Where to Start With AI If Your Manufacturing Business Has Never Used It
Start with one task, done by one person, that takes real time every week and doesn’t touch a customer or a specification. Run it for a month. Then decide anything else.
That’s the whole answer, and most of the reason manufacturers struggle with this is that it’s not the answer anyone selling to them gives. The pitch is always a platform, an assessment, or a strategy. The thing that actually works is embarrassingly small, and it works because it produces evidence instead of opinions. What follows is how to run that sequence properly. It’s part of a broader look at practical AI and automation for manufacturers.
Do you need an AI strategy first?
No. You need one working example first, then a strategy informed by what you learned.
This is backwards from how most business advice is written, and there’s a specific reason it applies here. An AI strategy written by a team with no hands-on experience is a document assembled from vendor material and conference talks. It will be confidently wrong about what’s easy, what’s hard, and what your people will actually adopt. Six weeks of a real person using a real tool on a real task teaches you more than any planning exercise, and it costs almost nothing.
The strategy conversation is worth having. It’s worth having second.
How do you pick the first thing to try?
Choose a task that meets four conditions. If a candidate fails any one of them, pick something else, because the failure mode is predictable and it will get blamed on AI rather than on the selection.
It happens repeatedly. Weekly at minimum. A task that comes up twice a year won’t generate enough repetition for anyone to build a habit or for you to judge results.
Someone can name how long it takes now. You need a before number. It doesn’t have to be precise. Without it, the after conversation dissolves into impressions.
The input already exists in writing. Documents, emails, transcripts, spreadsheets, records in a system. If the necessary information lives in an experienced person’s head, the tool will produce something plausible and wrong, which is the worst possible first impression.
Being wrong is cheap and visible. A bad meeting summary costs a minute to correct and the error is obvious. A bad quote costs a production run and the error may not surface until the customer opens the box. Start where mistakes are recoverable.
Candidates that usually meet all four in a manufacturing business: summarizing sales calls or internal meetings, drafting routine customer correspondence for someone to edit, pulling structured fields out of incoming RFQ documents, first-pass research on a prospect before a call, and turning a rough operational note into a clean written procedure.
Candidates that usually fail at least one: anything touching specification or pricing accuracy, anything that goes to a customer without review, and anything requiring three systems to talk to each other before it works. That last category is where budgets disappear, because the integration becomes the project and eighteen months later you have infrastructure and no result.
Who should run the first project?
Pick a curious person with real workload, not your most technical person and not your most senior one.
Technical staff tend to evaluate the tool. Senior staff tend to delegate it. What you want is someone who feels the weekly pain of the task personally, will actually use the thing on a Tuesday when they’re busy, and will tell you honestly when it isn’t helping. Enthusiasm matters more than credentials here, because the failure mode of every early attempt is quiet abandonment rather than dramatic breakdown.
Give that person explicit permission to spend time on it and explicit permission to conclude it doesn’t work. A pilot that can only succeed will report success.
What does the first ninety days look like?
Weeks 1-2: Baseline and setup. Write down how the task is done today and roughly how long it takes. Pick one tool. Don’t compare six. The differences between mainstream options matter far less at this stage than getting started at all.
Weeks 3-6: Use it on real work. Not test cases. Real work, with real consequences, reviewed by a human before it goes anywhere. Keep a running note of what worked, what needed heavy correction, and what was faster to do manually.
Weeks 7-8: Evaluate honestly. Time saved, quality compared against the old way, and whether the person would keep using it if you stopped asking. That third measure predicts adoption better than the first two.
Weeks 9-12: Decide and widen or stop. Either extend the same use case to more people, or pick a second narrow use case, or conclude it wasn’t a fit and take the lesson. Stopping is a legitimate outcome and treating it as failure is how organizations learn to hide results.
Notice what isn’t in that timeline. No platform selection, no integration work, no committee, no vendor engagement. Those may come later. They’re not how you start.
What has to be true underneath?
Two things, and this is where most manufacturing AI projects actually die.
Your data has to be trustworthy enough for the task at hand. Not perfect, and not consolidated. Trustworthy for the specific job. Analyzing customer buying patterns requires customer records that reflect reality. Summarizing a meeting requires nothing but the recording. Match the ambition to the state of your data rather than launching a cleanup project you’ll abandon in month five.
The underlying process has to work when a person does it. If quotes go unfollowed because nobody owns follow-up, automating follow-up produces automated messages nobody’s accountable for. If your CRM is untrusted, an AI reading it produces confident conclusions from bad inputs. The tool amplifies whatever process it lands on, including the absence of one.
That second point is the one worth sitting with. A significant share of what manufacturers hope AI will solve turns out to be a revenue system problem wearing a technology costume: no defined owner for a stage of the customer path, no trigger that surfaces work at the right moment, no number anyone watches. Those are fixable, and they’re fixable independent of any tool.
What about the people question?
Say plainly what the goal is, because the team will assume the worst if you don’t.
In a mid-size manufacturer, the realistic near-term effect of these tools is that experienced people spend less time on administrative work and more time on the work you actually hired them for. That’s a defensible thing to say out loud, and saying it removes most of the resistance you’d otherwise spend months working around.
What undermines this is announcing an efficiency initiative and then declining to say what efficiency means. People fill silence with the least reassuring available explanation, and they do it fast.
Also worth naming: your team is probably already using these tools. Some of them are pasting customer information into consumer chatbots right now without any guidance from you. A short written policy covering what may and may not be shared with outside tools is overdue at most manufacturers, and it costs an afternoon.
What to avoid in the first year
Buying a platform before proving a use case. The order matters. Evidence first, then infrastructure.
Starting with the hardest problem. The instinct is to point new technology at your biggest headache. Your biggest headache is usually complex, high-stakes, and dependent on judgment your best people carry in their heads, which describes the conditions under which these tools perform worst.
Running it as a committee initiative. Steering groups produce documents. One motivated person using a tool on real work produces information.
Treating a failed pilot as proof the category doesn’t work. One bad selection tells you about the selection.
Skipping the sales process question entirely. Much of the near-term value for a manufacturer sits in the revenue operation rather than on the floor. For a specific breakdown of where these tools help and hurt in a sales context, see where AI actually helps a manufacturing sales process.
For the full framework on where automation and AI fit in a manufacturer’s operation, see our complete guide to practical AI and automation for manufacturers.
Frequently asked questions
How much should a manufacturer budget to get started with AI? Less than most expect for a first use case, since mainstream tools are priced per user per month. Budget for the person’s time rather than the software. Larger commitments belong after you have evidence.
Do we need to hire someone to run AI? Not to start. A curious existing employee with permission to spend time on it is the right first move. Hiring ahead of a proven use case tends to produce a role in search of a mandate.
Should we clean up our data first? Only as much as the specific use case requires. A general data cleanup project with no application attached is a common way to spend a year and produce nothing anyone uses.
Is AI worth it for a company under $10M in revenue? For narrow tasks, yes, because current tools are inexpensive and require no infrastructure. For large platform investments, usually not yet at that size.
What’s the difference between AI and automation? Automation follows rules you write. AI generates output from patterns, so it can handle situations you didn’t anticipate and can also be wrong in ways a rule cannot. Most manufacturers get more near-term value from automation, because most of their gaps are missing rules rather than missing intelligence.
Pick a first project worth running
The hardest part of this isn’t the technology. It’s choosing a starting point narrow enough to prove something and useful enough to be worth proving. Schedule a Discovery Call and we’ll help you pick the first one, using how your business actually runs today.