Recently, a manufacturer launched a new product line based on the fact that the category was their top-selling line. No market study, or forecasting model, or even a consulting deck was used to support this decision.
According to the manufacturer, only a handful of customers were buying this particular product, despite it being a top seller. This proved that there was strong demand for the product and little competition.
The company wanted to move before the category got crowded.
They didn’t use a tool to tell them what to do. Someone was paying attention to a conversation most companies treat as small talk.
While many small and mid-market manufacturers may have a good tech stack, the mistakes stem from timing. A company finds a product line that sells well. They get known for it, so they keep producing it long after demand has flattened. Then sales slip, and the whole company turns into a fire drill as they scramble for the next big thing. Meanwhile, they are sitting on overstock that won’t move. Today, they may look for an AI tool to make sense of the mess.
Some manufacturers are growing by doing it right. They saw what was coming three-quarters ago from signals they were already collecting.
By the end of 2027, more than 40% of agentic AI projects will be abandoned, according to Gartner. Most of those projects will fail because the data underneath them can’t be trusted. Real optimization starts earlier: knowing what’s moving, what’s cooling off, and listening to what vendors are telling you about the market before your own sales data catches up. The software comes later.
Most SMB manufacturers can’t see the big picture because they aren’t set up to assemble it. The data is there: they have the open purchase orders, inventory turns, vendor conversations and slow-moving SKUs. But it sits in spreadsheets, email threads and in the head of company leaders. Nobody is being paid to pull it together, and by the time someone does, the window is closing.
This is also why bolting AI onto an SMB manufacturing operation often yields nothing useful. Deloitte’s 2026 Manufacturing Industry Outlook found that 93% of companies’ AI investments go to the technology itself. Only 7% goes to the people and processes underneath. Even though the tools are new, the foundation is the same.
That 7% buys the unglamorous work. Someone who owns data hygiene. Having sales and purchasing agree on part numbers. Making a habit of taking summaries of vendor conversations and feeding them back into demand planning. None of those are technology problems. They’re operating decisions, and they decide whether the AI purchased and implemented produces something with insight that you can act upon.
An AI tool that reads inventory data is only as good as the inventory data itself. If lot codes drift or bills of materials don’t match what’s actually on the production floor, the output will sound confident but scratch under the surface, and the insight will be fuzzy. Missing the big picture is worse – you may act on insight that’s just plain wrong. The companies doing this well treat AI as a layer to help them see straight. They earn the right to use it by first cleaning the underlying data.
The foundational work is specific. Reconcile what the system says you have against what’s actually in the warehouse. Get sales, purchasing and production to agree on a single set of part numbers. Know the Top 20 SKUs by margin contribution — not just by revenue — and watch the trend monthly. Treat vendors as market sensors, not only fulfillment channels. None of that is glamorous. All of it determines what happens when the curve flattens on the next bestseller.
The same logic applies to timing. Sales data is the most familiar number on the dashboard, and the slowest to turn. The earlier signals show up months before the revenue line bends. Inventory turn is slowing on a line that used to move weekly. Listen for a vendor mentioning new buyers in an adjacent category. Note when a customer is cutting order size but ordering more often. Make reading those signals a habit.
Before you evaluate another optimization tool, audit the data you already have. Can you count what’s in your business — what’s moving, what’s not, what your vendors are flagging — without making the answer up? Would the people on the floor bet their bonus on the numbers? Could your purchasing team defend the open PO list? When did anyone last walk into the warehouse and reconcile the shelves against the system? If the honest answer is no — or “a while ago” — the next piece of software won’t fix it. It will inherit the problem and run faster.
The manufacturer mentioned earlier launched the new product line without any proprietary forecasting model. They had a vendor with something to say, and somebody on their team who was listening. Somewhere in your own vendor list is the next move you should be making or the one you should walk away from. It can’t be found on a dashboard. Get your data right, then AI will help you soar.
