The R&D Credit for Job Shops: What Qualifies, What Doesn't, and the Records That Help

If you make parts to other people's prints, you have probably heard that the R&D credit is not for you. This guide explains why that answer is often wrong, which jobs tend to qualify and which don't, how customer-funded work is judged, and which records carry a claim.

This is general information, not tax advice. Your CPA decides how your return is filed.

Why job shops get told no

No engineers on staff. No lab. No patents. The customer sends the print, so the design belongs to the customer. All of that is true of many job shops, and none of it is the test.

The IRS uses a four-part test. It asks about the work, not the job titles or who owns the drawing:

1. Permitted purpose. You were trying to make a new or improved part, process or tool: one that works, works better, faster or more reliably.
2. Technical uncertainty at the start. You didn't know how to do it, whether it would work, or what the right setup was.
3. A process of experimentation. You tried options: drawings, models, prototypes, test runs, trial batches, simulations.
4. Technological in nature. The work relied on engineering, physics, chemistry, biology or computer science.

A job shop lives in parts two and three. The customer owns the part design. The way you learned to make it is yours. Say you had to work out how to hold the part, how to cut it and what sequence would hold the tolerance. You carried the cost when the first parts were scrap. That process development may qualify. Machinists, programmers and shop leads often do this work. You don't need engineers on staff.

Work that often qualifies

• Designing and programming toolpaths, fixtures and tooling for new or difficult parts
• First-article work and prototypes where tolerances or materials were uncertain
• Mold design and process development for new plastic parts
• Process changes to cut scrap or cycle time
• Designing custom machines, instruments or automation
• Testing new materials, coatings or joining methods

What these share: at the start, nobody knew whether the approach would work, and you found out by trying. Think of a first article that took several setups to pass inspection.

Work that usually doesn't

• Repeat production runs of proven parts
• Routine inspection and QC of finished parts
• Customer-funded work where payment didn't depend on success and you kept no rights

The same goes for cosmetic changes with no technical question, for market research, sales and general administration, and for work done outside the United States. In Union Carbide (2d Cir. 2012), supplies used in normal production runs generally don't count, even when an experiment rides along.

Customer-funded work, explained

This question often decides a job shop claim. Every job is paid for by a customer, so "customer-funded" cannot simply mean "a customer paid you." It turns on who carried the risk and who kept the rights.

Who carried the risk

If the customer paid for your time whether or not the part came out, the customer carried the risk. If you were paid for good parts that met the drawing, and the scrap, extra setups and reprogramming came out of your margin, you carried it. Meyer, Borgman & Johnson (8th Cir. 2024) is the warning: a structural engineering firm lost its credit because its contracts paid it regardless of success.

Who kept the rights

The customer owns the drawing and the part. The question is whether anything in the terms stops you from using what you worked out (the fixture design, the program, the process) on other work. If nothing does, you kept the rights, and the rights don't have to be exclusive. Populous Holdings (Tax Court, 2019) went the other way: an architecture firm's contracts can leave it bearing the financial risk of its design work, which keeps the research from being funded.

Purchase order terms

A fixed price per piece helps, but on its own it is not enough. Terms that pay you only for conforming parts the customer accepts, and let the customer reject parts that miss the print, point toward you carrying the risk. Billing for hours whatever the outcome, or terms that hand the customer everything you developed, point the other way. We review purchase orders and terms for the customer jobs that go into a study (see how we work).

The records that help

You do not need lab notebooks. The job packet already holds most of the evidence.

What you already keep	What it shows
Drawings, RFQs and quote notes	What you set out to make and what looked hard
Job travelers, setup sheets, program and design revisions	What you tried, in order
Scrap tickets, rework records, photos of failed parts	What didn't work. Failures are strong evidence of real trial and error.
First-article, inspection and test reports	How you settled it, and when development ended
Emails about a difficult job	The open question, in your own words, with a date
Timecards and payroll	Who worked on it
Material and tooling invoices	What the trials consumed
Purchase orders and contract terms	Who carried the risk if it failed

Three things most shops are missing:

1. A dated note of the problem. Five lines when a hard job starts: what are we trying to make work, and what don't we know how to do yet.
2. Time by person by project. One line per person per week, or a project code on the timecard.
3. Trial material kept apart from production material.

Records made while the work was happening carry far more weight than a story rebuilt a year later. Little Sandy Coal (7th Cir. 2023) shows what happens otherwise: a shipbuilder lost its credit because it could not show from its records which of its employees' work was experimentation. The full list and a ten-minute-a-week routine are in the Shop Floor Evidence Guide.

What the credit is worth, in plain terms

For most established claimants using the Alternative Simplified Credit, the federal credit works out to roughly 4% to 9% of qualifying spending. For a job shop, that means the wages of machinists, programmers, engineers and supervisors for time on qualifying work. It also covers materials scrapped or consumed in trials and prototypes, and 65% of US contractor costs for qualifying work.

Didn't claim in past years? Open years can generally be amended, usually 2023 onward; your CPA confirms each deadline (see amending prior years). From 2026 tax years, most claimants also report research costs by project on Form 6765 Section G (see Section G explained). The credit is separate from the Section 174A deduction (see Section 174A after the OBBBA). Your own CPA files; we don't prepare or file returns. Details are on the pricing page.

Next step

Book a 15-minute call and bring a rough headcount and two or three jobs that fought you. If the credit isn't worth pursuing, we'll say so on the call. More on the shops we work with: manufacturing and the FAQ.