows in a department (reporting, CRM updates, onboarding sequences, invoice processing), expect a build cost of between £3,000 and £8,000. Ongoing tool costs are typically £50 to £200 per month. Against a £35,000 salary, that means the automation pays for itself within two to four months of year one salary cost alone.

If you have a quote from an automation specialist, use that figure. If you are estimating, use £3,000 to £5,000 for a small business covering 2 to 3 core workflows, or £5,000 to £10,000 for a full department. A free AI Implementation Audit from Montaj Digital will give you a scoped cost estimate specific to your business within 48 hours.

That is exactly what these comparisons are for. Each one lays out the honest trade-offs on cost, speed, risk and ownership so you can see where you land. If you would rather not weigh it up alone, the free AI Readiness Audit maps your workflows and tells you which approach fits your situation, with no pitch attached.

You can, and for a handful of simple, low-stakes automations that is often the right call. The limits show up as you scale: brittle flows, no error handling, and a system only one person understands. We are happy to point you to what you can safely build yourself, and to take on the parts where a mistake is expensive.

Great, that usually makes projects faster, not redundant. A technical hire is brilliant at execution but rarely has the time or the cross-industry pattern library to design the whole system. Most firms get the best result when we set the architecture and hand your in-house person a system they can own and extend.

It happens often, and it is fixable. The usual failure is automating a process that was never properly mapped, so the tool amplifies a broken workflow. We can pick up a stalled DIY build, map what should have come first, and rebuild the parts that matter without you starting from scratch.

Both, chosen to fit the job rather than a preference. Where a no-code platform is reliable and cheaper to run we use it; where you need control, compliance or performance it cannot give, we build custom. Either way you own the result outright, with no lock-in to us.

Yes. Every written review, video testimonial and workshop reaction on this page is from a real client or a real attendee of our AI Week workshops. Our Google reviews are public and verifiable on our Google Business Profile.

If we've worked together, we'd love your feedback. Use the “Leave a Review” button at the top of this page: it opens our Google Business Profile so you can leave a rating and a few words in under a minute.

Often, yes. On a discovery call we can usually arrange a short reference conversation with a client in a similar field, so you can hear about the experience first-hand before committing to anything.

It varies by firm, but examples include going from a handful of leads a month to a pre-qualified pipeline, reactivating dormant data into six figures of warm deals, and reclaiming hundreds of admin hours a year. The video case studies on this page tell the specifics in each client's own words.

No. Most of our published stories come from financial services, healthcare and recruitment, but the approach applies to any professional service firm that runs on repeatable, admin-heavy workflows.

Not always for every seat. The roles on this page are the ones we grow into as the business scales, so we're not running open vacancies for all of them at once. If one of them is you, introduce yourself now and we'll come back to you first when it opens.

We're a UK team, remote-first, with a London-area HQ at Unit 105, Cuffley Place for the times when being in a room together helps.

Most of the team are full-time contractors, working remotely on their own kit. We pay properly for the level and review it as you grow, with a real path up as the client base scales. We'd rather be honest about where the business is than pretend to be a big corporate with a benefits brochure.

People who are genuinely brilliant at what they do, sweat the details, and want their work to count for more than a pay cheque. We're lean, so everyone has to be excellent and everyone touches every part of the work.

Introduce yourself through the section on this page. Tell us the role you'd fit and why Montaj, and it comes straight to us. If nothing's open for your role today, we'll keep you on file and come back to you first when it is.

Score each repetitive task on five weighted criteria: revenue impact, time saved, error and quality risk, feasibility, and repetitiveness. Multiply by the weights, total it out of 85, and sort descending. Anything scoring 70 or more is a build-now priority, 40 to 69 is worth planning, and under 40 can wait. The scoring forces you to compare the money and time behind each task rather than automating whatever felt most annoying that week.

Check the process is worth automating and is not broken. Run it through a simple gate first: does it make money, save time or money, or reduce risk? If it does none of those, do not automate it. Then confirm the underlying process actually works, because automating a broken process only makes the mess run faster. Fix the foundation, then score and build.

Usually not yet. Automation copies your current process and runs it faster and cheaper, so if the process is flawed you are simply reproducing the flaw at scale. Sort the process out first, get it to a standard you are happy with, and only then automate it. The exception is a task that is error-prone precisely because it is manual, like moving hundreds of figures by hand, where a machine is genuinely more reliable.

It is a weighted scoring tool for ranking which workflows to automate first. You list your repetitive tasks, score each from 1 to 10 on five criteria (revenue impact, time saved, error and quality risk, feasibility, repetitiveness), apply a weight to each, and total the result out of 85. A red-amber-green band then tells you what to build now, plan, or park. It turns a vague wish list into an ordered plan with a financial case behind each item.

