Request Routing for a Precision Fabrication Shop

The client

A precision machining and fabrication company whose customers range from one-off prototype shops to major aerospace and defense contractors. Dozens of requests arrive every day: quotes, engineering reviews, material questions, repeat orders, and custom specs. Every one of them landed in the same queue, where a senior engineer reviewed it, estimated its complexity, and routed it forward, with the biggest accounts getting priority from the most experienced people.

What everyone else tried

The AI pitch the company had already heard was to automate the review: train a model on past quotes, pre-fill estimates, flag anomalies, and surface similar jobs. It helped a little. Reviews got faster, but the queue still backed up behind the same engineers, and win rates didn't move.

What we found

When we looked at the data, the priority system was backwards. The largest clients were also the most organized. Their specs were tight and their orders predictable, so they rarely needed engineering judgment, just fast turnaround and accurate pricing. The smaller and mid-tier clients sent the messy, ambiguous requests where an engineer's eye catches a tolerance that doesn't make sense or a material that's overspecified. And buried in the big accounts' repeat orders were patterns nobody was looking for, like a customer ordering the same mounting plates every quarter while buying the matching dowel pins from another vendor.

What we built

We replaced the single queue with an AI triage and routing system that classifies each request by what it needs rather than who sent it:

• Direct to manufacturing. Clean repeat orders with standard specs and known materials are confirmed, scheduled, and acknowledged automatically. Most big-client orders end up here, and they get faster service because of it.
• Engineering review. Genuinely complex requests, unusual tolerances, and first-time specs go to the engineer best suited to that type of work, not whoever is assigned to the account.
• Sales and pre-bid. Requests that hint at an upsell, a larger scope, or a customer solving a problem piecemeal are flagged for sales with the context to act on them.

Every routed request comes with a brief drawn from purchase history, past communications, and the shop's adjacent capabilities, so whoever picks it up starts with context instead of starting cold.

Why it works

Skilled attention now goes where it creates value: to the mid-tier request that hides a much larger relationship, and to the large account whose tidy orders were masking the business it gave a competitor. The shop didn't need faster engineers. It needed to stop treating every request as the same kind of work.