If you run a company between $10M and $150M, my advice on AI strategy is simple: don’t start with a strategy. Start with the one workflow that costs you the most time or margin, build a tool for it on the systems you already have, and measure what changes. Then pick the next one based on what you learned. You don’t need a roadmap, a data warehouse or an AI committee. You need one task, one owner and a before number.
Why the big-company playbook doesn’t fit
Most AI strategy advice is written for large companies. It assumes a data team, an IT department and a year to get something working. A company your size usually has none of those. What you do have is a handful of people who know exactly where the hours go every week: re-keying orders, chasing accounts that have gone quiet, checking prices, building the same report every Monday morning.
My concern with the 40-page roadmap is that nobody owns most of what’s in it, so it tends to stall at the first project. One working tool that gives a real person their hours back does the opposite. It pays for the next project, and it teaches you more about your data and your team than any planning exercise will.
The one-page version
Start by asking each department head for the three tasks that eat the most hours or cause the most misses. You’ll usually have 15 to 25 candidates by the end of the afternoon.
Then score them. I look at five things: does someone do it often, can you measure it today, does the data already exist, can a person check the result before it goes out, and can a working version be built in about 30 days. A task with no data or no human check is out, no matter how big the payoff looks. From what’s left, pick the highest-scoring task that one person owns.
Build it on what you have. An export from your ERP, a spreadsheet or a shared inbox is usually enough to start. When it’s running, compare the before and after numbers and make a call: keep it, fix it or drop it. Then go back to the list and pick the next one.
I go deeper on the scoring step in how to pick your first AI project.
What usually makes the list
The same handful of tasks show up again and again. Watching for changes nobody has time to check, like commodity prices, competitor prices or new public bids. Spotting customers whose ordering has slipped before you lose the account. Pulling data out of purchase orders or supplier spec sheets that someone is retyping today. Building the weekly numbers with a short summary of what changed. Drafting routine writing that a person then reviews.
None of these are glamorous, and that’s the point. They’re where the hours and the misses are. You can click through working samples of several of them, built on invented data.
The rules that belong in the strategy
The one thing I’d write down before any build is the guardrails: which tools people can use, which data never goes into them, and what a person still approves. Prices and anything sent to a customer should always be on that last list. If you skip this, people will make up their own rules, and you won’t like all of them. There’s a one-page version in this AI rules template.
Where this approach breaks
It isn’t right for every situation. If nobody owns the task today, nobody will use the tool. If the data doesn’t exist, fix that first, because no build can invent it. If leadership expects AI to transform the business in a quarter, one workflow is going to feel too small, even when it’s the right first step. And if a spreadsheet or a setting in a system you already own solves the problem, do that instead and save the money.
Doing it yourself versus getting help
Plenty of companies can handle the list-and-score step on their own. Where I see it stall is the build, and then keeping the tool running week after week once the novelty wears off. That’s the part I do: a short Operational Diagnostic to rank the workflows, then a fixed-fee pilot on the one worth building first.
Book a 30-minute call and bring the task that eats the most of your team’s week.
