Life Sciences R&D Tax Credits: Common Pitfalls and Recent Tax Change

ADAM HUNTER: Hey everyone, welcome to Cherry Bekaert's Life Sciences Industry Podcast. We're continuing our conversation on R&D tax credits with Carolyn Smith Driscoll.

ADAM HUNTER: If you haven't already listened to our first episode, I highly recommend going back and checking it out. Carolyn, welcome back. As a director on an R&D tax credit team, you work with innovative companies across the life sciences industry to help them identify and maximize available credits.

ADAM HUNTER: Before we jump into common pitfalls, I do want to spend a few minutes on digital health. We've been seeing a lot more activity in the software as a medical device space, from mobile apps and algorithms to AI-driven diagnostic tools. Carolyn, does that work qualify the same way?

CAROLYN SMITH DRISCOLL: Adam, that's a great question. This is one of the fastest-growing areas of qualifying R&D we see.

Let me define it. The FDA borrows its definition from the International Medical Device Regulators Forum, defining software as a medical device, or as we call it, SaMD, as software intended for medical purposes that performs those functions without being part of a hardware medical device.

The key phrase there is "without being part of." SaMD runs independently. It's not what we see frequently, the firmware controlling a ventilator or embedded code inside of a CT scanner. That's a different category called software in a medical device, or SiMD.

Software as a medical device stands on its own.

ADAM HUNTER: Would you be able to give me some concrete examples?

CAROLYN SMITH DRISCOLL: Absolutely. Think about a smartphone app that lets a clinician view MRI images for diagnostic purposes, or computer-aided detection (CAD) software that helps detect breast cancer from imaging.

We see cloud-based algorithms interpreting ECG data to detect atrial fibrillation, or machine learning models analyzing skin lesion images to assess cancer risk. Those examples are all squarely software as a medical device.

What would not be software as a medical device is what I mentioned previously. Embedded software, often called firmware, like microcode driving motors in an infusion pump, is SiMD, not SaMD.

Also, general clinical workflow software, such as patient registration, billing, scheduling, and video calling, falls entirely outside of the medical device definition. Those company efforts could be eligible software development under the R&D tax credit definition, but they are not part of software as a medical device.

ADAM HUNTER: How does all of this map to the four-part test? Where do you typically find the qualifying R&D expenses?

CAROLYN SMITH DRISCOLL: That's a great question. Let's walk through the four parts.

First, permitted purpose. You're developing a new product or improving an existing algorithm to achieve better sensitivity or specificity.

Next, technical uncertainty. You've likely got a real unresolved question about model architecture, training data sufficiency, generalizability across different patient populations, and latency requirements for real-time clinical use.

Third, process of experimentation. You have model training and validation cycles, hyperparameter tuning, and benchmarking against ground-truth data sets.

Last but not least, technological in nature. The activities rely on computer science, data science, and biostatistics, all of which are explicitly recognized under Section 41.

ADAM HUNTER: Is that correct?

CAROLYN SMITH DRISCOLL: Absolutely.

ADAM HUNTER: What about the regulatory work itself, such as 510(k) preparation and all the documentation efforts?

CAROLYN SMITH DRISCOLL: A lot of the technical work qualifies, but not everything.

Algorithm validation studies, software verification under IEC 62304, cybersecurity testing, and human factors engineering all qualify because you are resolving technical uncertainty about safety and effectiveness.

But the purely administrative side, such as assembling the submission binder, paying user fees, and routine correspondence with the FDA, does not qualify. Qualifying activity is the technical experimentation that resolves uncertainty, not the paperwork that documents it.

ADAM HUNTER: I imagine with the rise of AI and machine learning, the FDA raises some unique issues there.

CAROLYN SMITH DRISCOLL: It does, and it's a conversation we're all having right now.

Adaptive algorithms continuously retrain. Companies assume that once the FDA clears the model, the R&D stops. It absolutely doesn't.

Each meaningful model update, drift detection, adjustment, or performance monitoring cycle typically involves a new round of experimentation that may generate fresh QREs year over year.

ADAM HUNTER: Are there any common traps you'd flag for SaMD companies specifically?

