Your first AI use case: how a small team gets one project to pay off
Most small teams start five AI pilots and finish none. Here is how to choose one first AI use case, tie it to a number, and actually finish it.
By Carl Chessum
The fastest way to waste a year on AI is to start five things and finish none. It is the most common pattern I see in small businesses. A tool gets trialled here, a pilot runs there, someone tests a chatbot, and twelve months later there is activity everywhere and a measurable result nowhere. The fix is not more pilots. It is one use case, chosen properly and finished: a single project, tied to a number, owned by someone, run to the point where you can say plainly whether it paid off.
The scattering is expensive. 42% of companies abandoned most of their AI initiatives in 2025, up from 17% the year before (S&P Global Market Intelligence, 2025), and 95% of AI pilot programmes fail to create measurable value (MIT NANDA Research, 2025). Those are not stories about bad technology. They are stories about businesses that started more than they could finish, and never tied it to an outcome they could measure.
Why do first AI projects stall?
Because nobody owns them, and nobody agreed what success looked like before they began.
A first AI project usually starts as enthusiasm, not a decision. Someone is keen, a tool looks promising, a trial begins. There is no named owner with the authority to see it through, no agreed number it is meant to move, and no point at which it is declared done. So it drifts. When attention moves to the next tool, the half-finished one is quietly dropped, and the lesson recorded is “AI did not work for us”. The patterns behind that are the subject of a companion piece, why AI projects fail. This post is the opposite argument: how to get one to work.
The strategy side is simple to state and hard to do. Decide what the project is for, in one sentence, before you start. If leadership cannot agree that sentence, the project is not ready, and no tool will rescue it. The Strategy pillar is about exactly that discipline.
How do I choose my first AI use case?
Pick the one that scores well on four things at once: a clear owner, a measurable outcome, clean enough data, and low dependency on other systems or teams.
Most businesses choose their first use case on excitement, which is how you end up automating the interesting thing instead of the useful one. Score your candidates honestly instead. List three or four jobs AI could help with, and rate each on the four criteria below.
| Candidate use case | Clear owner? | Measurable outcome? | Clean enough data? | Low dependency? |
|---|---|---|---|---|
| Drafting first-response emails | Yes | Reply time, hours saved | Yes | Yes |
| Lead scoring across systems | Unclear | Conversion rate | No | No |
| Summarising support tickets | Yes | Handling time | Mostly | Yes |
The one that scores well across all four is your first project. A use case that cannot tick “measurable outcome” is a hobby, not a project. One that cannot tick “clean enough data” will produce confident nonsense, which is the cheapest mistake to fix first. And one that sits on top of an undocumented process will automate the gaps nobody wrote down. That is why a first project should sit on a process you can describe end to end, which is the Process pillar in practice.
How do I run it from pick to payoff?
You do not need a project office. You need one owner, one number, and a finish line. Here is the sequence.
- Name the number first. Before any tool, write the sentence: “By [date], this will move [specific metric] from [baseline] to [target].” If you cannot fill the blanks, you are not ready to start.
- Name one owner. The person who does the work and has the authority to change how it runs. Not a committee. One name.
- Map the process as it actually runs. The real version, including the exceptions and the steps nobody wrote down. AI built on the documented fiction fails where the two disagree.
- Scope it small enough to finish. One task, one team, one month. A first project you can finish beats a grand one you cannot.
- Run it against the baseline. Use the tool on the real work for a fixed period. Measure the number you named, not how the tool feels.
- Decide in the open. Keep it, widen it, or stop, based on the number. A clear stop is a result, not a failure. A clear payoff is what earns the budget for the second project.
You can do all of this without a tech team, with the people and tools you already have. The detail on that is in AI without a tech team.
How do I know if it actually paid off?
You know because you wrote the number down before you started, and you can now compare it to the baseline. That is the whole test.
The reason most businesses cannot answer whether a project paid off is that they never set the number, so the answer becomes an opinion, and a generous one. A finished project gives you a fact instead. It moved the metric, or it did not. Either way you learned something you can act on, and you earned the right to the next one. Only 26% of organisations successfully transition AI from proof-of-concept to production (BCG, 2024). The ones who do are not luckier. They finished one thing properly before starting the next.
When the use case is genuinely bigger than one team and one month, that is the point to bring in help, not before. Our bespoke transformation engagement exists for the projects that have outgrown a single owner and a single month, once you know exactly what you are asking for.
Start with one honest score
Before you pick a use case, it helps to know where your foundations stand, because the project will sit on them. The AI Readiness Audit scores your strategy, process and the four other foundations in about seven minutes. 30 questions, an honest band for each pillar, no card and no sales call.
To see how those scores fit the wider picture, the AI Readiness Audit overview walks through all six pillars and the three ways to run it.
Frequently asked questions
How do I choose my first AI use case?
Score your candidates on four things at once: a clear owner, a measurable outcome, data that is clean enough to trust, and low dependency on other systems. List three or four jobs AI could help with, rate each honestly, and pick the one that scores well across all four. The exciting use case that depends on messy data and three integrations is not your first project, however appealing it looks.
What makes a good first AI project for a small business?
Small enough to finish, measurable, owned by one person, and sitting on a process you can describe end to end. A good first project moves a single named number within about a month. Grand projects that touch everything and finish nothing are the most common way a first attempt at AI gets written off.
How do I know if my first AI project paid off?
Write the number down before you start: which metric, from what baseline, to what target, by when. At the end, compare the result to the baseline. If you never set the number, whether it paid off becomes an opinion, and the answer is usually too kind. A finished project gives you a fact you can act on, whichever way it goes.
Why do most first AI projects stall?
Because no one owns them and no one agreed what success looked like before they began. They start as enthusiasm rather than a decision, drift without a finish line, and get abandoned when attention moves to the next tool. Naming one owner and one measurable outcome before you start removes the two most common causes of stall.
Carl Chessum is the founder of AI Readiness Partner and the author of AI Readiness for Marketing Leaders, available on Amazon. He has spent 25 years inside transformation programmes, client-side and consultancy-side, across PLC, VC-backed and private-equity-backed businesses.