A multinational manufacturer hired one of the largest accounting firms in the world to build its R&D credit study. It was thorough by the traditional standards. However, neither the company nor the accounting firm could provide documentation sufficient to support the credit once it was scrutinized by the IRS.
Kyocera AVX claimed $398,985 in research credit for 2018. A Big Four multiyear study pushed that number to roughly $1.7 million. To get there, the firm interviewed 36 subject-matter experts after the year had already closed and used what those people remembered to represent the work of about 1,200 employees, with no time tracking and expense allocations built on estimates. When the government subpoenaed the underlying source documents and survey responses, nobody could produce those either. The IRS moved for summary judgment and called the study hearsay on many levels. The case is now headed to Tax Court, and the government has countersued for a $13.36 million refund it says was paid in error. Nothing is decided yet, but Kyocera is now fighting on two fronts: the credit it claimed and the money it already has in hand.
I helped prepare R&D credit studies for several clients at a Big Four firms early in my career. Later I examined those same kinds of studies as a senior revenue agent in the IRS’s Large Business and International division. These days I’m working with engineers and data scientists to build tools meant to replace the interview-heavy model. What I took from my years as an examiner is blunt: the IRS doesn’t care whose name is on the cover of a credit study. They care whether the numbers tie back to something real. A polished report built on someone’s memory of what happened is, at the end of the day, just a memory (and often hazy memories, at that)!
An adage I picked up from a TEI conference was “pave the road with documentation, and fill the potholes with interview notes”. In Kyocera, the interviews were the road, and there was no pavement underneath. Interviews have a legitimate role - they can supplement and explain the gaps in the contemporaneous records - but they cannot be the record. The IRS has said as much for more than two decades, going back to Eustace in 2002 and reinforced in its own audit guidance. And the bar keeps rising: the four-part test demands activity-level proof that a real process of experimentation occurred, not a percentage an engineer estimated a year after the fact.
So why do traditional studies still lean so heavily on interviews? Not because anyone is cutting corners. Until recently, it was the only practical way to translate the work. Engineers document to build things, not to satisfy Section 41; their records are technical, non-linear, and written in the language of engineering, and someone has to re-map all of it to the four-part test. Sitting an accountant down with an engineer was the efficient way to do that - until the scale of a large multinational’s data made it impossible to do by hand. No team can manually read tens of thousands of tickets and commits, so they sample, interview, and estimate. It was a reasonable answer to a hard problem. It just isn’t the best one anymore.
But that mountain of data is also where the fix is. If your company runs on Jira, GitHub, or any modern project system, you’re already generating exactly the record the IRS prefers: timestamped entries showing what got built, when, by whom, and how it changed. Build the study on top of that, and interviews can go back to filling potholes instead of paving the road. Your best engineers stop losing a week in conference rooms trying to reconstruct work they finished months ago, and the study underneath the credit is stronger for it.
This isn’t about claiming a bigger number. It’s about claiming the right number – sometimes more than the interview method would find, sometimes less - and being able to stand behind every dollar of it if an examiner ever asks. That’s the standard I held taxpayers to at the IRS, and it’s the standard a credible study should meet today.
We built Neo.Tax’s platform around this, but you don’t need our software to take the lesson from Kyocera. Before you engage a firm to prepare your next R&D study, ask whoever prepares it one thing: if the IRS asked for the documentation behind every number in here tomorrow, could you produce it?
If the honest answer depends on what somebody remembers, you’re one audit away from finding out the hard way.