AI in Practice

AI content is either hype or homework. Neither helps you decide what to do on Monday morning.

This section is the practical end of AI. How specific tools work, where they fit in a real business, and how to use them well. You'll find pieces on Claude, Claude Code, APIs, image generation, automation patterns, and the working details that don't make it into vendor demos. The content here is more technical than the rest of the site by design. If you want to actually build something, this is where to start.

A founder at her desk in a small SaaS office looking thoughtfully at a laptop screen showing a Data Processing Agreement draft, a product roadmap with margin notes beside her

AI in B2B SaaS and tech firms in 2026

B2B SaaS is the one sector where AI sits inside the team and inside the product at the same time. Five internal jobs, five constraints unique to SaaS, and a 90-day rollout for a 20-person team at £6k to £15k a year.

A woman in her late 40s sitting at a quiet restaurant table before service, looking at a printed P&L and a roster on a tablet

AI in UK hospitality 2026: where the margin actually moves

UK hospitality is where AI maths now collides with the Employment Rights Act 2025. Five jobs are deployable today, with two that pay back inside six months for a 50-seat restaurant or 30-room hotel.

An operations director and a maintenance engineer in conversation on a precision-engineering shop floor next to a CNC machine, the director holding a tablet showing a sensor dashboard

AI in UK manufacturing in 2026: five use cases, six constraints, and Made Smarter as the route in

PrecisionParts cut downtime 30 percent, Brompton saved £60k a year on energy, Leyland Trucks inspects at 30 milliseconds. Made Smarter is funding the bridge. Here is what an SME manufacturer needs to know.

An agency founder at her standing desk in conversation with her chief creative officer, who is leaning on the desk holding a printed policy draft, mood boards visible behind them

AI in agencies: 30-40 percent productivity gains versus the billable hour

UK agencies are seeing real 15 to 40 percent productivity gains on content, paid media, and proposals. The harder question is what AI does to the billable-hour model and how to keep margin instead of giving it away.

A managing director of a UK fashion e-commerce business at her desk, looking at a printed P&L with a pen in her hand

AI in retail and e-commerce: what UK SMEs are actually deploying in 2026

UK retail is the sector with the densest SME-relevant AI evidence base in 2026. Footasylum, Iceland Foods, Boohoo, NOTHS and B&M are all named precedent. Here's what to actually start with at £2m to £10m revenue.

A managing director at a meeting-room table annotating two printouts labelled FCA examination response and AI vendor shortlist with a fountain pen

AI in compliance and risk for your business in 2026

Six compliance jobs UK SMEs are running on AI today, four that still need a qualified human, and what a 90-day rollout costs against the FCA, ICO and ISO/IEC 42001 governance the regulators now expect.

A managing director at her kitchen table early in the morning with a laptop, a notebook of priorities, and a cup of tea, working through her inbox before the workday begins

AI for the founder or MD: eight personal jobs, four boundaries, a 90-day rollout

Eight personal jobs AI does well for the founder or MD in 2026, four boundaries that still bite, and the 90-day rollout that keeps the productivity paradox off your desk.

A user researcher and a head of product reviewing tagged interview transcripts and unmoderated test results across two monitors at a desk

AI in product, design and customer research for your business in 2026

Where AI earns its keep across SME product and research work in 2026, where it falls short, and what a realistic 90-day rollout costs.

A head of delivery at a workstation with two monitors, reviewing a connected-search results panel and a vendor pricing spreadsheet alongside a printed shortlist of knowledge-management platforms

AI in knowledge management: six jobs that work, five failure modes that bite

Six knowledge-management jobs UK SMEs are running on AI today, the five failure modes that surface in regulated firms, and what a 90-day rollout costs against a 50-person services baseline.

An HR director at a desk with two monitors, reviewing an applicant tracking shortlist alongside a printed comparison of platform vendors with bias-audit notes

AI in HR: seven jobs that work, five the law still keeps human

Seven HR jobs UK SMEs are running on AI today, where the ICO and Equality Act draw the line, and what a 90-day rollout costs against the audit trail Article 22 expects.

