Your quality manager is writing the same Certificate of Conformance for the fifth time this week, pulling material certs from a shared drive, copying customer PO numbers by hand, and reformatting everything to match what the customer's incoming inspection team wants. It's 4:30 PM on a Thursday. The job traveler is done, the parts shipped two hours ago, and she's still at her desk finishing paperwork.

That's not a people problem. That's a systems problem — and it's costing you more than you think. At a fully burdened rate of $45/hour, 11 hours of manual quality documentation per week runs over $25,000 a year in labor that produces zero throughput. It doesn't move a single part. And if that quality manager ever leaves, the system doesn't just slow down — it stops.

The Documentation Problem in a Job Shop

The quality documentation stack in a typical 40-person CNC or sheet metal shop isn't glamorous, but it's load-bearing. Here's what's usually being produced manually, every week:

• Certificates of Conformance (C of C): Required on most defense, aerospace, and medical shipments. Customer-specific formats, part numbers, rev levels, material spec callouts, inspection results.
• Nonconformance Reports (NCRs): Written when parts fail inspection. Description of the defect, affected quantity, disposition (rework, scrap, use-as-is), and often a linked Corrective Action Request (CAR).
• First Article Inspection Reports (FAIRs): Dimensional and material results mapped to print callouts, often in AS9100 or customer-specified format.
• Standard Operating Procedures (SOPs): How the shop actually runs — setup procedures, inspection methods, handling requirements, operator instructions. Rarely written down, even when everyone agrees they should be.
• Inspection Reports: Job-specific records tied to a work order, serial number, or lot.

None of this is complex writing. It's structured, repetitive, and highly templated — which makes it exactly the kind of work Claude handles well.

What Claude Changes

Claude is an AI assistant made by Anthropic. You interact with it in plain English, and it responds like a knowledgeable colleague who has read everything you've given it. The feature that makes it genuinely useful for a job shop — not just a novelty — is Projects.

Claude's Projects feature gives you persistent memory across conversations. Instead of re-explaining your company's quality system every time you open a new chat, you set it up once inside a Project. You upload your C of C template, your NCR form, your SOP format, a short description of how your shop runs — and Claude holds all of that context. Every conversation inside that Project starts with Claude already knowing your part numbering convention, your customer-specific requirements, your ISO 9001 scope, and your preferred language.

This is the difference between using a general-purpose AI and having a documentation assistant that actually knows your operation.

An SBE Council report from 2026 found that 82% of SMBs have now invested in AI tools. The gap isn't adoption anymore — it's setup. Most shops that try Claude once use it like a search engine, get mediocre results, and move on. The shops getting real value built a Project. That's what this post walks you through.

Step-by-Step Setup

This takes under an hour. You do not need an IT background.

1. Create a Claude account and start a new Project. Go to claude.ai, sign up for the Pro plan ($20/month), and click "New Project" in the left sidebar. Name it something like "Quality Documentation — [Your Shop Name]."

2. Write a "how we work" description in the Project instructions. In the Project settings, there's a field for custom instructions. Write 3–5 sentences describing your shop: what you make, what certifications you hold (ISO 9001, AS9100, ITAR, etc.), what ERP you use (JobBOSS, E2, Epicor), and how your job travelers are structured. Example: "We're a 38-person CNC machine shop in Ohio. We hold ISO 9001:2015 certification. We run JobBOSS for job tracking. Our part numbers follow the format XXXXX-Rev-XX. Most customers require a C of C with every shipment referencing our cert number and the specific material heat number."

3. Upload your existing templates. Attach your C of C template, your NCR form, and one or two completed examples. Claude will learn the format, the fields, and the language you already use.

4. Generate your first C of C. Open a new chat inside the Project. Paste in the relevant job details — part number, revision, quantity, customer PO, material cert number, inspection results — and ask Claude to draft the C of C. Review it. Correct anything that's off. Claude will hold those corrections for the rest of the conversation.

5. Build an SOP through a Claude interview. Instead of writing an SOP from scratch, let Claude interview you. Type: "I need to write an SOP for our deburring process. Ask me questions one at a time until you have enough to write a complete procedure." Claude will ask about equipment, PPE, inspection criteria, reject conditions, and sign-off requirements. When it has enough, it writes the draft. This takes 15–20 minutes instead of two hours.

6. Speed up NCRs with a fill-in prompt. Create a saved prompt (keep it in a text file on your desktop) that reads: "Write an NCR using our standard format. Here are the details: [defect description], [affected quantity], [job number], [disposition]. Include a placeholder for the CAR number." Paste and go. NCR drafts in under two minutes.

What the Results Look Like

A 40-person contract manufacturer running a mix of CNC turning and sheet metal work ran this setup starting in Q1. Before Claude, their quality manager was spending roughly 11 hours per week on documentation — C of Cs, NCRs, inspection reports, and the occasional FAIR. After four weeks with the Project configured and the team trained on the prompts, that number dropped to just under 3 hours.

The audit prep shift was even more noticeable. Their previous AS9100 surveillance audit required two full days of document gathering and formatting. With SOPs already drafted and stored, the next audit prep took four hours — mostly verification and cross-referencing, not creation.

The other result nobody talks about: when their quality manager took two weeks off, a production supervisor was able to generate C of Cs without calling her. The knowledge was in the Project, not just in her head.

What This Doesn't Cover

Claude drafts documents. It does not pull live data.

If your part number, revision level, or inspection results live in JobBOSS or E2, Claude cannot reach in and grab them. Someone still has to copy that information into the prompt. Claude also cannot update your ERP, attach documents to work orders automatically, or trigger a CAR workflow when an NCR is created. Manual data entry is still part of the picture.

Claude also won't catch a bad inspection call. It will write what you tell it. Garbage in, garbage out — same as any template.

This is step 1. It's a real hour of work with a real return: 8 hours of weekly documentation time recovered, institutional knowledge captured in a system instead of a single person's memory, and audit prep that doesn't require a fire drill. When you're ready to connect this to your ERP, your inspection tools, and automate the data flow — that's where architecture matters.

---

Book a discovery call at cloudbeast.io/schedule. We'll look at your current documentation stack, what's manual, and what a connected system could actually look like for your shop.