By 2026, most small businesses we talk to have either tried "AI" once and given up, or are paying a consultancy for a strategy deck nobody implements. Both outcomes have the same root cause: the project was framed as transformation when it should have been framed as doing one specific boring job well.
The teams who actually get value from AI - the ones who, a year later, can point at hours saved or revenue earned - all start the same way.
Pick the most repetitive thing you do.
Not the most strategic. Not the most exciting. The most repetitive.
Look at last week's calendar and inbox. What did you do more than three times that took you more than ten minutes each time? That's your candidate.
For most small business owners we've worked with, the list looks something like:
- Drafting the same kind of reply to the same kind of email
- Copy-pasting between two tools that should talk to each other
- Researching prospects before calls
- Writing weekly social posts
- Pulling end-of-day numbers out of three systems
- Triaging inbound enquiries
None of those will land you on a magazine cover. All of them, automated or AI-assisted, free up real hours - the kind that compound into actual capacity.
Build a tool that does that one thing.
Not a "platform". Not a "system." A tool. One job, one interface, one obvious place to look when it breaks.
If the boring job is "drafting replies to enquiry emails," the tool is a small AI assistant trained on your past replies that drafts new ones for you to edit and send. If it's "researching prospects," it's a one-click summariser that takes a company name and produces a short brief from public sources. If it's "weekly social posts," it's a content generator that drafts five posts in your voice from a single bullet-point input.
Each of these is small. Each of them ships in days, not quarters. Each of them is reversible if you decide it isn't worth it.
Live with it for a fortnight.
Two weeks is the right test window. Long enough to actually use the thing on real work, short enough that you'll spot if you're tolerating something annoying.
At the end of two weeks, ask: has this saved me time, and would I be sad if it stopped working? If yes to both, keep it and move to the next boring job. If no, kill it - that's not a failure, that's a successful experiment.
Then - and only then - do the next one.
Compounding is the entire game. One small AI tool that saves an hour a week is worth about a working week a year. Five of them is a junior hire's worth of capacity, with no recruiter, no salary, no NI.
But you only get the compounding effect if you don't try to do them all at once. Pick the boring job. Build the tool. Live with it. Ship the next.
The "human in the loop" rule.
For every AI tool we've built and kept, the human is in the loop where it matters. The AI drafts, the human ships. The AI suggests, the human decides. The AI scores, the human acts.
This isn't a hedge against the technology - it's the design pattern that actually works. AI is exceptional at the first 80% of repetitive cognitive work. People are exceptional at the last 20% that requires judgment, taste or nuance. Build tools that play to both.
What "ready" looks like.
You don't need a "data strategy." You don't need an enterprise contract. You don't need to know what RAG stands for. You need a boring job, two weeks, and a tool small enough to throw away if it doesn't work.
That's the practical AI playbook. Everything else is a slide.
If you've got a boring job and want to scope what an AI tool for it might look like, that's exactly the call we like having.