A finance director at a workstation with two monitors, reviewing an invoice queue and a cashflow forecast alongside a printed shortlist of platform vendors

AI in finance and analytics: six jobs that work, four that still need a human

Six finance jobs UK SMEs are running on AI today, where the boundary still sits, and what a 90-day rollout costs against the audit trail HMRC and ICAEW expect.

An operations director at her desk in a professional services firm reviewing a printed budget paper, coffee in hand, a colleague visible in soft focus at a nearby desk

AI in operations and back-office for your business in 2026

Six operations jobs AI does well in 2026, three it does not, and what a 90-day rollout actually costs a £1m to £10m UK firm.

A founder writing in a notebook at her kitchen table with a laptop open beside her

AI in marketing for your business in 2026

A grounded look at where AI earns its keep across SME marketing, where it still fails, and how a 90-day rollout actually plays out.

A managing director at her desk in a consultancy office, looking at a printed pipeline report with handwritten margin notes, a closed laptop beside her, a thoughtful expression

AI in sales for UK SMEs in 2026: six jobs that work, three that don't

Sales is the function with the densest 2026 vendor noise. Six AI jobs are deployable today at SME scale, three are still genuinely human work, and the brochure won't tell you which is which.

A support manager at her desk holding a printed ticket export with handwritten intent tags, a chat dashboard open on the laptop beside her

AI in customer service for your business in 2026

Customer service is the function with the densest 2026 evidence and the sharpest line between what AI does well and what it quietly fails at. Six jobs to give it, two to keep it well clear of, and a 90-day rollout with real numbers.

A founder in his mid-forties at a desk in late afternoon, holding a printed contract, looking out the window thoughtfully, laptop open beside him, mug of cold tea on the desk

Your AI rollout is expanding output. Measure it that way.

The 13-hours-a-week saved figure is real. The behavioural data showing every category of work expanded is also real. The reconciliation changes how you should justify the spend at renewal.

A founder sitting at a kitchen table on a Friday evening with a laptop, a notepad and a glass of wine, looking past the screen in a moment of realisation.

AI rollouts don't stall in the tech, they stall in the middle

When AI adoption goes quiet, it is rarely the technology. It is the management layer running the same play on purpose, because their incentives reward them for slow-rolling you.

A founder at a desk reviewing a printed consulting proposal with margin notes and glasses on the page in late afternoon light

Why a single-number AI ROI claim is a red flag

Consultants anchor high (2x) at proposal time and SMEs see 1.4x at year-end. The gap is structural. A defensible proposal-stage claim is a range with confidence intervals and named risks, not a single number in a clean blue box.

A founder at a desk reviewing a printed board pack with handwritten margin notes in late afternoon light

How to defend the AI spend at the next board meeting

When the chair or CFO asks 'what have we got to show for the AI spend?', the questions are predictable. The CFO has three. The chair has two. The structured defence has four pieces.

Three people around a meeting room table mid-discussion with printed reports and notebooks in afternoon light

The 12-month AI review that decides continue, expand, contract, or kill

Most SME AI investments hit the renewal point and get extended on gut feel. A post-implementation review with an EOS-style sunk-cost check produces an explicit four-way decision, named owners, and a written record that prevents the next renewal from drifting back into gut feel.

A finance manager at a desk with a printed one-page dashboard and laptop open to a spreadsheet in late afternoon light

What an SME-scale AI ROI dashboard actually looks like

Most SMEs build a 25-metric AI dashboard nobody reads, or no dashboard at all. The middle ground is one page, five to eight metrics, refreshed quarterly. Buildable in a working week using tools the firm already owns.

An operations director at a desk with a laptop dashboard and notebook with handwritten cohort tracking in late afternoon light

Four proxy metrics when AI financial ROI is unmeasurable

Some AI deployments resist clean financial measurement because the data is messy or the tool touches several processes at once. Four proxy metrics give a credible read when financial ROI cannot be measured directly.

A managing partner reviewing a printed case-study deck across a meeting room table with another person mid-conversation

How to read AI vendor case studies sceptically

Every vendor case study you read shows top-decile customer outcomes. Five biases stack to inflate the published numbers. The reasonable response is not to discard case studies but to read them as existence proofs, not as expectations.

