Guide  ·  What qualifies

New to your firm is enough.

The R&D credit rewards work that solves a technical problem your company had not solved before. Most of it happens on ordinary projects. Here is the test the law applies, in plain English.

Four questionsmake up the whole test, set by Section 41 of the Internal Revenue Code. Pass all four, stay clear of the listed exclusions, and the work qualifies.
No lab requireddrafting tables, shop floors, clinics, and job sites can all count.
Three years backcredits you never claimed can still be collected by amended return.

The four-part test

What the law asks
01

A new or improved business component

A product, process, technique, formula, invention, or piece of software. It has to be new or better in function, performance, reliability, or quality. One of those is enough.

02

Technical uncertainty at the start

You did not know whether it could be done, how to do it, or what the final design should be. Permits and budgets are administrative questions. Load paths and tolerances are technical ones.

03

A process of experimentation

You worked through alternatives to get there. Modeling, calculations, prototypes, bench tests, and trial installs count when they test options against a real unknown.

04

Grounded in hard science

The work relies on engineering, physical science, biology, or computer science. Marketing studies, market research, and management changes stay out.

Every project is held to all four. A Foundry R&D credit study documents how each one was met. Read the test in the statute's words.

The standard is new to you

Where owners get it wrong
01

It does not have to be new to the world

Congress wrote the credit to push companies to do work they had not done before. If a competitor solved the same problem last year, your work still counts. No patent is needed.

02

Your product can be the component

For an architect the business component is the drawing set. For an integrator it is the engineered system. For a practice it is the protocol that folds a new device into patient care.

03

Unique projects carry the most

A design you have never solved before is full of technical uncertainty. A repeat of a settled design has little. The study sorts your project list on that line.

04

Failure still qualifies

The test is about the attempt to resolve uncertainty. A prototype that did not work, or a design that was abandoned, is evidence of experimentation.

Where it shows up

Work you already do
ArchitectsStructural, site, and systems problems solved in the drawing set. Read the brief.
AV integratorsControl systems engineered into real buildings, programming included. Read the brief.
Boat buildersHull, build, and model iteration on every new design. Read the brief.
Medical practicesNew scanners, devices, and treatments worked into clinical protocols. Read the brief.
Software teamsCustom applications, and internal tools that clear an extra bar for innovation and risk.
Job shops and fabricatorsCustom parts, fixtures, and processes developed to hit a spec.

Not on the list? Browse the industry briefs, or book a call and describe the work.

What stays out

The lines the study draws
01

Routine production

Work after the uncertainty is resolved. Building the tenth unit of a settled design is production, and the study leaves it out.

02

Cosmetic and market changes

Style, taste, and seasonal changes. Surveys, advertising, and management studies. None of it rests on hard science.

03

Work done outside the United States

Only activity performed in the United States and its territories counts. Where the business is registered does not matter. Where the work happens does.

04

Research someone else paid for

If the customer pays whether or not the work succeeds, the work is funded and you can't claim it. If you get paid only when it works and keep rights to what you learned, the credit is yours. The contract decides it, not whether the fee is fixed.

Common questions

Does the work have to be patentable?

No. A patent isn't required, though an issued patent is strong proof of the work. The standard is new to your company, and most qualifying work would never be patented.

We do the same kind of project every year. Does it still count?

Each project with its own technical problem counts on its own. Two custom homes can be two sets of uncertainty when each site posed its own technical problem. Repeats of a settled design carry little.

Do we need a lab or a research department?

No. The people directly doing the work count, and so do the people directly supporting and supervising it. A principal reviewing redlines is part of the experiment.

Does software we built for ourselves count?

Software you build to sell or license is held to the standard four-part test. Software built only for internal use has an added bar, and the study checks it before anything is claimed.

Do we need documentation before we start?

The law sets no required format, but the records have to back up each project, and the IRS now asks for project-by-project detail on Form 6765. Drawings, models, change orders, test results, and project files are the evidence. The study organizes what you already keep.

What if the project was never finished?

It still qualifies. The credit rewards the attempt to resolve technical uncertainty, whether or not the result shipped.

How far back can we claim?

Usually the past three tax years by amended return, counted from when each return was filed. Some state programs reach further. Credits you never claimed are still there to collect.

How do we know if our work passes?

Thirty minutes on the phone with the person who knows the work. We walk the four questions against your projects and tell you plainly whether a study is worth doing.

Check your inbox. Click the link in our email to confirm.

That didn't go through. Check the email address and try again.

Thirty minutes shows whether you qualify.

Book a call and bring the person who knows the work. We will cover whether your projects pass the test and what a credit could be worth.

Book a call Or write to inquiries@thefoundryfirm.com