Don't start with a moonshot. Find one repetitive, high-volume workflow, automate the decision inside it (not the whole job), keep a human in the loop, and measure the hours you save. That's the entire playbook for a first AI win that actually pays for itself.
Every week we talk to a small-business owner who's been told they "need AI." They rarely need what they've been sold. What they need is one specific, expensive problem solved — and AI is sometimes, but not always, the cheapest way to solve it.
This is the framework we use to find the projects worth doing. It's deliberately boring, because boring is what works.
1. Start with a workflow, not a technology
The mistake is starting from "what can AI do?" The right question is "where do we lose the most time or money to repetitive decisions?" Look for work that is:
- High-volume — it happens dozens or hundreds of times a week.
- Repetitive — the same kind of decision, over and over.
- Rules-ish — an experienced person could explain how they decide.
- Expensive — it eats senior time, or mistakes cost real money.
Triage, routing, categorisation, first-draft writing, data entry, summarisation. Unglamorous, and exactly where the money is.
Automate the decision inside the job — not the whole job.
2. Keep a human in the loop (at first)
The fastest way to lose trust in an AI system is to let it act unsupervised before it's earned it. We almost always start with the AI making a recommendation a human approves. Confidence is high? Auto-action it. Low? Flag it for review.
3. Measure one number
Pick the metric that maps to money before you build anything. Usually it's hours saved per week, or turnaround time, or error rate. If you can't name the number you expect to move, you're not ready to build yet — you're ready for a consultancy conversation.
4. Integration is the real project
The model is a commodity now. The hard, valuable work is connecting it to your data and the tools your team already uses, and making it reliable enough to depend on. Budget for that, not for the demo.
So, is your project worth it?
Three questions: Is it high-volume and repetitive? Can you keep a human in the loop while it earns trust? Can you name the number it should move? Three yeses, and you've probably found your first win.
Want a second opinion on where AI fits in your business? That's exactly what our first call is for — and it's free.
Frequently asked questions
If you have a repetitive, high-volume decision that an experienced person could explain the rules for, and you can name the number it should move (hours saved, turnaround, error rate), you're ready. If you can't name that number yet, start with a consultancy conversation, not a build.
Automate the decision inside the job, not the whole job. Have the AI make a recommendation a human approves at first — auto-action only the high-confidence cases. This is the difference between a tool your team trusts and one they quietly stop using.
Triage, routing, categorisation, first-draft writing, data entry and summarisation. They're unglamorous and high-volume — which is exactly where the money is for a first project.