The Lean AI Method

    MIT said up to 95% of generative AI pilots return no measurable financial value. That number, from the NANDA study published in 2025, isn't an AI problem. It's a process problem wearing AI clothes. Teams buy tools before they define the work, then blame the tool when nothing changes. We've watched it happen in factories, agencies, and marketing ops teams for a decade. The pattern is always the same.

    The Lean AI Method is the answer we've built from that decade of work. It's Lean Six Sigma applied to AI implementation, five principles that keep mid-market businesses in the 5% who actually get a return. It's the playbook we run on every Workshop, Audit, and Sprint. The rule underneath all of it: process first, technology second.

    Why Lean AI exists

    The hype is loud and the budgets are real. A lot of MDs we talk to have already bought three or four AI tools by the time we meet them, often before anyone wrote down what the tools were supposed to do. We've sat in operations meetings where someone said, "we need an AI for that," and nobody in the room could explain what the underlying process actually looked like. Tools first, problem later. That's how you end up with 11 AI subscriptions and a team that uses two of them.

    The other half of the failure rate is change, not code. AI projects fail for the same reason most operational projects fail: the people doing the work weren't consulted, the workflow wasn't mapped, and the ROI was a hope, not a baseline. BCG put the general transformation failure rate at around 70% in 2025, and they were measuring this long before generative AI arrived. AI doesn't break the pattern. It accelerates it, because the technical complexity hides the process gaps for longer.

    Lean Six Sigma was built to solve exactly this. It's a methodology for spotting waste, mapping how work actually flows, and improving processes through small, measured experiments. We've spent over 10,000 hours running it across manufacturing floors, digital teams, and marketing ops engines. The frameworks transfer almost cleanly. A Gemba walk turns into a screen-share through your CRM. A value stream map turns into a workflow diagram of how a lead becomes a customer. Continuous improvement turns into a 90-day sprint with a measured before-and-after. The names change, the discipline doesn't.

    The Lean AI Method packages that discipline into five principles you can hold a project to. Process first. Pilot before platform. Skin in the game. Skill the team. Measure ROI from day one. Each one is a guard against a specific way AI projects go wrong, and each one maps to a step in our offer ladder. Definitions for the terms we use, AI pilot, AI ROI, AI skill gap, three-question filter, live in the Lean AI glossary.

    Who this is for

    The Lean AI Method is built for marketing leaders at UK mid-market businesses, £3M to £10M turnover, typically 20 to 80 people, with a marketing team of 2 to 8. Three sectors specifically: agencies, B2B businesses with in-house marketing teams, and professional services firms with a lead-gen engine. The buyer is the MD, the Marketing Director, or the Head of Marketing, the person who owns the marketing budget and feels the drag of manual work most directly. The pattern we see is consistent: a busy, profitable business that has bought a handful of AI tools, has a backlog of things AI "could help with", and no internal capacity to figure out which use case to start with or how to measure whether it worked.

    A note on scope. The five principles below are universal, they apply to any process with waste in it. Marketing operations is where we start because the workflows live on screens (a digital Gemba works), the metrics move in days not months, and the spend on AI tools is already happening and mostly missing the mark. The method generalises beyond marketing, the conversion focus on this site does not.

    It isn't built for enterprise. Enterprise AI buying is its own world: dedicated AI teams, multi-year platform contracts, a different risk profile, a different set of stakes. The five principles still apply, the implementation rhythm is different. At the other end, very small teams under £1M turnover tend to need a single sharp contractor, not a methodology. The 90-day sprint is sized for the gap in the middle, where the business has enough revenue to justify the work and not enough internal capacity to do it themselves.

    The five principles of Lean AI

    1. Process first, technology second

    Every AI project we've seen fail had the same opening move: someone picked the tool before anyone mapped the process. The chatbot got bought. The automation platform got signed. The seat licences got allocated. Then, six months later, the question came back: why isn't this saving us any time? Because the underlying workflow was still broken, the AI just made the broken workflow run faster, and now there were two systems of record instead of one. Tools without process are shelfware. Always have been.

    We always start with the work. Before we talk about models or vendors, we map the process as it runs today, the actual steps, the actual handoffs, the actual time spent. Most of the time, two things become obvious in the first hour. One: there are steps in the process that exist only because nobody's questioned them. Two: a chunk of the "AI problem" is really a workflow problem that a clearer process and a spreadsheet formula could fix on its own. The job of AI is to do what only AI can do, language at volume, pattern matching, prediction, personalisation, repetitive cognitive work. The rest of it is a process question, and a process question deserves a process answer.

    Try the AI Necessity Test, eight minutes, no sign-up, scores whether a process actually needs AI or whether a clearer workflow would do the job.

    2. Pilot before platform

    The pattern that costs the most money is buying the platform before proving the pilot. Enterprise AI vendors sell on 12-month contracts and price by seat or by API call, which means the moment you sign, the meter starts running on a use case nobody has tested. We've watched businesses spend £40,000 to £80,000 on annual licences for AI tools that ended up serving a single team running a single workflow nobody adopted. The contract was bigger than the problem.

