Find Your
First
AI Win
The first AI programme has to survive more than a demo. Operations needs an owner, finance needs a credible value case and IT needs to know what data and systems it will touch. We choose a use case all three can support and measure.
95%
OF GENAI PILOTS RETURN ZERO
1
USE CASE, PICKED PROPERLY
Baseline first
VALUE MEASURED BEFORE BUILD
A first AI project needs a clear business case
MIT Project NANDA analysed 300+ public AI initiatives in 2025 and found 95% had no measurable P&L impact. Gartner had already warned that at least 30% of GenAI projects would be abandoned after proof-of-concept by end of 2025. McKinsey's March 2025 State of AI says 78% of organisations use AI in at least one function, but only 17% say 5%+ of their EBIT is attributable to GenAI.
The pattern's familiar. A board asks for AI. Someone picks a use case that sounds interesting, not one that's worth building. The slide deck looks fine. Nothing changes on the invoice run.
The first one matters because it teaches the business what a useful AI project looks like.
TYPICAL FIRST PROJECT
- Chatbot on the website, nobody asked for
- "AI for everything" platform, six-figure licence
- Picked by IT, not by the team doing the work
- Data isn't ready, so the demo is faked
- No P&L owner, no number it has to move
- Stops after the pilot
A FIRST WIN
- A task done daily, by a real team
- Bespoke build, sized to the saving
- Picked with the people who do the work
- Uses data you already have, cleanly
- One number it has to move, named up front
- A contained first phase with a credible payback case
How we choose the first project
We score each candidate on financial value, delivery size, available data, failure risk and who will own it. Weak ideas leave the list before they consume a quarter.
Visible on the P&L
Hours saved, deals closed, costs avoided. If finance can't point at the line it's moved, the project shouldn't be the first one.
Small enough to test quickly
If the first version takes months, you've picked something too big. Split it down until it can be tested quickly, or pick a different job.
Data you already have
First wins use data that exists, in a system you already use. If it needs a new data warehouse first, it's a second-year project.
Safe to get wrong
A human checks the output, or the cost of an error is small. The team needs room to spot what's working without a board-level incident every Friday.
An owner who wants it
A named operator whose number it moves. Not IT, not a vague business owner. The person who's emailed about it on a Monday morning.
What the leadership team gets
A focused review produces a scored shortlist and an evidence plan for the strongest candidate. The business approves the first implementation against that value, risk and readiness case.
We don't sell licences and don't earn referral fees from model providers. The right answer is whichever AI tool, or no AI at all, moves the number.
BOOK A DISCOVERY CALLDiscovery
We work with the affected teams, observe the process and review the systems they use. Candidate use cases are recorded in the same format so leadership can compare them.
Score and shortlist
Every candidate's run through the five tests and plotted on value against feasibility, like Gartner's use-case prism. You get the full list with our reasoning, a top three, and a recommendation for which one to do first. If the right answer's no, we'll say so.
Launch the first one
The first phase has a defined user group, workflow, systems, baseline and stop criteria. It runs on representative business data and is measured against the agreed result.
Hand over a real roadmap
After the first project is live, you have a baseline and a team that knows how to test the result. We use the same scoring criteria for the next year's plan and leave them with you.
Why the first result matters
Four numbers from credible research. They all point at the same problem: the use case has to be chosen for value and feasibility before anyone builds.
95% return nothing.
MIT Project NANDA's State of AI in Business 2025 analysed 300+ public AI initiatives. 95% delivered no measurable P&L impact, despite $30bn to $40bn of spend.
30% killed after PoC.
Gartner: at least 30% of GenAI projects will be abandoned after proof-of-concept by the end of 2025. Cited reasons: poor data quality, weak risk controls, costs, unclear business value.
78% adopt, 17% see 5%+ EBIT.
McKinsey's State of AI (March 2025): 78% of organisations use AI in at least one function. 17% of respondents say 5% or more of their EBIT is attributable to GenAI.
14% of micro firms, 36% of large.
DSIT's AI Adoption Research found larger UK businesses were more likely to use AI, but most adopters had yet to see a revenue change. A first programme needs its own value case rather than relying on the adoption trend.
Sources: MIT Project NANDA (State of AI in Business 2025); Gartner press release, 29 July 2024; McKinsey State of AI, March 2025; UK Department for Science, Innovation and Technology, AI Adoption Research.
People and data matter more than the model
BCG puts the split at roughly 10/20/70. Ten per cent of the value comes from the algorithm. Twenty comes from the data and technology. Seventy comes from people and process.
If you've already paid a vendor for the ten per cent and nothing's moved, that's why. The other ninety per cent is what we pick for, and what we build around.
McKinsey's March 2025 numbers say the same thing: out of 25 attributes tested, workflow redesign had the biggest effect on EBIT impact from GenAI.
The model
The bit everyone talks about. Pick a sensible one, move on.
The data and plumbing
Get the inputs clean, the outputs into the system the team already uses.
The people and the workflow
Who does what, when. What gets cut, what gets kept, who checks the output. This is where first wins are won and lost.
The first useful workflow is usually very specific
For AM2PM, candidate assessments had to return graded reports to the recruitment CRM. For Crystal, one deal had to pass the same compliance checks with the evidence retained. Both started with a defined job and owner.
When this is worth discussing
We work best when there is a real operating problem, enough volume to measure and people from the affected teams who can make decisions.
Usually a good fit
- An established UK business, usually with annual revenue above £10m
- A repeated process with a known cost, delay, error rate or capacity problem
- A senior sponsor and a day-to-day owner who understand the work
- Access to the relevant staff, systems, sample records and security requirements
We may point you elsewhere
- A standard product already covers the process well
- The requirement is a one-off small build with no wider operating case
- There is no owner or access to the people and data needed to test the result
- The plan relies on AI making high-impact decisions with nobody responsible for review
Questions before committing
We've already done a pilot. Is this for us?
Especially. A previous chatbot, Copilot rollout or no-code agent is useful evidence. The first win is the first one that moves a number, which is rarely the first one anyone tried. The previous attempt shows what your business will and won't use.
How is this different from a big consultancy engagement?
This is narrower. One working system, on your data, with a number it has to move. We do the discovery and the build ourselves, so the advice has to survive contact with the actual workflow.
What if AI is the wrong answer here?
Then that's what we say. A lot of the time the answer's a small internal tool, a tidied-up workflow, or a piece of automation that's barely AI at all. We aren't paid commission by model providers, so we'll happily recommend the cheapest thing that moves the number.
Our data's a mess. Should we sort that out first?
The first project runs on the data you have, with the cleaning that job needs. A company-wide data cleanup would delay the result and create a much larger programme.
What size of business is this for?
Established UK businesses, usually above £10m revenue, with repeated workflows, named department owners and enough volume to measure the result. Very early businesses are usually better served by an off-the-shelf tool.
Who needs to be in the room?
We need a leader who can approve the priority, the department owner and people who perform or receive the work. Finance, IT, privacy or security join where their decisions affect the use case.
How much does it cost?
The review and first implementation are scoped separately. The proposal states the delivery cost, expected running cost, dependencies and the saving or revenue assumption used in the investment case.
What happens after the first win?
You keep the shortlist, scoring and baseline, so the next decision does not depend on hiring us again. Your team can run the next use case or ask Vu to scope it as a separate phase.
Talk to us about the first AI project
Tell us what has already been tried, where the cost sits and which measure leadership wants to change. We will test whether there is a suitable first programme and what evidence would justify it.