The best first AI project for a $10M to $150M manufacturer or distributor is a task someone on your team already does often, usually weekly, that you can measure today, using data you already have, where a person checks the result before it goes out. Score each candidate against the five questions below. Two of them are gates: if the data doesn’t exist or nobody can check the output, pick a different task no matter how the others score.
Most owners I talk to don’t need an AI strategy. They need one workflow that pays for itself. Here’s how I score the options.
The five questions
| Question | Why it matters |
|---|---|
| Does someone do this task every week? | Frequent work usually pays back faster, and people notice when it gets better. A rare task can still qualify when the stakes are high and the steps repeat, like an annual price review. |
| Can we measure it today? | Hours spent, days late, dollars missed. Without a before number there’s no after number. |
| Does the data already exist somewhere we can read? | ERP exports, spreadsheets, a shared drive and email are enough to start. You don’t need a data warehouse. |
| Can a person check the output before it goes out? | Prices, quotes and customer commitments stay with your people. |
| Can a working version be built in about 30 days? | Long first projects lose their sponsor. |
Four or five yeses: strong first project. Three: worth a closer look at the payoff. Two or fewer: keep looking. Either way, a no on the data or the human check rules it out.
Tasks that usually score well
- Watching for things that change. Commodity and input prices, competitor prices and stock, new public bids. Somebody used to check these when they had time.
- Pulling data out of documents. Supplier spec sheets, RFQs, equipment documents. Someone retypes it today.
- Building the same report every week or month. Forecasts, cost updates, board numbers.
- Spotting customers who went quiet. Accounts that stopped reordering on their normal rhythm.
- Prep work before a conversation. A one-page brief on an account before a sales call.
You can see working examples of several of these on the Work page.
Tasks that usually score badly as a first project
- Anything that talks to customers or changes prices on its own. Start with tools that draft and flag, not tools that act.
- Problems with no owner. If nobody does the task today, nobody will use the tool.
- “Clean up all our data first” projects. Part of a good pilot is figuring out what’s usable as it stands.
- Rare, high-stakes decisions. Once-a-year tasks don’t pay back a build.
What you need before you start
- Read access to wherever the work lives today: the ERP, CRM, spreadsheets or a shared drive
- 12 to 24 months of history for anything that forecasts or flags a change
- The person who does the task now, for about an hour a week during the build
- One person who decides what the tool is allowed to do
If customer or item IDs don’t match across systems, mapping them is part of the work, not a reason to wait.
Where AI breaks, so you can plan for it
- Customers who order rarely don’t have a pattern. A good tool says so instead of guessing.
- Stockouts and supplier outages can make demand look like it vanished.
- Matching your products to a competitor’s is wrong often enough that a person should review the matches.
When to skip AI
If a spreadsheet, a report your ERP already has, or a change in who does what fixes it, do that. I’ll tell you if that’s the answer.
Next step
If you have two or three candidate tasks and can’t pick, that’s what the Operational Diagnostic is for: two to three weeks with your team and your numbers, ending in three to five ranked use cases and a 90-day plan. If you already know the task, a Pilot Build gets you a working tool in about 30 days.
Manufacturers: where I’d start. Distributors: where I’d start. Or book a 30-minute call.