    A pilot fixes that. A pilot is a small, time-boxed experiment, four to twelve weeks, one workflow, one team, one measurable outcome. You build the minimum version of the solution, run it through the people who'll actually use it, and measure whether it moves the needle. If it works, you scale, and now you know exactly what you're scaling and what it's worth. If it doesn't, you've spent a fraction of what a platform would have cost and you've learned something useful about your own process. Either way, you bought information, not just software.

    The trick with pilots is to keep them honest. We see two failure modes. One: the pilot is so small it can't fail, a five-user trial of a tool that ships value at fifty. That doesn't prove anything except that the demo works. Two: the pilot has no exit criteria, no defined "we keep it" or "we kill it" outcome, so it drifts into a permanent project that nobody owns. A proper pilot has a baseline metric, a target metric, a decision gate, and a date. Hit the gate, scale it. Miss the gate, stop, and write down what you learned. Either outcome is a win, the only loss is letting it drift.

    See: AI Pilot (glossary).

    3. Skin in the game

    A lot of the AI work in the UK mid-market is sold on 12-month retainers with no tied-back outcome. The vendor gets paid whether or not anything ships. The buyer waits for "discovery" to end, watches the quarterly invoice land, and tries to justify the spend at the next board meeting. By month nine, half of those contracts are limping, and the other half are about to be renegotiated. The incentives are pointing the wrong way.

    Our default engagement is a 90-day fixed-scope sprint. One use case picked from the audit, scoped, built, deployed, measured. The scope is fixed so the budget doesn't drift. The window is short so the decision-makers stay close to the work. The guarantee is tied to the outcome: if the implemented use case isn't showing measurable positive ROI by day 90, we keep working at no extra cost until it does. Skin in the game runs both ways. The point isn't bravado, it's alignment. When the people building the solution share the outcome risk, the conversations get sharper and the waste gets cut.

    The 90-day window does another thing that matters more than people expect: it forces the scope to stay honest. Twelve-month projects accumulate features the way old hard drives accumulate files, every passing meeting adds a "wouldn't it be nice if". By day 60 the original use case is buried under nice-to-haves nobody can remember the business case for. A 90-day sprint can't carry that weight. The scope at day 1 is almost exactly the scope at day 90, because there isn't time for it to drift. Constraint, in this case, is a feature.

    See: 90-Day Implementation Sprint.

    4. Skill the team, don't replace it

    The headlines say AI is coming for the jobs. The reality on the ground is that AI is coming for the tasks, and the people who own the tasks are the only ones who know which tasks matter. We've never finished a sprint where the right answer was to remove a team. The right answer was always to give the team a sharper tool and the skill to use it well. Replace the team and you lose the institutional memory that made the process work in the first place, the unwritten rules, the customer quirks, the edge cases nobody documented. AI doesn't know any of that. Your team does.

    The skill gap is closeable, and faster than most people think. In a marketing ops team of 4 to 8 people, we typically get the core users productive on a new AI workflow inside two to four weeks, and confident inside eight. That's not because the technology is simple, it's because we train against their actual work, not against a generic curriculum. Every sprint includes a training programme built around the workflow we shipped, so the team that has to live with the tool also knows how to extend it. The handover isn't a PDF, it's competence.

    The principle has a second-order effect that compounds over time. A team that has built one AI workflow successfully has an instinct for the next one. They know what a good use case looks like. They can spot a vendor demo dressed up as a strategy. They can scope a pilot themselves and ask the right awkward questions about ROI. After the first sprint, the second use case takes less of our help. After the third, the team is running the rhythm without us. That's the goal. The retainer exists for the high-stakes decisions and the next-stage planning, the day-to-day muscle lives inside the business.

    See: AI Skill Gap (glossary).

    5. Measure ROI from day one

    The most expensive sentence in AI is, "we'll figure out the ROI later." Later doesn't come. The pilot ends, the team moves on, the budget cycle resets, and the only evidence anyone has is a vague sense that "things feel a bit better." That's not a business case. That's a feeling, and feelings don't unlock the next round of investment.

    We baseline the metric before the pilot starts, not after. If the use case is reducing the time it takes to qualify a lead, we measure today's cycle time in days and hours, for a real sample of real leads, before we touch the workflow. If the use case is improving the response rate on outbound, we measure today's open rate, reply rate, and meetings booked, against the same target list and the same script. Every sprint ends with a measured before-and-after, presented to the board with the working solution beside it. Numbers, not impressions. That's how the second use case gets approved, and the third, and the rollout that follows.

    The baseline does another job that's harder to put a price on. It surfaces the real cost of the current state, which a lot of teams have never actually quantified. We've run baselining exercises where the finance director discovered the qualification process was eating 601 hours of senior sales time a quarter, against an assumed 200. The AI use case was a clear win. The bigger win was the conversation that started after, about whether the process should exist in that shape at all. Measurement is where the obvious questions get asked, and the obvious questions are usually the expensive ones.