A director at a desk reviewing a vendor proposal with calculator and notebook in late afternoon light

What the 3.5x to 8x AI ROI benchmark actually represents

Vendor decks quote 3.5x to 8x AI ROI as if it were a baseline expectation. Decomposed honestly, the headline range comes from best-case aggregated data. The realistic benchmark for a competent SME deployment is 1.5x to 2.5x cumulative over two years.

A founder at a desk reviewing a quarterly report and renewal contract in late afternoon light, hand resting on the page

Why year-one AI ROI is structurally suspicious

Brynjolfsson's J-curve work shows technology productivity dips while workers learn, then climbs. Year-one AI ROI sits on the dip. Most founders feel ambushed at month nine; the curve explains why.

A finance director at a desk reviewing spreadsheets and a margin chart in late afternoon light

Why your AI hours saved haven't shown up as margin

Time saved doesn't equal margin expansion. The productivity-paradox literature names five pathways through which freed-up hours leak away from the bottom line. If you didn't plan where the time would go, the answer is into the noise.

An associate reviewing two stacks of marked-up documents side by side at a desk in late afternoon light

How to measure AI output quality without a research project

Most SMEs measure how much faster AI is and not whether the output is at least as good. A blinded rubric on thirty samples is the missing discipline, and it is achievable in a working week.

A finance manager at a desk with a clipboard time-log and laptop in late afternoon light

Why the AI hours-saved survey is not measurement

Asking your team how many hours AI saves them is the most common ROI methodology and the most indefensible. Time-study and activity logs are achievable at SME scale and produce numbers a CFO can actually trust.

A managing partner at a desk reviewing a printed report and notebook in late afternoon light, glasses laid on the page

Where you sit on the AI ROI maturity ladder

Most owner-led firms work AI ROI from numbers a CFO would dismiss in 40 minutes. The five-level maturity ladder, where most firms sit, and what climbing two levels actually involves.

A founder at a desk reviewing a printed contract with a pen in hand, mid-review

The six contract clauses that distinguish honest AI consulting from opportunistic billing

Six structural clauses protect the buyer in any AI consulting contract regardless of price. If two are missing, you should not sign.

A founder at a desk with two open notebooks side by side, one older and one new, with a pen between them

What CRM, ERP, and BI cost SMEs in the 2010s, and how AI is repeating the pattern

The 2-to-3x licensing TCO multiplier on CRM, the 70-80% ERP failure rate, the BI tools that sat unused. Founders who lived through the 2010s already paid for these lessons. AI is charging them again.

Two professionals in a meeting room across a table reviewing a document together

The regulatory line item your AI consultant should already have priced in

Legal, accountancy, healthcare, and financial services AI engagements have real regulatory cost. If the consultant's proposal does not name it, they are hiding it or do not know it exists.

An empty chair at a meeting table with a closed laptop, mug, and printed report

The full cost of a failed AI engagement, decomposed

A £15k engagement that fails costs £75k to £300k once internal time, productivity disruption, and the cost of waiting are added. The consulting fee is around a fifth of it.

A founder at a desk reviewing a printed proposal with a pen in hand, marking notes

What to demand from any AI pilot proposal

Most AI pilot proposals are sized correctly and specified poorly. Nine specs separate one that protects your investment from one that does not.

A team member at a kitchen table with a laptop, notebook, and coffee, working through AI experiments

The DIY trap: why £240 of ChatGPT a year quietly costs you £30k

The £240 ChatGPT licence looks like prudence. The senior staff member spending five hours a week on it is burning £25k to £50k of capacity a year.

A founder at a desk reviewing a printed spreadsheet next to a laptop and notebook, mid-decision

When in-house AI hire makes sense, and when it's still consulting territory

An in-house AI lead costs £56k to £176k year one with a 6 to 12 month ramp. Consulting only loses on cost when you have 3 to 4 concurrent AI projects.

The AI subscription stack that quietly costs more than your consultant

By year two, most SMEs are paying more for AI tooling than they spent on the strategy that selected it. Here is what the stack costs and how to plan for it.

Two people in a meeting room with a laptop, notebook, and sticky notes, in mid-conversation

The 100-hour staff time tax hidden inside every AI engagement

A typical AI strategy engagement absorbs 100 to 150 hours of your team's time. None of it appears on the consultant's invoice, and all of it is real money.

