What we've done, how well it worked, and what it taught us. Honestly, with the numbers in front of us.
Five people leaving the room with the same picture of the business — and the same answer to what we're not doing.
An 18-month plan we can actually execute — with the numbers behind it, not just the intent.
No scheduled breaks. We're all grown-ups.
Take five between sessions whenever you need it — grab a drink, step outside, take a call. We won't be timetabling it. Lunch is on the premises at 12:00. The one ask: when a session is running, phones down and be in the room.
One thing each about yourself that nobody in this room knows. You're allowed to stretch it. We'll guess whether you did.
We have been trading through MIXE Group Ltd on a founders agreement — deliberately light, on the basis that we would not add structure until there was something worth structuring. That test has been met. So if we agree, there are three things to settle.
The original plan was to incorporate something new. Worth revisiting.
Change the plan — rename MIXE Group rather than start again.
A rename is a resolution and a filing. Everything else stays where it is, and we keep the trading history rather than resetting the clock. Either route leaves the name open: nothing obliges us to be Lean Growth Labs.
Scott reduces his shareholding to 47.5%
SibbertCo — Nigel and Libby — increases to 17.5%
The employee pot reduces to zero and is redistributed
All five of us become directors of the company
And the other way round: if anyone would rather step back, today is the cleanest day to say so — before the agreement is drafted.
Worth asking properly while everything else is being settled. The legal entity is MIXE Group Ltd — "Lean Growth Labs" is a trading name, so changing it touches no shares, no filings, and no structure. It will never be cheaper to change than today.
The deciding question isn't whether we like it. It's whether it's the name we want on the ad that has to sell us to someone who has never heard of us.
Why cash sits £3,778 below profit: take the £6,837 of profit, deduct the £4,000 Aura invoice that has not been paid yet, add the £262 of cash carried in from last year, and deduct £40 of non-business spend on the company card. That leaves £3,059. Source: Xero, accrual basis, as at 9 September 2026. Nothing owed to suppliers.
Not values on a wall. The five things this year has told us to hold onto.
Solve one or two real problems, not twenty. A small fix they understand beats a grand system they do not.
Lots of small clients beats one big one. A single large account swallows every hour, then ends.
The wrong client costs more than no client. Wrong sector, wrong size, wrong chemistry — walk away early.
Every job starts with money in, covering our costs at a minimum. If we share risk, we share profit, never costs.
Look for the ongoing element in every deal, and take it where we can. A one-off is a good month. A retainer is a business.
One engineer, Harry, running a small UK practice and building every engagement himself. Positions against the field as “engineers, not resellers” — agencies who glue no-code templates together and vanish when it breaks.
Automation sprints, plus a productised AI receptionist that answers every enquiry, qualifies it, books the appointment and updates the CRM. Six industry pages: letting agents, dental and physio, recruiters, accountants, e‑commerce, tech teams.
“Processes that live in one person’s head. When they are off sick or on holiday, things stop. That is not a process — that is a risk.” Northloop homepage
Our own thesis, on someone else’s homepage. We say it just as well — “your best people are the process” — so the gap is not the words. It is that theirs sits on six sector pages next to a fixed price and an ROI calculator, at an eighth of our ticket.
Oliver, on his own, in Buckinghamshire and the Thames Valley. Sells to residential building, renovation and high-ticket home improvement firms — and says so in the first line.
A free Comparative Lead Leak Audit: he submits a real enquiry to the client and to four named local rivals, times every reply, and shows the comparison. That leads to a costed diagnostic, then a build, then ongoing management reported against jobs won.
“You never see the enquiry that went to a rival because they replied the same morning and you replied on Thursday.”Loxvik homepage
Nothing about AI appears above the fold, and there is a published list of who this is not for — including his own geography. Narrower than either of our personas, and the most disciplined audience writing we found.
A Chesterfield build shop selling “whole-system automation” to anyone from 1 to 200 staff. Shipped work in field-service dispatch, racing analytics and property planning data.
A tool menu: n8n, Zapier and Make, AI agents, WhatsApp and website chatbots, CRM automation, data pipelines, dashboards, Stripe, deliverability. No strategy layer — you arrive knowing what you want built.