Tasks that rely on human judgement, nuance, or a personal relationship, where handing the work to AI would drop the quality noticeably or damage trust. A good example is chasing invoices for a low-volume, high-value client you have known for years. If automating something takes your output from a 9 to a 5, the quality risk alone should veto it, however much time it would save.

For most service businesses, off-the-shelf tools cover it: Zapier to start, Make as you grow, and n8n when you need real control, plus GoHighLevel for CRM and a model like Claude where genuine language work is involved. You rarely need a custom-built AI model. In the scoring test, the easier a task is to build with these existing tools, the higher it scores on feasibility.

Usually it isn't the ads, it's the system around them. The five most common reasons are: targeting through audience settings instead of creative, sending cold traffic to a homepage rather than a dedicated lead capture, following up too slowly, running a generic offer that blends in with every other firm, and tracking vanity metrics like clicks and cost per lead instead of the full funnel through to revenue. Fix those and the same ad spend performs very differently.

Send it to a dynamic lead capture built around the specific intent of the ad. A homepage builds credibility but tries to do too many things at once, so cold prospects with short attention spans leave. A focused lead capture asks the right qualifying questions, adapts to the prospect's answers, and delivers something concrete like a diagnostic or a custom report. Done well, these convert well above the 2 to 5% industry average for financial services.

In minutes, not hours. High-intent interest decays fast, and waiting 24 to 72 hours is how good leads go cold. A basic follow-up system fires an instant confirmation with clear next steps, runs a short five to seven touch nurture over 7 to 14 days that reframes the cost of inaction and handles objections, then adds a light qualification step before booking. Ads create attention; follow-up is what turns it into a conversation.

Structured clarity. Safe language like trusted, tailored and holistic is indistinguishable from every competitor, so it converts poorly. A strong offer is specific about who it's for, clear on the problem it solves, honest about the process, and framed around outcomes without over-promising, which matters in a regulated market. Diagnostics, benchmarks and assessments work well because they give the prospect something concrete before asking for a commitment.

Track the full funnel, not vanity metrics. Clicks, impressions and cost per lead look good on reports but don't tell you whether leads convert. The numbers that matter are cost per lead, lead-to-appointment rate, appointment show-up rate, appointment-to-close rate, cost per client, and return on ad spend. Together they show exactly where the system is breaking, so every optimisation is informed rather than a guess.

Paid ads tend to make sense for high-ticket financial services firms doing over roughly half a million a year in revenue that want predictable growth without depending entirely on the founder. Below that, or with a weak offer and no follow-up system, ads usually just expose the gaps faster. The sensible first step isn't more ad spend, it's a diagnosis of the current acquisition system to find what to fix first.

An AI opportunity audit is a structured review of how a business spends its week, followed by a map of which tasks AI or automation could do better, cheaper, or faster. It looks at the data you already own, your admin drains, and your workflows, then gives you a prioritised shortlist of what to automate first. The output is a plan, not a sales pitch.

Export the data you already own, your LinkedIn network, inbox, and CRM, and run it through an AI model with a clear description of your ideal client. The model reads far more than a person can and surfaces contacts and threads that have gone cold. In our live audit this dormant-data approach identified around £110k of pipeline the firm had forgotten it was sitting on.

It's identified pipeline, not banked revenue. The AI analysis estimated it could generate roughly £110,000 from opportunities inside the firm's own network, and they've started working through them. We're careful to call it a projection off owned data rather than money in the bank. A separate recruitment firm we ran the same inbox play for signed around £60,000 in deals, and that figure is confirmed.

The highest-value uses are mining your own network and CRM for dormant opportunities, scoring candidates against a defined matrix instead of skim-reading, automating CV formatting, running structured outreach, and re-engaging cold contacts. Used well, AI in recruitment removes admin and hands time back to the human part of the job: building relationships and placing people.

Less than most people expect. In this audit, roughly 80% of a £40,000 job description could be handled for under £400 a month in software. Individual tools typically run from around £8 to £100 a month each. The point isn't the cost of the tools, it's the swap: a fraction of a salary for work that used to need a person.

No. Most CRMs, including the one this firm used, have an API connector that lets your CRM talk to an AI model without migrating anything. You keep the system you know and add a layer on top that can read your data, score it, and answer questions about it. Replacing the CRM is rarely the right first move.

An AI-first company is one where AI is part of how work is designed, measured and improved, rather than something a few people reach for when they remember. Every tool has a clear job, an owner and a success metric, the knowledge is packaged so it plugs into any model, and improvement is tracked against a road map. The payoff is two to five times the revenue per head from the same team, but only for the businesses with proper systems behind it.