A founder seated at a kitchen table with a laptop, notebook, and printed proposal, paused in thought

The 2-to-4x rule: why your AI engagement budget multiplies after sign-off

The consulting fee is the smallest cost in any AI engagement, and the 2-to-4x rule explains why your real budget is twice to four times larger.

A founder unwell on a Saturday morning at a kitchen table, a laptop closed beside him, a kettle and a cup of tea, his hand resting on his forehead

When AI becomes the new founder dependency

Building a sophisticated AI stack only the founder understands rebuilds the dependency it was meant to break. Five conditions separate leverage from theatre.

A founder and her M&A advisor sitting at a conference table with papers between them, both looking at a printed valuation summary, mugs of coffee, the conversation in progress

The 12-month AI-accelerated exit-readiness programme

AI compresses the synthesis work that produces the four documents buyers pay for, taking exit-readiness from a 24-month £80,000 consulting engagement to 12 months at a fraction of the cost. The discipline is honesty, not speed.

A founder at her desk on a Monday morning, looking at a calendar on a laptop, an open notebook with a list of names, a mug of coffee in hand, the office quiet around her

When the senior person resigns, what AI can absorb and what it cannot

The 30-day handoff playbook: which parts of a senior role AI can hold (history, precedents, knowledge base) and which die with them (judgement, relationships, cultural memory).

A founder at a leadership meeting, a colleague leaning in to look at a laptop screen, two other team members at the table, coffee mugs and papers visible

When the founder is the only one in the firm using AI

If only the founder is AI-fluent, every AI-assisted decision routes back through them. Leadership-team AI literacy is the gating condition for genuine leverage.

A founder at his desk in lamp light at night, looking thoughtfully at an email draft on his laptop, a cup of tea nearby

Where AI in client communication erodes trust, and where it multiplies it

AI-mediated client comms can give a founder ten hours a week back without the client noticing, or it can erode a long relationship over weeks of slightly-off responses. The line is whether the founder still authors the message.

A founder at her kitchen table on a Sunday evening, laptop open with a document visible, a hand resting on a notebook, a mug of tea on the table

When recorded SOPs work, and the three places they fail

Recording-based AI-structured SOPs cut documentation from 20 hours to 3, but capture mechanics and not judgement, decay quickly without maintenance, and mislead when applied to founder-judgement work.

A founder at his desk on a Friday afternoon, looking at a Notion page on his laptop, an unopened notebook nearby, a mug of coffee on the desk

Why your prompt library is sitting in a Notion that nobody opens

Generic prompt libraries plateau at 15 to 20 percent adoption. The lever is workflow embedding and custom-trained GPTs, not storage.

A founder at her desk, laptop open with a workflow diagram on screen, an open notebook with handwritten notes, hands paused over the keyboard mid-thought

Designing AI as your operational integrator

Too small to afford the £150,000 integrator hire, too big for the founder to do everything alone. The agentic-stack pattern fills part of the gap, with named limits.

An MD at a meeting table with an adviser opposite, a laptop showing a calendar with three highlighted dates, water glasses on the table

The 30-minute monthly AI governance cadence for a 10-person business

AI governance does not need a committee, a board pack, or quarterly slide decks. At 10-50 staff it needs a rhythm. The actual cadence: 15-30 minutes monthly informal, 30-45 minutes quarterly formal, 1-2 hours annually for the full audit.

An estate agent MD and a marketing manager looking at a tablet showing a property image, the marketing manager pointing at the screen, brochures on the desk

When you must label AI-generated content

The EU AI Act and ICO guidance both require disclosure of AI-generated content in specific contexts. Most SMEs using AI for marketing or chatbots have not asked the question.

An MD and an operations lead reading printed documents at a meeting table, a laptop closed to one side, a notepad with handwritten notes

Article 22 and the human-review rule for AI decisions about people

UK GDPR Article 22 gives individuals the right not to be subject to decisions made solely by AI where those decisions have legal or similarly significant effects. Most SMEs deploying AI for hiring, credit, or eligibility have not met the rule and do not realise.

An MD and a commercial lawyer at a meeting table looking at a laptop, the lawyer pointing at the screen, two coffee cups on the table

The 'we're too small to be regulated' mistake on AI

Many SME owners assume regulators only pursue large firms. The empirical record on AI governance says otherwise. ICO, FCA, FTC, EU AI Act extraterritorial reach. The Samsung leak is the parallel-scale lesson.