CAROLYN SMITH DRISCOLL: There are a few.

First, wage allocation. Most SaMD teams are cross-functional. They're a blend of software engineers, data scientists, clinicians, and QA functions.

Time tracking is important because under the new Section G of Form 6765, the IRS wants wages broken down by direct performance, direct support, and direct supervision by business component.

Because SaMD teams are so interdisciplinary, it can be a challenge to allocate QREs to each business component and each employee contributing.

Second, cloud and compute costs. GPU compute for model training and cloud infrastructure for validation environments may be treated as supplies or computer lease costs, depending on how they are utilized.

If they are consumed in the research, they could be considered a supply. If you are renting space, similar to cloud hosting, they could be considered computer lease costs. The analysis is very fact-specific, so don't assume they are either all in or all out.

Third, funded research. A lot of software as a medical device work happens under collaboration agreements with hospitals or pharmaceutical partners.

You have to examine who owns the IP and who bears the financial risk to identify what qualifies.

ADAM HUNTER: I think I got all of that. Bottom line, if you are building a SaMD product, while there's a lot of R&D credit opportunity, you've really got to document it at the business component level from day one.

CAROLYN SMITH DRISCOLL: Exactly, Adam.

Given how visible AI in healthcare has become, I'd expect software as a medical device claims to attract real IRS scrutiny going forward. Build the file the right way from the start, and you'll be able to withstand that scrutiny.

ADAM HUNTER: Let's talk costs. What can you actually capture here?

CAROLYN SMITH DRISCOLL: The R&D tax credit is an expense-based credit, and there are three buckets of qualified research expenditures, or QREs.

First, wages. Salaries paid to scientists, engineers, lab technicians, QA or QC staff, and project managers who perform, directly supervise, or directly support the research.

For example, if an engineer earns $120,000 a year and spends 75% of her time on qualifying work, her wage QRE would be $90,000.

The second bucket is supplies, which are tangible property consumed in the research. Examples include reagents, cell culture media, chromatography resins, or prototype components.

The third bucket is contract research. Statutorily, only 65% of what you pay third parties, such as a CRO, CDMO, or academic collaborator, is eligible, and only if you keep substantial rights to the results and bear the financial risk. We always dig into the contract.

You'll hear "rights and risk" discussed quite frequently.

ADAM HUNTER: Thanks for that. I know when I'm talking to a lot of my clients about R&D tax credits, they're often early-stage companies that don't even have revenue yet. They're still going through preclinical or clinical work. What would you say to those companies?

CAROLYN SMITH DRISCOLL: That's a great question, and this is a phenomenal opportunity for newly starting companies.

Even if you're pre-revenue with zero income tax liability, the credit can still put cash in your pocket.

Under Internal Revenue Code Section 41(h), qualified small businesses, generally those with less than $5 million in gross receipts in the current year and no gross receipts before the five-tax-year period ending with the current year, can apply up to $500,000 a year of R&D tax credits directly against employer payroll taxes.

That's $250,000 against the employer portion of Social Security tax and another $250,000 against the employer portion of Medicare tax. It's very lucrative.

For example, a startup with $1.5 million in qualifying research wages could generate a payroll tax offset approaching $100,000 or more annually. It's a real reduction in quarterly cash outflows.

ADAM HUNTER: We talked a little bit earlier about pitfalls around SaMD companies, but where do you see other companies stumble in claiming the R&D tax credit?

CAROLYN SMITH DRISCOLL: There are a few.

The first is underclaiming. A lot of life sciences companies assume only late-stage or commercialized projects qualify, or they think they have to be profitable to benefit. Neither is true.

We routinely find companies leaving approximately 30% of their potentially eligible credits on the table.

For example, one of my clients was a medical device company that was only claiming credits for its dedicated R&D engineering team. They also had a manufacturing process development group spending significant time evaluating new fabrication techniques, testing materials for biocompatibility, and optimizing sterilization for new product configurations.

None of that was being captured because they were focusing only on the R&D engineering team.

Once we pulled in those activities and traced the wages, supplies, and contractor costs to the business component, their annual credit almost doubled.