Building a Parametric BOM Calculator With AI Sizing Rules

Every custom manufacturer I've talked to has the same story. They looked at the big enterprise software systems, got sticker shock, and went back to their spreadsheets. That's exactly where a custom manufacturing client of mine was — a window and door shop doing somewhere between $500K and $2M a year, building made-to-order products where every single order has different dimensions. They needed a smarter way to figure out what parts go into each product and what it all costs. Nothing on the market fit.

The Problem Nobody Solves for Small Manufacturers

This client builds custom windows and doors. Every order is different. A standard window might have 47 parts. Change the width by 6 inches and 23 of those parts change — different lengths, different quantities, different costs. Every order is essentially a unique product.

Here's what their process looked like before we started. A sales rep takes the order, manually calculates each part size using memory and a reference spreadsheet, looks up material costs from three different suppliers, builds the parts list in Excel, and hand-writes instructions for the shop floor.

One parts list took 45 to 90 minutes. Mistakes were common. And every mistake that made it to the shop meant redoing the work — at roughly $200 a pop.

They'd looked at the big-name software. SAP. Oracle. The licensing alone would have eaten their margins. The mid-tier tools could handle simple products with fixed parts lists, but couldn't handle what this shop needed: a system where changing one dimension automatically recalculates everything else.

This is a gap nobody talks about. Small custom manufacturers are too complex for spreadsheets and too small for enterprise software. They fall right through the middle.

I'd seen this pattern before. When I built an AI production system for my own DTC fashion brand in San Diego, the lesson was the same: custom tools built around your actual workflow often outperform expensive off-the-shelf software for businesses under $5M. The software you need doesn't exist because your process is specific to you. So you build it.

What I Built and How It Works

Think of this system like a very smart calculator. You punch in the customer's window dimensions and options. The system instantly figures out every single part needed, the exact size to cut each piece, and the total cost — in about 20 seconds instead of 45 minutes.

Here's a real example. Customer orders a 48x72 window. The system automatically knows that a 48-inch wide window actually needs a 46.25-inch horizontal frame piece (because you have to account for where the corners join). It knows the glass isn't exactly 48x72 — it's slightly smaller to fit the seals. It knows that windows taller than 60 inches need four hinges instead of three. It calculates weatherstripping based on the total distance around the frame. And it flags if any piece is too long to get from standard suppliers.

All of that happens automatically. Every time. No mental math, no reference sheets, no guesswork.

Now, here's where AI comes in — and this is important to understand. AI does not guess the part sizes. You absolutely do not want software "estimating" that a frame piece should be 46.5 inches when it needs to be 46.25. Wrong means scrap metal.

What AI did was build the rules in the first place. This client had about 800 past orders in spreadsheets. Each one was a parts list a human had calculated by hand. That's 800 examples of "when the window is this size, the parts are these sizes."

I fed all those historical records through AI and asked it to find the patterns. The results were specific: "When width increases by 1 inch, the horizontal frame piece increases by 1 inch minus 0.375 inches." "When height exceeds 60 inches, add a fourth hinge." AI pulled out 140+ rules like this from the data. Doing this manually — sitting with the fabricator, testing each relationship, documenting every offset — would have taken weeks. AI did the pattern-finding in hours.

Then the AI found something even more valuable. When it checked the 800 past orders against its extracted rules, it flagged 23 cases where the human fabricator had broken the normal pattern. These weren't mistakes. They were real exceptions — reinforcement bars added for extra-wide spans, different hardware for triple-wide setups, thicker seals for coastal buildings exposed to salt air. The fabricator had this knowledge in his head and had never written it down. AI found it. I confirmed each one with the fabricator. We added those exceptions to the system.

The Money Part

Generating the right parts list is only half the problem. Costing it accurately is the other half.

Before this system, the client was using material prices that were 3 to 6 months old. Updating prices was a full-day chore that kept getting pushed off. They were unknowingly losing margin on more than 30% of their orders — sometimes quoting below their actual material cost on larger jobs where the price drift added up across dozens of parts.

The system I built handles pricing at three levels. Company-wide cost changes (aluminum up 8% this quarter — one update adjusts every quote). Supplier-specific pricing (Supplier A charges $3.20 per foot, Supplier B charges $2.85 but takes three weeks instead of five days). And individual part pricing for special items or volume discounts.

Real example from month one: a particular window configuration cost $847 from Supplier A but $791 from Supplier B, with 12 extra days of wait time. The system shows both. The sales rep makes an informed decision with the customer instead of quoting from memory.

After putting this system into action, quoting accuracy improved from roughly 85% to about 97% against actual material costs. That's the difference between knowing your margins and hoping.

One more piece that matters: the system prints labels for the shop floor. Each label has the exact cut dimensions and a scannable code that pulls up the full order on any phone. This replaced handwritten tags that got covered in sawdust and became unreadable by mid-afternoon. Cutting errors — someone misreading a handwritten measurement — dropped from about 1 in 15 to 1 in 60.

The Results

Before: 45-90 minutes per parts list. 15-20 lists per week. Stale pricing. Fabrication errors costing ~$200 each. One person spending roughly 20 hours a week on this work.

After: Parts list generated in under 30 seconds. Accurate real-time costing. Automated shop floor labels. 20 hours per week reclaimed. Rework costs cut by about 75%.

The build took about 3 weeks. The annual value the client estimated: $50K-$80K, between labor savings, fewer mistakes, and margins they'd been losing without realizing it.

I want to be honest about what the system doesn't do yet. It doesn't automatically place orders with suppliers. It doesn't plug directly into their accounting software (they export files for now). And adding an entirely new product type still requires manual setup. It's not magic. It's a well-built tool that does the actual work.

This kind of system makes sense if your products have variable dimensions, you're building 10+ unique quotes per week, someone is manually calculating part sizes, and you suspect you're losing margin you can't see.

Thinking About AI for Your Business?

If this resonated, let's have a conversation. I do free 30-minute discovery calls where we look at your operations and identify where AI could actually move the needle.

Book a Discovery Call