An MD at a desk with a laptop showing a blank document, a notepad with handwritten draft categories beside, papers and a coffee cup on the desk

The one-page AI risk register that fits an SME

A risk register is a governance instrument. At enterprise scale it can run to dozens of categories and a dedicated GRC tool. At 10-50 staff it should fit on a single page.

An MD and a compliance adviser at a meeting room table looking at a laptop together, the adviser pointing at the screen, a notepad on the table

When you legally need a DPIA for AI, with five worked examples

Most SMEs do not run a DPIA before rolling out an AI tool. UK GDPR Article 35 makes one legally required for most AI processing of personal data, and the SME version fits 4-6 pages.

An operations lead at a standing desk with a printed reference table in front of her, a phone and a laptop nearby, a colleague visible in the background

Public, Internal, Confidential, Restricted: the four-tier rule that drives your AI policy

Most SME AI policies fail at the moment of practical decision: an employee with a document open and a tool tab open, asking 'can I put this into ChatGPT?' A four-tier classification scheme gives a one-line answer.

An accountancy firm owner at a desk with a printed checklist and a laptop showing two browser tabs, a coffee cup beside

The 12-question vendor due diligence sheet for any AI tool

SME owners typically buy AI tools the way they buy other software: a demo, a free trial, a credit card. The vendors most likely to fail you on data privacy or compliance are also the ones least likely to surface those failures in a sales call. Twelve questions surface the answers in 90 minutes.

An MD at a desk reading a printed report with a laptop open beside her, a notepad with handwritten figures and a calculator on the desk

The £150-a-month AI decision that changes your data risk

Free LLM tiers train on user inputs by default. Paid commercial tiers ship with Data Processing Agreements and training opt-out. The threshold between the two is roughly £15-30 per employee per month, and it is the highest-leverage governance spend an SME can make.

An MD and a project manager talking in an office hallway, both holding coffee cups, an open-plan office visible behind them

Why default-ban on AI tools backfires, and what to do instead

Most owners' first instinct on AI policy is to ban what hasn't been approved. The data on shadow AI says default-ban is the posture most likely to leave you with worse visibility than no policy at all.

An operations lead at a desk marking up a printed template with a red pen, a laptop open beside her with a blank document

The two-page AI policy that fits a 10-person business

An AI policy at SME scale should fit on 2-3 pages. If it runs to 30, it was copied from somewhere it does not belong. Here are the seven sections that need to be on the page.

A managing director at a desk with a closed printed playbook beside her and a laptop open showing a blank document, a notepad with handwritten notes in front

AI governance written for a 10-person business, not a 10,000-person one

Most published AI governance frameworks were written for organisations with a CISO, a DPO, and a risk committee. The owner-led 5-50 person firm has none of those, and reading the wrong source material is making the problem look unsolvable.

A senior practice manager at a desk reviewing a printed list with a red pen and a laptop showing a Notion page open

Why your knowledge base goes stale six months in

Owners deploy a knowledge base to capture institutional memory and watch it die in six to twelve months. The cause is rarely the platform. It is that nobody is responsible for keeping it current.

A practice owner at a desk with three printed pages laid in a row with arrows drawn between them and a notebook beside

The three AI processes that compound when you sequence them right

Most owners deploy AI one process at a time and get linear gains. A few processes compound when sequenced correctly. Calendar plus meeting plus knowledge base, in that order, generates more value together than the sum of their parts.

A founder at a kitchen table with a printed planning page showing process columns and estimated hours, a laptop showing a calendar

The owner's year-one AI time tax, the hours behind the time savings

Vendor demos talk about hours saved per week. They don't talk about hours invested per process to get those savings. The honest year-one cost is 80 to 160 hours of senior-person time.

A practice owner at a desk with multiple printed reports laid out and a notebook open with arrows drawn between documents

The data-readiness step every AI use case fails without

Across onboarding, financial reporting, invoice processing, and knowledge bases, the same prereq pattern shows up. AI does not work on the data layer most SMEs already have.