“Three freelancers, three half-finished systems. Or one that fits together.”The Automation Agency homepage
A good line — but it names a buying problem, not a business one. It only lands on someone who has already bought automation once and been burned. Strong at catching demand, useless at creating it.
Sunderland-based, badged “SMEs only” and “North East based” on the homepage. Free 30-minute AI audit at the top of the funnel, builds live within two weeks.
Done-for-you agents and workflow automation, two-hour team training workshops, AI SEO, managed social media, and a design-render studio. Breadth, not depth — and training sold openly as a product, not as a downsell.
“Stop losing hours to tasks AI can handle in minutes” — cut admin by 10+ hours a week, boost output by 40%, ROI in six months.KlarifAi homepage
Percentages with no source, stock boardroom photography, no sector named. At £99 a month that is survivable. The interesting part is the ladder: they can transact with businesses that will never write us a cheque.
“AI consulting and managed outbound” for UK B2B. Named client work in electrical contracting, glazing, flooring, drinks and film. Structurally the nearest thing to us in the set: consulting plus lead generation.
A consulting roadmap of where AI is worth applying — explicitly portable to another implementation partner — and an Outbound Sales Engine they run for you across email and LinkedIn. The client rents the pipeline rather than owning it.
“Better decisions. Smarter growth.” — and, further down, “not every problem needs a new tool. Sometimes the answer is a better process.” Elevate AI homepage
Calm and credible, and that second line is one we could have written. But the page never names a person, a week or a stuck process — it reads like a firm expecting to be met by referral rather than found by a founder in pain.
We checked all five across the Meta ad library — Facebook, Instagram and Audience Network — the LinkedIn ad library and Google’s Ads Transparency Centre. Nothing is running today. One of them has run a campaign: Elevate AI, from December to June, selling a free AI audit.
Five video ads on Facebook, Instagram and Audience Network. The offer was a free 30-minute AI Automation Audit, booked straight into Cal.com, behind a lead form that qualified on role, headcount and timeline. Started 23 December, stopped 2 June.
The live Meta ads in this category are “start your own AI agency” offers, GoHighLevel resale packs and course funnels. Almost none of it is aimed at the business owner who actually has the problem.
Nobody is competing for this attention right now. But the one firm that tried our exact motion stopped after six months, and Meta does not publish what they spent or got. Worth knowing before we treat the channel as free ground.
What the libraries can and cannot tell us: Meta keeps an archive, so “ran ads” is solid; LinkedIn and Google show a much shorter window, so read those as no ads found, not never advertised. Spend and results are not published for ads like these, and no public tool reports traffic for sites this small.
“Where it breaks” is as sharp as anything in this set — Northloop says the same thing, no better. The bit still unsaid, by us and by them: what founder-dependency costs when you come to sell.
We already name the moment — the lead that arrived Monday, unconfirmed until Friday. What we don’t name is the sector. Northloop has six pages of it. We have one page for everyone.
Loxvik disqualifies people on the homepage and reads more confident for it. Our not-for list is currently internal.
£20,000 fixed for the build, £2,500 to build it with us, “no hourly billing, no scope creep” — all on the product pages. Elevate AI publishes nothing and reads vaguest for it. The only gap: our number is one click in, not on the homepage.
“Yours to own and control, nothing to host yourself” — already on the product page. Two of the five keep running what they built instead. Nobody else in this set claims it, so it should be a headline, not a bullet.
Most of this is already done. The words, the price and the ownership promise are all there — they just sit a click in, behind a homepage that speaks to every sector at once. The work is sector proof, and pulling what we already say to the front.
The model barely matters on its own. Pair it with a harness that holds memory and state, and MCP to reach real tools — one standard, adopted across Anthropic, OpenAI and Google in a year — and it stops answering and starts working.
Businesses aren’t stuck because the AI isn’t clever enough. They are stuck because they can’t be sure it got the answer right. 75% say that is what stops them letting it near their customers. Checking the work is the job.
AI can’t tell that “Acme Ltd” and “Acme Limited” are the same customer. The firms getting anything out of AI are the ones who already kept their data in order — 41% of them against 26% of everyone else.
