If you run a manufacturing business between 20 and 100 employees, you've probably been told AI will change everything about your operation. You've also probably ignored most of that, because the people writing it have never spent time on a shop floor or talked to your customers.
This is the other version. We recently sat down with the operations lead at a manufacturer we work with — a company that makes physical products, ships from their own warehouse, drop-ships some items from vendors, and serves both consumer and commercial customers. We talked through what was actually slowing them down. Five themes came back over and over. They map cleanly to things AI can do today, in production, for businesses like theirs.
None of these require ripping out your ERP. None require a technical team. All of them are real.
1. Answer order status questions — including the hard ones
The most common message coming into customer-facing channels at this manufacturer was some version of: "Where's my order?". The team's process for answering it was three steps: open the ERP, look up the order, copy the tracking info, paste it into a reply. Multiplied across dozens of inquiries a day, this was a meaningful share of someone's week.
That's the easy version. AI handles it well, in seconds, by reading the message and pulling live data directly from the ERP.
The harder version is partial shipments — when some items shipped from your warehouse and others are drop-shipping from a vendor whose ETA is unreliable. The customer gets a box, sees three of five items, and writes in. A human handling this has to look at the order, identify what shipped, identify what didn't, check vendor status, and craft a response that explains the situation without overpromising on the missing pieces.
AI handles this, too — it's just a more carefully designed version. It pulls the line items, identifies shipped vs. pending, surfaces vendor info where reliable, and explicitly flags uncertainty rather than guessing. When the answer requires actual judgment ("the vendor is two weeks late and the customer is angry"), it escalates with full context.
2. Answer product spec and compatibility questions
The second category was product questions. What are the dimensions of this part? Does this replacement piece fit the model I bought three years ago? What finish options does this come in?
For most manufacturers, this information already exists — but it's scattered. Some lives in ERP item attributes. Some on the website product page. Some in spec sheets sitting in a shared drive. Some, honestly, in the heads of two or three people in the warehouse who've been there long enough to know.
AI is unusually good at this kind of work — pulling from multiple sources, reconciling them, and answering a customer's actual question rather than dumping a spec sheet at them. The implementation involves connecting it to the existing data sources (ERP, website, spec sheet library) and giving it explicit rules about when to cite the source vs. summarize.
For the manufacturer we talked with, an estimated 60–80% of inbound product questions could be answered from data they already have. The other 20–40% — the ones that genuinely require asking someone in the warehouse — get escalated, and the AI captures the answer once it's given so it doesn't have to ask again.
3. Triage by customer value, automatically
Not every customer message deserves the same treatment. A consumer asking whether a small part is in stock is a different urgency than a commercial contractor asking for a quote on a $20,000 install.
The team we talked with described this in clean terms: under about $1,000, the question is great for AI to handle. Above that, it should go to sales — fast, with full context.
AI handles this kind of triage well. It looks at who the customer is (returning contractor account vs. first-time consumer), what they're asking about (small part vs. large quote), and routes accordingly. Small inquiries get an instant, helpful response. Large inquiries get a Teams or Slack notification to the right person with a summary of what the customer wants and what context the AI has already gathered.
The win here isn't just speed. It's that high-value leads stop getting buried in the same queue as routine product questions, which is what kills speed-to-lead at most growing businesses.
The instinct is to think AI either replaces a person or doesn't. The more useful frame: AI handles the work that doesn't need human judgment, and routes everything else to the right human, with the right context, faster than you could do manually.
4. Surface what your data is missing
This is the one nobody talks about, and it might be the most valuable.
When you put AI in front of customer questions, it doesn't just answer the ones it can — it logs the ones it can't, and why. Over a few weeks, you get a clean list of: here are the questions customers are asking that we don't have good answers to in any system. Here's the product data that's missing. Here's where the website disagrees with the ERP. Here's the compatibility info that lives only in someone's head.
For most manufacturers, this list is gold. It tells you exactly which item attributes to fix in the ERP, which spec sheets to write, and which tribal knowledge to capture before the person who has it leaves. You'd never invest the time to compile this list manually. AI generates it as a byproduct of doing its job.
This is where AI starts paying for itself a second time — not by saving labor on customer questions, but by giving you a prioritized, real-world list of what to fix in the systems your business runs on.
5. Bridge the tools your team actually uses
Most manufacturers we talk to have the same problem: customer messages live in one tool (a website chat, an email inbox, a HubSpot queue), the team works in another (Microsoft Teams or Slack), and the data lives in a third (the ERP, plus a website CMS, plus a half-dozen spreadsheets). Nothing talks to anything else, and someone — usually whoever has a free moment — is the bridge.
AI is well-suited to sit in the middle of this. It picks up the customer message wherever it comes in. It pulls the data it needs from wherever that data lives. It posts escalations into the tool the team actually checks (Teams, Slack, whatever). It logs the resolution back to the ERP or CRM so the record stays clean.
The point isn't replacing your tools. It's stopping the human from being the integration layer between them.
The honest caveat
There's a version of this post that ends with "and that's why every manufacturer should adopt AI tomorrow." This isn't that.
The operations lead we talked with said something at the end of our call that we keep coming back to: "This will only work if our data is accurate." She's right. AI inherits the quality of the systems it pulls from. If your ERP has half-correct item attributes, AI will give half-correct answers. If your spec sheets are three product generations out of date, AI will confidently quote outdated specs.
The implication isn't that you need a perfect data house before you start. It's that the right starting point is usually narrower than the full vision — pick the area where your data is cleanest (often: order data, because the ERP enforces it), launch there, use the AI's logs to find and fix data gaps in adjacent areas, and expand. That sequencing matters more than any single technology choice.
For the manufacturer we talked with, the starting point is order status and partial shipments. Then product compatibility. Then routing by customer value. The data quality work happens in parallel, informed by what the AI surfaces. That's a realistic plan. The "AI everything in one project" version is not.
Curious what this looks like for your shop?
That's exactly what our free call is for. We'll look at your operation, your systems, and where your team is spending time on questions AI could handle — and tell you where to start.
Book a Free Call