An accounts person at a desk with a printed exception report marked up in pen and a laptop showing an invoice approval queue

Where AI invoice processing actually pays back, and where it doesn't

Invoice and document extraction is the strongest single ROI case in the SME AI landscape, with payback measured in weeks not months. It also has clear failure modes when deployed wrong.

A sales lead at a desk with a printed proposal marked up in pen and a laptop showing an empty new-proposal screen

Why proposal AI doesn't change your win rate, and what it does change

Sales tooling vendors imply proposal AI lifts your win rate. The data does not support this. What it actually changes is deal velocity and the time the sales lead can spend on qualification.

A team lead at a desk with a printed weekly calendar marked up in highlighter and a laptop showing a calendar view

AI scheduling fails on calendar hygiene, not on the technology

Motion and Reclaim deployments die at week six in a lot of small firms. The cause is rarely the tool. It is that the team does not keep their calendars current enough for the AI to make sensible suggestions.

A consultant on a video call at a desk with a laptop open and a notebook with handwritten action items beside a phone face-down

Bot vs local capture, why client calls aren't a place for Otter

Most meeting AI tools work by joining the call as a visible bot. Some clients hate this. The tool category that fits a client call is not the same as the one that fits an internal stand-up.

A practice manager at a desk with coloured sticky notes on a screen labelling email categories, a notebook open to a triage flowchart

Triage before draft, the inbox AI sequence most SMEs get wrong

Owners read the Klarna chatbot story and deploy autonomous response generation. The right deployment for SMEs starts at classification, not at drafting. The order matters more than the tool.

A finance lead at a desk with two monitors and a marked-up printed chart of accounts beside the keyboard

Clean your data before you buy the financial AI tool

Owners deploy AI forecasting and reporting tools expecting cleaner numbers. They get faster, more confident wrong numbers. The data layer is the pre-step the vendors do not sell.

A senior solicitor reviewing a printed contract marked up in red pen with a laptop open beside it

Why AI extracts contracts well and drafts them dangerously

AI is good at one half of contract work and bad at the other. Owners who treat the two halves as the same task get sanctioned, get sued, or get expensive-but-safe.

A practice manager at a desk with a printed onboarding checklist marked up with red pen and a laptop open beside a mug

The bottleneck in client onboarding isn't data entry

Owners deploy AI to clear the onboarding admin pile and watch the cycle time barely move. The bottleneck was never the data entry. It was the decisions that came before it.

A founder at a kitchen table with three vendor proposals fanned out, hand resting on the closed cover of one of them, looking out the window with a considered expression

When not to try a second time

The default after a failed AI engagement is 'try again, smarter'. Sometimes that is right. Four conditions where it is wrong, and a way to tell which kind of pause you are in.

A founder at a desk reviewing three printed reports laid out side by side, a finger on a number in the middle one, mid-comparison, notebook open with handwritten cross-references

What the data actually says about second-engagement success

Three bodies of research that do not usually sit together: CHAOS on IT project recovery, McKinsey on digital transformation stalls, Prosci on change management ROI. Read together they give a useful answer to 'if we do this again, what are the odds'.

Two people across a desk, one explaining a hand-drawn timeline divided into milestones, the other listening with pen poised over a notebook

The four posture shifts in a second AI engagement

Second AI engagements that succeed share four structural shifts in posture: budget, governance, staff, scope. The goal stays the same. The structure changes. Concrete numbers throughout.

A founder at a desk with a printed contract spread out, a yellow highlighter mid-stroke on a clause, a notebook with notes alongside

Vendor diligence after a burn: the post-burn checklist most SMEs do not run

After a vendor relationship that did not work, the SME instinct is 'be more careful'. The vendor-management literature has clear answers about what 'more careful' actually means.

Two people across a kitchen table mid-conversation, one listening with a notebook open and pen down, the other speaking with hands resting on the table

Rebuilding trust after a botched AI rollout

Trust is a function of four things. AI tools wobble on all four at once. Here is how to rebuild it before the second engagement starts, with the three feedback tools that actually work.

A founder at a desk with a printed SaaS spend report, highlighting subscriptions with a marker, mug and laptop pushed to one side

Why you should consolidate your software stack before adding more AI

After a stalled AI engagement, the counter-intuitive move is to step backward first. Cut your application count and stabilise the stack. The data on what fragmentation costs makes the case clearly.