Microsoft has 30m paid Copilot seats, but 30–40% go unused in the first 90 days. Google put Gemini into Workspace at no extra charge. Whichever suite a client runs has already picked their AI for them.
The biggest cheques are going to compute, data‑centre power, chips and defence — the scarce inputs. Generic AI applications are not where capital is concentrating; specific, high‑value workflows are.
All stock, closed 14 August 2026, with xAI folded into SpaceX alongside it. A coding agent valued at $29bn nine months earlier. Models, compute and the application under one roof — that is the strategy being bought.
DeepSeek, Qwen, Kimi, GLM and MiniMax now score ~77–80% on SWE‑bench Verified at a fraction of frontier pricing. The gap that remains is long‑horizon agent reliability, multimodal work and knowing when they are wrong.
Agent loops burn 5–30× the tokens of a chat request, so subscriptions stopped cross‑subsidising. Anthropic capped weekly use in Aug 2025 — naming 24/7 Claude Code running — then locked third‑party harnesses out of subscription auth entirely by April 2026.
The squeeze arrived from somewhere else entirely — and it closed off the answer they used to reach for first, which was to hire somebody.
+£835 in employer NI on a single £25k salary. Firms scaling back recruitment went from 31% to 45% in a year.
Entry‑level accountant postings down 29%, legal assistants 14%. The work still needs doing; nobody is being hired to do it.
11–14 hours a week on admin and finance. 68% hit by late payment, losing 86 working hours a year chasing invoices.
When Google answers the question itself, people click a result 8% of the time instead of 15%. Far fewer of them ever reach a website.
Around half of UK SMEs say they use AI — but only about one in ten have built anything, and 95% of those using it report no change in workforce size. Four reasons why.
The firm or its people do not have the knowledge to use AI successfully. The largest single barrier — and not one a better tool fixes.
Data privacy is the top concern among non‑adopters, with 31% citing security risk. For an accountant or solicitor that is an obligation, not a worry.
A quarter do not believe AI is appropriate for their business. Not hostile — unconvinced, which is a completely different sale.
ONS finds identifying a use case and cost are the common obstacles. They are not short of tools — they are short of someone saying “start here”.
None of those four barriers is about the technology, and their data is split across Microsoft or Google, Xero or QuickBooks, HubSpot or Salesforce and a lot of spreadsheets. Every one of them is a reason to hire somebody rather than buy something — which makes this a diagnosis problem before it is a build problem.
Capture live. Nothing leaves this room as "someone should".
We're off and running. The right team, the right pieces, pointed at the right part of the market. Money is coming in.
What we haven't found is the widget — the one thing we can build a thousand times over. The rest of this session is about naming it and building the line that produces it.
Immediate value — recommendations, tools, what they should be doing. And it surfaces the uncomfortable truth that they have neither the capability nor the capacity to act on it. We're not selling at this stage. We're being useful in public.
A long-term partnership making incremental improvements, one problem at a time. Light touch, repeatable, and the only tier that compounds. This is the widget.
Chunky, transformational, expensive. They don't scale and they never will — but they're extremely profitable, and sitting inside the business every month is how we see them coming.
One warning: tier three is where our instincts pull hardest, and it's the tier that can quietly eat the other two. It works as the upside on top of a retainer book. It does not work as the business.
A monthly retainer for a set number of hours. They bring us problems, we solve them with them — live, on a call, one at a time. They watch it happen and learn to do it themselves, so they're part of the solution rather than dependent on us.
This is the thing we can build a thousand times. Same shape for every client, different problems inside it. It recurs, it compounds, and it's the only asset here that's worth something at exit.
| 2026 | 2027 | 2028 | |
|---|---|---|---|
| Audit revenue | |||
| Retainer revenue | |||
| Project work | |||
| Total revenue | |||
| vs target | |||
| Retainer clients | |||
| Corporation tax | |||
| Distributable | |||
| Per quarter |
Engine starts October 2026 and ramps to full rate over 18 months. Marketing at 10% of revenue; project work at £2,500 per retained client per year. CT 19% to £50k, marginal relief to £250k, 25% above. Three directors at the personal allowance, GM from mid-2027. Illustrative — confirm every rate with the accountant.