Section 174 & 174A · Brief · Working level
R&D in financial services: Section 174 and the credit for banks, insurers, and asset managers
Financial institutions write enormous amounts of software — trading systems, underwriting models, risk engines — and that development is SRE under Section 174/174A even though the industry sells no technology. The Section 41 credit is harder: most of this software is internal-use, so the three-part high-threshold-of-innovation test stands between the same costs and a credit.
Nobody at a bank calls it R&D, which is why financial services chronically misses — and mismeasures — both halves of the research-cost analysis. Section 174 is indifferent to industry: developing or improving software is a specified research or experimental activity whether the developer is a chip company or a regional insurer, and the TCJA-era statute said so expressly. A bank's core-platform rebuild, an insurer's policy administration modernization, an asset manager's portfolio analytics engine — all of it was capitalizable SRE during 2022–2024 and is domestic-deductible now under Section 174A, with offshore development centers still generating 15-year foreign amortization. The credit is a different, narrower question, because so much financial-sector software is internal-use.
The SRE scope: wider than the industry assumes
The costs hide under other names — "change the bank" budgets, digital transformation, IT capital projects. What counts is the activity underneath: planning, designing, building, and testing software improvements, per the framework in identifying SRE expenditures. That sweeps in payments infrastructure, fraud detection systems, regulatory reporting engines, actuarial platforms, and the middleware stitching acquired banks' systems together. What stays out follows the ordinary lines: configuring vendor platforms as shipped, content and UX copy, data migration executed with existing tools, and maintenance that fixes rather than improves — the same configuration-versus-customization boundary that applies everywhere.
Financial institutions also felt the situs rule harder than most. Offshore captive development centers — a fixture of the sector — put large cost pools on the 15-year foreign schedule, where they remain even after the OBBBA. Institutions still unwinding 2022–2024 domestic balances take the one- or two-year catch-up like everyone else.
The credit and the internal-use software wall
Here the industry's position inverts. Software developed primarily for internal general-and-administrative functions is "internal-use software" (IUS), and Treas. Reg. §1.41-4(c)(6) (eCFR Title 26) makes IUS earn the research credit the hard way — the ordinary four-part test plus the high threshold of innovation:
The three additional hurdles for internal-use software:
| Test | What it demands |
|---|---|
| Innovative | Intended to yield a substantial and economically significant reduction in cost or improvement in speed |
| Significant economic risk | Substantial resources committed with substantial uncertainty of recovery, for technical reasons |
| Not commercially available | No off-the-shelf product could do the job without significant modification |
The regulations' own carve-outs matter as much as the tests. Software developed to be used by third parties — customer-facing mobile banking, client portals, trading interfaces used by counterparties — is generally not IUS and faces only the normal four-part test. Nor is software developed as part of a hardware-software product or sold commercially. A sensible credit study therefore sorts the portfolio first: customer-facing and dual-function systems on the ordinary track, back-office G&A systems against the high threshold, with the dual-function safe harbor available for mixed systems. Worked examples are in internal-use software and the credit.
Model development
Quant and actuarial modeling is the sector's most genuine research — and its most contested. Building a new credit-risk model, a novel pricing engine, or a machine-learning fraud detector involves algorithm design, architecture experimentation, and performance engineering that can satisfy the process-of-experimentation requirement. But the credit's technological-in-nature prong wants hard science and computer science, not economics: the defensible claims center on the engineering — making the model compute at scale, integrating it into production systems — rather than the financial theory. And routine life is not research: quarterly recalibration, parameter refreshes, regulatory-mandated validation runs, and production scoring resolve no uncertainty. For 174 purposes the same sort applies with a lower bar; new-model development is SRE, model operation is not.
Frequently asked questions
- Do banks and insurers have Section 174 research costs?
- Yes, usually large ones. Section 174 turns on the activity, not the industry: developing or improving software — core banking platforms, policy administration systems, pricing and risk models, trading infrastructure — is SRE activity. During 2022–2024 those costs were capitalized over five years (fifteen for offshore development centers); under Section 174A, domestic development is deductible again while foreign work stays on the 15-year schedule.
- Can financial institutions claim the R&D credit on internal software?
- Sometimes. Software developed primarily for internal general-and-administrative use must pass the internal-use software rules of Treas. Reg. §1.41-4(c)(6): beyond the ordinary four-part test, it must be innovative, involve significant economic risk, and not be commercially available. Customer-facing platforms — mobile banking, client portals — are generally not internal-use software and face only the normal four-part test.
- Is building or calibrating a financial model 'research' for tax purposes?
- Developing a new model or materially improving one — new methodology, new architecture, engineering to make it perform at scale — can be. Routine periodic recalibration, refreshing parameters on an existing model, and running models in production are not; they resolve no technological uncertainty. The line matters for both the 174 cost population and the much narrower credit population.