A founder at a kitchen table with three printed documents fanned out, a notebook open to a hand-drawn six-cell grid, pen mid-stroke

The six-dimension diagnostic audit you should run before a second AI engagement

The four-question diagnostic on the first engagement is a start. The audit you actually need looks at the organisation across six dimensions, in three phases, in an afternoon.

A founder at a desk reading a printed report covered in pencilled notes, a closed laptop and coffee cup beside it

The 73 percent figure, decomposed: where most first AI engagements fail

Beneath the headline 73 percent failure rate sits a more useful breakdown: where the modal first AI engagement actually falls off, and what that implies for the second move.

A founder in her mid-forties at her desk on a Friday afternoon, laptop open next to a paper diary, leaning back in her chair holding a pen, late afternoon light through the window.

When the competitor moved on AI and you did not

The AI-acceleration panic is a bandwidth problem, not a tooling problem. Adding AI on top of a founder-dependent business accelerates depletion rather than producing leverage.

A founder at a desk reading a printed proposal with a calculator and notebook beside it

What 10k, 25k, and 50k actually buys in AI advisory

Three budget tiers exist for SME AI advisory and they buy three different things. Owners regularly pay one tier and expect the deliverables of another. Here is what each tier really gets you, and how to buy small first.

A founder at a desk on a phone call holding a pen over an open notebook with a numbered list of questions visible

The reference call most AI buyers conduct badly, and the nine questions that fix it

Most reference calls produce boilerplate positive feedback because most buyers ask the wrong questions. A nine-question structure produces real signal. The question that breaks the pattern is the sixth one.

A founder at a desk reading a printed CV with a notebook open beside it showing handwritten notes

What to demand from an AI consultant who has never run a business

Operator-backed consultants are not universally better than career practitioners. The question is consequential for an SME, and rarely asked. Here is the LinkedIn check, the integrating question, and where each background fails.

A founder at a desk comparing three printed quotes laid side by side with a notebook and calculator nearby

Hourly, fixed-fee, or retainer: which AI consulting pricing model actually serves you

AI consulting prices in five different models, each with its own incentive structure. Three of them work at SME scale, two are usually wrong. The pricing model is itself a signal about the consultant.

Two people in conversation across a small office table, one holding a notebook with handwritten notes

Three things on the first call that should end the AI consultant conversation

Most failed AI engagements were predictable on the first sales call. Three specific behaviours, each tied to a documented failure pattern, are enough to walk away early.

A founder at a desk reviewing a printed proposal with handwritten margin notes visible on the pages

What a real AI roadmap includes, and what you should refuse to pay for

An AI roadmap engagement at £5,000 to £15,000 should produce five named deliverables. Half the engagements at this price tier produce a slide deck and not much else. Here is what to demand and where to push back.

A founder at a desk on a video call with a notebook open beside the laptop showing handwritten notes

Why your AI consultant might be selling you a platform, and how to tell

Vendor-aligned advisors steer 44 percentage points more of their clients to their vendor's platform than independent advisors do. The bias is structural, not deceptive, and detectable in five specific questions.

A founder at a kitchen table with three printed proposals laid out side by side and a notebook open with handwritten notes

Fractional CAIO, project, or roadmap: matching the AI consultant format to where you actually are

Three engagement formats. Three different starting positions. Most owners pick on monthly cash flow and end up with the wrong format for the question they actually have.

A managing partner at his desk in a boutique consulting office, reviewing a printed resourcing plan with handwritten margin notes, a laptop open beside him, a thoughtful expression

AI is restructuring the consulting model. The 'obelisk' reshape and what it means for boutiques.

Harvard Business Review's 'obelisk' framing names what's actually happening to consulting firm structure under AI. McKinsey's Lilli copilot shows what scaling looks like at the top. The harder question for a 28-person boutique is the staffing model itself.

A compliance officer at her desk in a financial services office, looking thoughtfully at a printed marketing draft with annotations, a laptop open beside her, a pen in her hand

If any of this sounds familiar, let's talk.

The next step is a conversation. No pitch, no pressure. Just an honest discussion about where you are and whether I can help.

Book a conversation