A 70-person contract machine shop in the Mon Valley was losing work it should have won. Their estimators were buried under 40–60 RFQs a week — PDFs, emails with photos, the occasional STEP file dropped into a shared inbox. By the time a quote went out, three days had passed and the buyer had already moved on. The owner didn't need another ERP. He needed his existing people to stop retyping data between systems.
This case study walks through how AI automation for manufacturers in Pittsburgh actually gets deployed in a shop like that — what was built, what it cost in effort, and what changed in the first 90 days. Names and identifying numbers are anonymized; the workflow pattern is real and repeatable for similar shops within 75 miles of 15220.
The challenge: a 70-person job shop drowning in quote requests
The shop ran a mix of CNC machining and light fabrication for energy, medical device, and a handful of DoD-adjacent primes. Three problems stacked on top of each other:
- Inbound RFQs arrived in five formats. Estimators manually opened each, transcribed part numbers, and looked up prior jobs in an aging ERP.
- Two of their primes had started asking CMMC Level 2 readiness questions, which meant any AI tooling touching drawings had to respect controlled unclassified information (CUI) boundaries.
- Tribal knowledge — "we quoted something like this for [customer] in 2022" — lived in one senior estimator's head and a folder of spreadsheets.
Leadership had looked at off-the-shelf "AI quoting" SaaS tools. Most either wanted drawings uploaded to a public cloud the shop couldn't justify under CMMC, or priced per-seat in a way that didn't fit a 4-estimator team.
The bottleneck wasn't a missing tool — it was six manual handoffs between an inbox, a file share, and an ERP that nobody had time to fix.
How it was solved: AI automation for manufacturers, Pittsburgh-built
The engagement started with a two-week discovery: shadowing estimators, mapping the RFQ-to-quote path, and classifying which data could and couldn't leave the shop's tenant. From there, we built the workflow in stages.
TL;DR: AI automation for manufacturers in Pittsburgh works best when the model sits behind the firewall of the customer's existing Microsoft 365 tenant, not in a third-party SaaS that re-hosts drawings.
The stack:
- Intake normalization. A Power Automate flow watched the RFQ inbox, extracted attachments, and handed them to an Azure OpenAI deployment inside the shop's own tenant. The model parsed PDFs and emails into a structured record — customer, part number, quantity, due date, material callouts, tolerances flagged for human review.
- Historical match. A retrieval layer indexed ten years of past quotes and travelers from the file share. When a new RFQ came in, the system surfaced the three closest historical jobs with prices, cycle times, and the estimator who ran them.
- CUI-safe boundary. Drawings flagged as ITAR/CUI-adjacent were routed to a separate workflow that kept content inside GCC-aligned services and never reached general-purpose models. This was non-negotiable for the DoD-adjacent primes.
- Estimator cockpit. Instead of a new app, estimators got a Teams-based view that showed the parsed RFQ, the historical matches, and a draft response — which they edited and sent. Humans stayed in the loop on every quote.
- ERP write-back. Once a quote went out, the workflow wrote the job record back to the ERP automatically, killing the double-entry step that had been costing 15–20 minutes per quote.
Outcomes: faster quotes, cleaner data, happier estimators
After roughly 90 days in production:
- Average time from RFQ receipt to quote sent dropped from about three business days to under one for standard parts. Complex jobs still got senior review, as designed.
- The estimating team handled a meaningfully higher RFQ volume without adding headcount, freeing the senior estimator to focus on the complex DoD-adjacent work.
- Quote data finally lived in one place. The owner could, for the first time, run a clean win/loss report by customer and material.
- CMMC posture improved as a side effect — the project forced documentation of data flows the assessor was going to ask about anyway.
We don't claim a magic percentage here. The honest answer is that the shop got back the equivalent of a part-time estimator and stopped losing fast-turn jobs to faster competitors.
Who this applies to
This pattern fits small and mid-market manufacturers across the Pittsburgh metro — Monongahela, McKeesport, New Kensington, Washington, Beaver, Butler, and the shops along the Route 51 and Route 28 corridors. If you have between 25 and 300 employees, an ERP you can't easily replace, and at least one customer asking compliance questions, the case study above is roughly your situation.
What's included in a PGH Networks engagement
A typical AI automation engagement covers discovery and process mapping, a tenant and data-boundary review, building the workflows in Azure and Microsoft 365, ERP and file-share integration, estimator and shop-floor training, and ongoing managed support so the automations don't quietly break six months later. Compliance work — CMMC, ITAR data handling, NIST 800-171 alignment — is built in rather than bolted on, because we run the underlying managed IT and cybersecurity for most of our manufacturing clients already.
Why PGH Networks
We are a Pittsburgh-based MSP with a dedicated AI-workflow practice. That combination matters: most shops we talk to have been pitched AI by vendors who don't run their network, and managed IT by providers who don't build automations. Sitting on both sides means we can wire an AI workflow into your ERP, your identity layer, and your compliance documentation without three vendors pointing at each other when something breaks.
AI automation for manufacturers in Pittsburgh isn't an experiment you bolt onto IT — it has to live inside the same identity, security, and compliance fabric your shop already runs on.
We work on-site across the 75-mile radius around 15220, which covers Allegheny, Westmoreland, Washington, Beaver, Butler, and Armstrong counties, plus the northern West Virginia and eastern Ohio shops that sit inside that ring.
Next step
If the case study above sounds uncomfortably familiar, the right next step is a 30-minute discovery call. We'll ask about your RFQ volume, your ERP, and which of your customers are pushing compliance requirements, and tell you honestly whether an AI automation project makes sense this year or whether you have a more pressing IT problem to clear first. Contact PGH Networks at pghnetworks.com to schedule.