How to use AI in manufacturing, one business problem at a time.
By Chris McCarthy, Founder & Chief AI Officer, OB.1
You've decided to try AI. Now the question is practical: how to use AI in manufacturing without turning it into a big, messy program. This piece gives you a simple working method you can run with your team. Still picking where to begin? Read my starting guide first.
How to use AI in manufacturing without a big program
The method has five parts. Each one is small on purpose.
- Choose one business problem. Make it specific, like slow replies to customer emails.
- Name one owner. That person answers for the trial and reports back.
- List the data the tool needs. Also list the data it must never see.
- Run a small trial for a short, agreed window.
- Check the results against a measure you set before you started.
There's no company-wide rollout and no long vendor search. When one trial works, you move to the next problem.
Company-level uses worth a look
Here are uses a leader can weigh from the executive seat. Each one ties to a process and needs an owner.
- Drafting replies to customer emails. The tool writes a first draft. Someone on the team edits and sends it.
- Summarizing long specs or contracts. The tool pulls out the key terms. A reviewer confirms them against the source.
- Pulling weekly reports from the ERP. The tool gathers the numbers into one page. The owner checks them before the meeting.
- First drafts of quotes. The tool prepares a starting point. An estimator checks every line. More on this in the quoting guide.
- Sorting incoming orders. The tool flags orders that need a second look. A person makes the call.
Every one of these carries the same guard. A named person reviews what the tool produces. Nothing goes out to a customer or supplier until that review is done. The tool drafts. Your people decide.
Set the owner, the data and the limits before the trial
Before anyone opens a new tool, write three things down.
First, the owner. Pick the person closest to the work. They know what good output looks like, and they'll spot errors fast.
Second, the data the tool may use. Be specific. Name the folder, the report or the inbox.
Third, the data it must not touch. Customer drawings, price lists and anything covered by a contract usually belong on that list. Your own contracts and policies decide the rest. For help writing those rules, see the AI rules guide.
Bring the people doing the task into this step. They should help pick the trial and help judge it. Their judgment is what makes the tool useful. Want an outside view of this setup? A Blueprint Session covers it.
Run a small trial and judge it plainly
Pick a window your team agrees on, and keep it short. Agree up front what "better" means for your company, in your own terms. It might be time saved on a task, fewer cases of typing the same data twice, or faster replies to customers.
Write that measure down before the trial starts. Then have the owner track it and report back. Plain notes are enough. You don't need a special system to judge a small trial.
At the end, sit down with the owner and the team. Ask what worked, what didn't and what surprised them. Ask the people who used the tool whether they would keep it. Their answer matters as much as the measure. Write down what you learned, even if the trial stops, so the next trial starts smarter.
Decide what to do next
A trial ends in one of three ways.
- Keep it. Write the new step into how the work runs, with the owner and the review step named.
- Change it and try again. Adjust the task, the data or the prompt, then run a second short trial.
- Stop. Some problems are better solved without AI. Stopping is a fine answer, and it costs little when the trial was small.
Then pick the next problem. Want a clear view of where to look next? The mapping guide shows how to lay out a full process with your team.
Does AI make sense for a smaller manufacturer?
Yes, and the method is the same. In a smaller company, one person often covers several roles. A well-chosen trial can give some of that time back. Start with a narrower first trial. Pick one inbox, one report or one type of order.
Keep the owner close to the work, and keep the data list short. A smaller company can often move faster, because fewer people need to agree.
How OB.1 helps
Some leaders want help choosing that first problem and setting up the trial. We begin by writing out how one area of your business runs. That includes who owns each part and where the data sits. Then we lay out a plan for what to try first. We call it an Operational Blueprint.
It begins with a Blueprint Session. You can read what the session covers on the Blueprint Session page.
Using AI in manufacturing: what leaders ask
What is a practical first use of AI in a manufacturing business?
Pick one office task with a clear owner, such as drafting replies to customer emails. Run a short trial, and have a person review every output.
Which AI uses make sense for a manufacturing company's front office?
Drafting emails, summarizing specs or contracts, pulling ERP reports, first drafts of quotes and sorting orders are all worth a look. Each one needs an owner and a review step.
Can a smaller manufacturing company use AI?
Yes. Use the same method with a narrower trial. One owner, a short data list and a clear measure are enough to start.
How do we know if an AI trial worked?
Set the measure before the trial begins. At the end, compare the results to that measure with the owner and the team.