How Arvo’s Evidence Engine Saves Up to 80% of R&D Credit Prep Time

"Send over a list of your R&D projects."

That is the email. You open a blank document. Then you stop. Your team shipped a lot last year. Which of it counts? Where is it written down? Who has time to write it up?

That list is often the start of an R&D tax credit study. The credit is a federal incentive worth 7 to 10 percent of what you spend on qualified research. The study is the record that backs it up.

This post covers three things: what a study is, where the time goes, and how Arvo’s new Evidence Engine starts the work for you.

What Is an R&D Tax Credit Study?

An R&D tax credit study is a written record of which projects qualify for the federal R&D credit, why they qualify, and what you spent on them. It is the backbone of your claim on IRS Form 6765. For most companies the credit is worth 7 to 10 percent of qualified research spending.

The credit is a dollar-for-dollar offset. It can reduce income tax or AMT. Qualified small businesses can also use it against up to $500,000 a year of employer payroll tax. Unused credit carries forward 20 years.

A project qualifies when it passes the four-part test in IRC §41(d). In plain words:

• Purpose. The work aimed to build or improve a product, process, or piece of software.
• Uncertainty. At the start, you did not know if it would work, or how.
• Experiment. You tried things, tested them, and ruled some out.
• Science. The work relied on engineering, computer science, or another hard science.

That is the whole test. For the long version, read R&D Tax Credit Qualified Activities.

A finished study has three parts. First, the list of projects. Second, a short write-up for each one that shows how it passed the test. Third, the costs tied to each project: wages, supplies, contractors, and cloud. The R&D Tax Credit Qualified Expenses Guide covers each cost type. The totals go on IRS Form 6765, which we walk through in the IRS Form 6765 Ultimate Guide.

Why Do a Study If the IRS Does Not Require One?

The IRS does not require a study. Arvo does one anyway to protect clients if the IRS asks questions later. A study shows that each project passed the four-part test and that every dollar claimed ties back to real work. Think of it as audit protection, not a rule.

The risk is simple. A credit with nothing behind it is the one that gets questioned. If the IRS asks, you need to show which projects qualified, why, and what you spent. The IRS Audit Techniques Guide for the research credit tells examiners what to look for. A good study answers those questions before anyone asks them.

Arvo has been doing this since 2016. In that time, $0 has been disallowed.

Every Arvo study moves through four steps:

• Qualify. Find the projects that pass the four-part test. The Evidence Engine works here.
• Calculate. Tie wages, supplies, contractor costs, and cloud spend to each project.
• Document. Write the study and prepare Form 6765.
• Monetize. Coordinate with your CPA and your payroll provider so the credit lands on your return or your payroll filings.

That last step matters. Arvo prepares the credit and hands it off cleanly. Your CPA files the return. Your payroll provider files Form 941. Nobody has to switch systems.

The Blank Page Problem

Most of a study is prep, not math. Teams chase project details across GitHub, Jira, and old release notes, argue about what counts, and write every project narrative from scratch. Providers commonly quote four to eight weeks for a study. The blank page, not the tax code, is what takes the time.

The time goes to three places.

• Finding the work. Last year's projects live in commits, tickets, specs, Slack threads, and a few people's memories. Someone has to pull it all into one place.
• Debating what counts. Was the payments refactor research, or just maintenance? Every borderline project turns into a meeting.
• Writing from nothing. Each project needs a short narrative that maps to the four-part test. Engineers do not want to write it. Finance cannot write it without them.

Here is what providers say a study takes. These are their own published numbers, viewed on Sept 25, 2026.

None of those weeks go to the tax code. They go to the blank page.

So what if the project list started as a draft instead of a blank page?

How the Evidence Engine Starts Your Study for You

The Evidence Engine is a feature inside Arvo's R&D Portal. Connect GitHub or Jira, or upload release notes, specs, or road maps. It reads titles, descriptions, and dates from the credit year and groups related work into suggested projects. You pick which ones go in. Arvo's experts review each one before it enters your study.

It works in four steps.

• Connect your sources. GitHub, Jira, release notes, product specs, road maps. Use one or all five.
• We find the work. The Evidence Engine reads what you connected and groups related items into suggested projects. Your first draft is ready in hours.
• You choose. Every suggestion starts as "not included." You decide what goes in.
• Experts review. An Arvo tax expert checks each project you picked against the four-part test.

Here is an example. The numbers are made up to show the shape, not a client result.

A dev team closes 340 Jira tickets and pushes 1,200 commits in a credit year. The Evidence Engine reads the titles, descriptions, and dates. It groups the work into 11 suggested projects. The team looks at the list and includes 7. The other 4 were bug fixes and routine upgrades, so they stay out. An Arvo expert reviews the 7 against the four-part test. The study starts with 7 real projects. Nobody stared at a blank page.