    See: AI ROI (glossary).

    Anti-patterns we keep seeing

    The five principles are easier to remember once you've watched the alternatives fail. These are the patterns that come up in nearly every diagnostic conversation, the shapes that signal an AI project is heading for the 95%. None of them are individually fatal. Stack two or three of them and the result is predictable.

    The tool stack with no map. A business has 8 to 12 AI subscriptions, half of them paid for by individual managers on company cards, with no central record of what each one does. Nobody can answer the question "what would we stop doing if we cancelled this tomorrow?" A lot of those subscriptions are quietly auto-renewing on charges that add up to five figures a year, against a usage rate of under 10%.

    The pilot that became permanent. A six-week proof-of-concept that has been running for fourteen months with no decision gate. The original sponsor moved on, the new owner inherited it, the metrics nobody agreed on at the start are still the metrics nobody can agree on now. Cancelling it feels like an admission of failure. Keeping it feels like throwing good money after bad. Both are right.

    The vendor-led roadmap. The AI strategy is the vendor's product roadmap. Every new feature in the platform becomes a new use case the business "should" be doing, regardless of whether anyone needs it. Strategy by feature release isn't a strategy. It's a backlog being read out loud.

    The hero AI champion. One enthusiastic team member, often technical, builds three impressive prototypes. None of them survive the move from their machine to a production workflow because nobody else was trained, the integrations weren't built, and the prototype assumed clean data the business doesn't have. The champion leaves. The prototypes go with them. The business is back to zero.

    The "save time" use case with no time saved. The pilot ships, the team uses it, and when the finance director asks where the savings landed in the P&L, nobody can answer. The hours saved went into other tasks, which is fine, but it isn't ROI. ROI is a number on a report. If you can't show it, the next budget round won't be friendly.

    How the method maps to our offer ladder

    The five principles don't live in a binder, they run on a four-step offer ladder. Each step does one job, time-boxed, with a guarantee tied to a specific outcome. Each step credits forward into the next, so committing at the top of the ladder is cheap and the path to the full Sprint is short. Workshop, Audit, Sprint, Retainer. The names map to the principles. The ladder is the method made tangible.

    1. AI Strategy Workshop, half-day

    A four-hour facilitated session for up to five people from your leadership team. We run an opportunity scan against your operations, surface at least three specific AI use cases that fit your business, and leave you with a prioritised action plan. The fee credits in full toward the Audit if you continue. If we don't find three real opportunities in your business, you don't pay. This is principle one in action: define the work before anyone shops for tools.

    2. AI Readiness Audit, 14 days

    A 14-day deep dive into three of your highest-waste workflows. Two digital Gemba sessions, screen-shares through the tools your team actually uses, then a full process map, an opportunity scorecard with projected ROI per use case, and a 90-day roadmap. The deliverable is a board-ready briefing your finance director can sign off on. The audit fee credits in full toward the Sprint. The framework that drives it lives on the AI readiness page.

    3. 90-Day Implementation Sprint

    The Sprint takes the number-one use case from the Audit and ships it. Four phases in 90 days: discover, design, deploy, scale. We build the workflow, integrate it into your systems, train your team, and produce a live ROI dashboard. At day 90, your board sees the working solution and the measured before-and-after. If the ROI isn't positive, we keep working at no extra cost until it is. One use case. Built. Tested. Delivering.

    4. AI Advisory Retainer, monthly

    Once the first use case is live, the retainer keeps the momentum. Monthly strategy calls, quarterly roadmap reviews, priority email support, an annual training refresh for the team. The point isn't to replace your team with us, it's to keep a sharp AI brain on call so you can roll out use cases two, three, and four without losing the discipline that made the first one work. A 12-month minimum, then 90 days' notice. You renew because the work is paying for itself.

    Where to start

    The honest answer for most teams is: don't start with a tool, start with clarity. A lot of the businesses that come to us are sitting on three or four candidate AI use cases and can't tell which one actually has a return behind it. The cheapest move is to score the candidates against the five principles before anyone signs anything. Pick the one with the clearest process, the cleanest data, and the largest cost baseline. Pilot that one. Measure it. If it works, scale. If it doesn't, you've learned something useful and you haven't blown the budget. Three steps, in order of commitment, each one earning the next.

    A note on the name

    "Lean AI" is also the title of a 2024 book by Lomit Patel published by O'Reilly. It is unrelated to our methodology. The term "Lean" is also used by the Lean programming language at lean-lang.org, which is a different thing again. The methodology described on this page is the LeverageAI approach: Lean Six Sigma applied to AI implementation, documented in our book of the same name.

    The book

    The full method lives in our book, Lean AI: The 90-Day Transformation Playbook for Growing Businesses. It's written for the broader audience the method applies to, business owners, MDs, and department heads at any growing UK business asking where AI actually fits, not just marketing leaders. The book's job is to give the method away. The consulting on this site is for the narrower marketing-ops slice where we deliver it. Join the waitlist for the launch.