An AI-using company uses AI when it remembers. The team is scattered across ChatGPT, Claude and Gemini with no agreed use case, the knowledge lives in personal chats, outputs vary from person to person, and nothing is owned or measured. An AI-first company has decided where AI goes and why, packages its knowledge for reuse, gives every tool a job and an owner, and measures whether any of it's working. The gap isn't the software, both have the same models. It's the system underneath.

Strategy, workflows, tools, knowledge and adoption, in that order. Strategy finds where the business is losing margin, skilled time and speed. Workflows sort each task into cruise, co-pilot or human. Tools are hired for specific jobs rather than collected. Knowledge is packaged into a company brain any model can read. Adoption gets the team using it, led by an owner, with a review cadence and a success metric. Buy tools before you have a strategy and you just collect subscriptions you never use.

A company brain is a single, structured store of everything that makes your business itself: your processes, your clients, your one-page strategic plan, your ideal customer, your tone of voice. Mine is just a set of linked markdown files, and it plugs into any AI tool. It matters because the AI you use is only as good as the brain behind it: without one, a model gives you the average of everything ever written; with one, every output starts from your reality. Build it in a store you control and you're never tied to a single platform.

Because adoption is a management problem, not a technology one. Nobody owns the initiative, so there's no champion driving it. The team doesn't trust the outputs because the knowledge base and scoring are weak or missing. There's no review loop, so the first failure convinces everyone AI is useless. And success is never measured, so nobody can tell whether it's working. Fix those with clear ownership, a review cadence and a metric, and adoption follows.

No. If your first instinct with AI is to cut the team, you have probably got the wrong person in the role to begin with. The aim is to amplify the people you have by putting systems, processes and infrastructure behind them, so a pod that once capped out at twenty clients can carry three to five times more without burning out. That's where the revenue-per-head gain comes from: the tools free your people to do the work only they can do.

Because you can't automate a process you haven't defined, and automating a bad process just makes the mess run faster. AI is only as good as the input you give it, so mapping the workflow first, every step, handoff and decision, is how you make that input consistent. Businesses that map before they build get a far faster return, because the automation has something solid to work with.

A good process is structured, documented, consistent, scalable and constantly improving. That means a defined order of steps, written down outside anyone's head, producing the same output whoever runs it, still working at a hundred clients as well as ten, and getting a little better over time. A bad process is the reverse: unstructured, undocumented, inconsistent, unscalable, untrackable and static.

The five whys is a root-cause technique from Toyota. When something goes wrong, you ask why, then keep asking, usually about five times, until you reach the underlying cause rather than the symptom. A machine stopping might trace back through a blown fuse and poor lubrication to a missing filter. Fixing the symptom leaves the real problem in place; fixing the root cause solves it for good.

Score your processes on time saved, revenue impact, the risk of quality dropping, how repetitive the task is, and whether it's feasible with current tools. High-repetition, low-risk tasks are the best candidates. Anything where quality is likely to fall is usually a job to keep human. A practical shortcut is to start with the process that annoys you the most, because it tends to give the best return on the effort.

They describe how much of a task AI should handle. Cruise is a clear, repeatable task software can run on its own with light oversight. Co-pilot is a complex task where AI drafts and a person reviews before anything goes out. Captain is a consequential or highly complex task the human handles directly with the systems off. Sort each mapped step into one of the three, and always keep a human in the loop.

For the mapping itself, a whiteboard, pen and paper, or an online tool like Miro or Lucidchart. If drawing the boxes is the blocker, record yourself doing the task with Loom, hand the video to Gemini and ask it to write out the process, then turn that into a diagram. Claude can draw diagrams in the chat too. To build the automations once mapped, the platforms to look at are Make and n8n.

Because AI is already here and already doing a share of everyday knowledge work. More than half of recurring hours in a typical service business are automatable with tools everyone can now access, and people who use AI well are producing several times the output of those who do not. Ignoring it does not keep things as they were; it just means being out-competed by people and businesses who have folded it into how they work.

AI is taking tasks, not whole jobs, at least for now. The more useful question is which parts of your work are easy to explain, repeat, check or outsource, because those are the parts AI compresses first. The real risk is not AI itself but a person who uses AI well replacing someone who does not. You protect yourself by leaning into the work AI is bad at: complex, high-context, high-stakes and expert judgement.

Work that is genuinely complex, needs nuanced context from many sources, carries high stakes where mistakes are costly, or is hard to judge without real expertise. AI predicts the most likely answer from what already exists, so it struggles with novel situations, subtle judgement calls and anything where there is no tested best practice. Deep expertise, taste and the ability to prevent serious mistakes are what stay valuable.

