The Dangers of R&D Pop-Ups

IRS, itself has stated that “taxpayers should not rely on AI-generated responses to complex tax questions.” Complex tax incentives like the R&D Credit require taxpayers demonstrate that every dollar claimed is directly related to a qualified project. There are only two options to do so: evaluate every project individually or use an IRS-approved sampling methodology. Unscrupulous providers of R&D Credits would have you believe claiming the credit is as simple as checking some questions off a checklist, exporting a few spreadsheets, and then letting their AI fill in the blanks.

Oftentimes, a business owner can only provide a list of their projects. That project list almost never has any corresponding data that will describe the work done, the employees that worked on it, or the time spent by each of those employees on the project. AI models are only as good as the data they are trained on – how can a provider generate a better conclusion with just a small fraction of the information necessary? Untrained models used by inexperienced providers will not draw the correct conclusions for how projects qualify and how much in R&D expenses should be allocated to them. Since the onus for substantiation ultimately falls on taxpayers, they should be very cautious not to rely on an AI that has not been trained with sufficient data.

In addition, as the name would suggest, the R&D Credit is an incentive for increasing investment in R&D activities. A given year’s R&D expenditures must be compared to some set of prior years and must show an escalation in qualified activity to generate a credit. Thus, an R&D Credit study for even one tax year is actually a study for several, consequently multiplying the effects of a faulty methodology.

Only a provider with extensive experience in dealing with the IRS on audit defense issues, as well as one with industry experts and former members of the government, can successfully implement and defend a complex credit study. There is no substitute for their judgement, honed by years of experience and knowledge of how claims will be treated by the IRS (which is run by humans…not by AI).

By moving away from a consultant-based service model, today’s R&D pop-ups also promise accelerated completion times and a lighter load for their clients. This is prevalent not only with new R&D providers, but also payroll processors who already have access to employee wages and want to quickly tack on a tax credit. Established providers looked on in disbelief as Employee Retention Credit (“ERC”) pop-ups utilized this approach, to the point that the IRS instituted a processing moratorium to stem the tide of dubious claims.

The time saved by skimping on diligence on the front end, however, is offset by the additional review and confirmations that business owners must provide on the back end. Even the pop-ups admit that their AI generates predictions, so their clients still have to go through every conclusion and match it with reality. In this way, AI-generated studies are actually built on the backs of business owners. This is not a recipe for good client service.

Experienced R&D Credit providers serve clients well by minimizing their load, taking on the difficult and time-intensive qualification process themselves. More importantly, experienced providers take the burden of making legal determinations off of their clients’ shoulders. This means getting the information from the source: reviewing available documentation and interviewing employees where needed. They also draw from their expertise to parse out which activities qualify and how much in costs to assign to them.

The savvy business owner must decide if they want to provide project information once or if they want to waste time confirming conclusions generated by a poorly-trained AI.

The most reliable and highest-quality providers all have several aspects in common. The following are questions a decision maker should ask before choosing a provider:

• How experienced are they?
  Pop-ups with limited experience do not have enough data to train their AI. Their predictions will be all over the place and require hours of client time to clean up. If the IRS isn’t yet able to successfully utilize AI on its own dataset (see the Government Accountability Office’s May 2024 report TAX GAP: IRS Should Take Steps to Ensure Continued Improvement in Estimates), a twenty-person shop opened a couple of years ago has no business using theirs to run an R&D study.

• Where did their “experts” come from?
  A diverse team of experts will not only be well-versed in accounting lingo – they will also have worked in the industries that they serve and be able to speak their language. AI-driven providers do not provide this depth of knowledge and rely on models trained from limited datasets. The end result is continued reliance on their clients for legal determinations.

• How far do they go to defend their work?
  Inexperienced pop-ups defending credit studies create a data training issue: a successful AI is trained on historical facts and resulting outcomes. Faulty training results in incorrect conclusions. Moreover, resource-strapped pop-ups who cap the audit defense hours included in their engagements (if any are included at all) will very quickly exceed these hours in all but the most basic situations, resulting in more invoicing. Audit defense should be provided until the job is done, not until an hourly quota is fulfilled.

The Section 41 credit is a critical way to mitigate the rising taxes on businesses conducting R&D, and every eligible business should seriously consider it. However, business owners must understand that the decision to employ an established consultant over an unproven pop-up will result in more reliable results, better client service, and ultimately higher quality. The choice is yours: do you want to see your refund on a check, or your name on a disallowance letter?