This is the mix Arvo has always run: software plus people. Jeff Coombs, Controller at Origin Financial, said his team's experience was "Outstanding every step of the way. We had interactions with both tax experts and the software." 

What the Evidence Engine Reads (and What It Never Reads)

The Evidence Engine reads three things: titles, descriptions, and dates. It never reads or copies source code. It uses that metadata to spot related work and suggest projects you may have forgotten. Every suggestion starts as "not included" until you choose to add it.

Most CTOs ask the same question first: does this thing read my code? No. It reads the words your team already wrote around the code. A commit message. A ticket title. A line in a release note. That is enough to see that six tickets in March and forty commits in April were the same project.

That is also how it finds the work you forgot. Teams remember the big launch. They forget the three weeks spent on a data pipeline that never shipped, or the tagging system that quietly became the core of the business. Both can qualify.

Joe Biel, founder and CEO of Microcosm Publishing, did not think a book publisher did R&D. His team had built its own inventory software and a custom buying algorithm. That work earned a $244,201 credit. "Arvo is a great partner for Microcosm because they know about all these weird tax incentives that we wouldn't have even had on our radar if they hadn't approached us about them." 

Four Things the Evidence Engine Does Not Do

The Evidence Engine speeds up discovery. It does not replace judgment. It does not decide your credit. It does not file your claim. It does not read your source code. And it does not skip expert review. A tax expert checks every project against the four-part test before anything is published to your study.

The Evidence Engine speeds up discovery. It does not replace judgment.

It does not decide your credit.

It does not file your claim.

It does not read your source code.

It does not skip expert review.

That is why we do not call this automation. Software finds the candidates. People make the calls.

Fast and Defensible: Why the Human Review Step Matters

AI-only tools are fast but thin. Legacy firms are careful but slow. Arvo pairs software suggestions with a tax expert who reviews 100 percent of studies before anything is filed. Since 2016, Arvo has had zero credits disallowed. You should not have to choose between fast and defensible.

Fast. The Evidence Engine handles the slowest part of the study: finding the work. Your first draft of suggested projects is ready in hours, not weeks.

Reviewed. A tax expert looks at every project in every study. Not a sample. All of them. Suggestions do not become claims until a person says so.

Defensible. Arvo helps you build each study around what the IRS Audit Techniques Guide says examiners look for. Since 2016, $0 disallowed.

The market splits into two camps. Software-only tools move fast and leave the judgment to you. Big firms bring judgment and take weeks to deliver it. Arvo sits in the gap. 

The review step also catches what a generalist misses. Kevin Gray, CEO of ApproveMe, had his CPA look at the credit first. Then he came to Arvo, which found $59,096. "Their incredible team helped us identify tax refunds that our CPA missed."

Who It Fits and What to Expect

The Evidence Engine fits teams with real development work spread across several tools and little time to write it up. Plan on two to three hours of your team's time for the whole study. Your first draft of suggested projects is ready in hours, not weeks.

You are a good fit if most of these are true:

• Your dev work lives in GitHub, Jira, or shared docs, not in one person's head.
• Your finance team is small, or it is one person wearing four hats.
• You are a qualified small business under §41(h), so the credit can offset payroll tax. See R&D Tax Credit & Payroll Taxes.
• Arvo confirmed your eligibility on a call.

Fees are 20 percent or less, and transparent. The full schedule is at arvotech.com/pricing/rd.

What you get back is not just a tax line. Forrest North, co-founder and CTO of TELO Trucks, described a $652,723 credit in the terms an engineering leader uses: "It's basically like having the payroll for one to two extra engineers, and that's hugely valuable for startups. We'll probably build more value with those one or two people in this time period than the tax credit itself is worth." 

Frequently Asked Questions

How long does an R&D tax credit study take?

Most providers quote four to eight weeks. With Arvo, your first draft of suggested projects is ready in hours. Plan on two to three hours of your team's time for the whole study.

Is an R&D study required by the IRS?

No. The IRS does not require one. Arvo does a study for every client anyway. It protects you if the IRS asks questions later.

What records do I need for an R&D study?

For the project list: GitHub, Jira, release notes, product specs, or road maps. Any one of them works. For the cost side, you will need payroll data and expense records for the Calculate step.

Does Arvo’s Evidence Engine read my source code?

No. It reads titles, descriptions, and dates. It never reads or copies source code.

Who decides which projects qualify?

You choose which suggested projects to include. Then an Arvo tax expert reviews each one against the four-part test. Nothing enters your study without both.

Start With a Draft, Not a Blank Page

The hardest part of the credit is the list. The Evidence Engine starts it for you, and an expert checks every line. Talk to an R&D specialist to see if your work qualifies.