It is a way to decide how much of a task to give AI, based on how planes are flown. Cruise is for repeatable, low-variance tasks software can run with light oversight. Co-pilot is for valuable, nuanced work where AI drafts and you review, like take-off and landing. Captain is for high-stakes work you keep in your own hands, such as pricing, negotiation or legal claims. Before using AI, check the human time, the odds of a good result, and the time to review and fix.

Use five habits: apply the flight deck check to decide AI's role in each task; amplify your workflows by building systems with context rather than doing everything by hand; curate taste by studying the best work in your field; learn to tell stories that make people care, using frameworks like and-but-therefore and SCQA; and think for yourself first before reaching for AI, so you keep your own judgement sharp.

Only if it changes your habits for the worse. Plato feared writing would erode memory, yet writing let humanity pass knowledge on. AI is the same: a tool that helps if you keep thinking, and harms if you outsource all your thinking to it. Forming your own view before you ask, making your own decisions, and challenging the machine's answers keeps your judgement strong while still getting the leverage.

There's no single best tool, because it depends on the job. The practical answer is to cover six buckets: an everyday chat model (ChatGPT, Gemini or Claude), a search and synthesis tool (Perplexity, NotebookLM or Poppy), a creative studio, an autonomous agent like Manus or Claude Code, a smart-assets tool such as Google Workspace or Gamma, and an automation tool like Zapier, Make or n8n. Hold one tool per bucket and you've covered most of what a business needs.

They're close on raw capability, so pick by superpower. ChatGPT is the most obedient and follows instructions to the letter. Gemini wins on modality, handling video, audio and large files natively, and it comes bundled with Google Workspace. Claude tends to produce the best first drafts of writing and code, and can build interactive tools inside the chat. Reach for the one that fits the task in front of you.

Fewer than the lists suggest. One tool per bucket, so around six, covers the vast majority of the work, and about ten tools handle roughly 90% of mine. Every new tool should have to beat the one already sitting in its bucket, and if you can't name the job it does, the person who owns it, and what success looks like at 30 days, don't adopt it.

Make it pass four questions before it earns a place: what recurring task does it do, who in the business owns and improves it, what output does it produce, and what does success look like at 30 days? Treat tools like hires with a defined job, not a collection to complete. Most run generous free plans, so you can test one properly before you pay.

No, and trying to is a common mistake. One of my most valuable workflows saves 16 hours a week and contains no AI at all: it just moves data between apps on a schedule. Use AI where genuine language or judgement is needed, and let plain automation handle repeatable, rule-based work. Forcing AI into everything tends to break things that already worked fine.

For quick research and fact-checking, Perplexity, because it pulls from live sources and cites every one, and its site operator lets you restrict a search to a single place like Reddit. For querying your own documents with no risk of made-up answers, NotebookLM, because it only answers from the sources you give it. They do different jobs: one searches the open web, the other stays inside your own material.

Start with small, repeatable admin, not a grand plan. The seven quick wins from the workshop are: record and transcribe every meeting, dictate to your AI instead of typing, connect your email so you can search and draft in seconds, set up reusable projects that hold your context, use pull prompting, turn your knowledge base into assets like SOPs and training, and build a personal assistant that briefs you each morning. Each takes minutes to set up and needs no technical skill.

Pull prompting is asking the AI to interview you before it does the work, instead of pushing it a request and hoping. Rather than "write me a proposal," you say "I need a proposal for a new client, before you start ask me any clarifying questions you need to do this well." It then asks its questions one at a time and builds from your answers, which are far better than its guesses.

Fewer than you think, and most are free. Fathom transcribes video calls and an app called Easy Recorder handles in-person ones. Wispr Flow is strong for dictation, though most everyday models now build it in. For the core work, one everyday model such as Claude, ChatGPT or Gemini does most of the heavy lifting, and NotebookLM turns your documents into something your team can query.

No. Every win here runs through settings menus and plain-English instructions, not code. Connecting your email is a few clicks under settings and integrations. Setting up a project means writing a paragraph about your business and uploading a few documents. Attendees who called themselves complete AI novices had these running the same week, without a developer or a new hire.

The average knowledge worker loses around 15 hours a week to admin, and McKinsey's data says more than half of that recurring work is automatable with tools that already exist. In one real example, a weekly reporting process that took my media buyer 8 to 16 hours went down to about 30 minutes of me reviewing the output once it was automated.

Score each process on five things: how much time it saves, its revenue impact, how often it goes wrong by hand, how rule-based it is, and how feasible it is with your current tools, then weight them so time saved and revenue count for more. A quicker rule of thumb is to start with the repetitive task that annoys you most, but never automate a broken process, because you'll only make the mess faster.