Platforms, APIs, software architecture and technical experimentation
Estimate your credit →Looking for model development, machine learning or robotics? That guidance now lives in AI + Robotics. This guide focuses on general software and SaaS development.
Building, improving, and scaling software involves continuous technical uncertainty resolved through iterative development and testing — the core of qualified research for modern software companies.
The credit is calculated on qualified research expenses (QREs) — primarily wages for engineers and technical staff who spend time on qualifying activities, plus supplies, cloud/compute, and 65% of qualified contract research.
Qualifying work is more specific than writing code. A team might compare database designs to meet an unfamiliar throughput requirement, prototype a fault-tolerant API, or test model architectures against an uncertain accuracy target. The useful question is what technical capability, method, or design was uncertain at the outset, and which alternatives the team evaluated to resolve it.
Work on a new component and substantial improvements to an existing one can both be relevant. Keep a record of the business component, experiments, results, and people involved rather than assuming every engineering hour qualifies. The table below shows possible components and the technical uncertainty that might support a claim; the facts of each project still matter.
| Business component | Example | Technical uncertainty |
|---|---|---|
| New platform development | Multi-tenant SaaS, marketplace, or vertical platform | Which architecture scales to the target load and reliability? |
| Significant product enhancement | Reporting-engine redesign, new billing or onboarding module | Can the new design hold performance at production scale? |
| Scalability & performance | Monolith-to-microservices, query and latency optimization | How to cut latency while preserving reliability? |
| Complex integrations & APIs | ERP, payroll, banking, or EHR integrations | How to reconcile disparate data models and handle failures? |
| Cybersecurity & data protection | Encryption, identity, role-based access control | Which security architecture withstands the threat model? |
| Real-time data & pipelines | Event streaming, ETL and analytics at volume | How to process at low latency without data loss? |
| Internal workflow automation | Internal workflow engine, back-office process software | Which automation logic performs reliably? (internal-use bar applies) |
Routine maintenance, cosmetic interface changes, ordinary bug fixes, and installing or configuring an existing product generally do not become qualified research just because engineers performed the work. Copying a known implementation into a new codebase is different from evaluating uncertain technical alternatives. Market research, sales experiments, and changes driven solely by a desired business result also need to be separated from technical experimentation.
A project can contain both qualifying and nonqualifying tasks. For example, the experiments that resolve an uncertain scaling problem may be relevant, while subsequent routine deployment or customer training may not be. Time records and project notes help distinguish the two. Work performed outside the United States generally does not count for the federal credit, even when it is directed by a U.S. team.
Software developed primarily for a company's own general and administrative functions can face additional requirements. A back-office workflow, finance tool, or internal operations dashboard is not automatically eligible merely because it was custom-built. Internal-use projects generally must satisfy a heightened innovation test in addition to the usual research requirements; the intended use and project facts determine how the rules apply.
Document what the team could not achieve with available approaches, which alternatives it tested, and the technical risk involved. Do not treat an internal tool as customer-facing software based only on its architecture. If a product includes both customer-facing and internal functions, review the components separately with a qualified tax advisor. See the four-part test for the underlying research criteria.
A startup does not necessarily need income tax liability to benefit from the federal R&D credit. If it meets the qualified small business rules and makes the required election, it may apply the credit against employer payroll taxes. The existing overview describes an offset of up to $500,000 a year; that is a statutory ceiling, not a promised credit for any particular company.
Eligibility depends on the business's gross receipts history and other facts, and the amount still depends on documented qualifying research expenses. Engineers working on a new SaaS platform or experimental model may have qualifying time, but their entire payroll is not automatically included. MainStreet prepares the R&D study and Form 6765; your CPA files the return and helps determine whether the payroll election fits your situation.
Find your own job titles below. Each row shows the wage category a role usually falls under and the kinds of work that typically do and do not count. Titles alone never make wages eligible; the activity must meet the four-part test.
13 of 13 roles shown
| Role | What they do | Wage category | Typically qualifies | Typically does not |
|---|---|---|---|---|
| Builds the services, APIs and data layers behind the product. | Direct research | New service architecture, API design and versioning +2 more | Routine bug fixes, Dependency bumps +1 more | |
Typically qualifies
Typically does not
|
||||
| Builds the interface customers click, type and scroll through. | Direct research | New interface architecture, Performance and rendering work +2 more | Copy changes, Marketing page updates +1 more | |
Typically qualifies
Typically does not
|
||||
| Ships features across the interface, services and database. | Direct research | End to end feature development where the approach was uncertain | Ticket triage, Routine maintenance | |
Typically qualifies
Typically does not
|
||||
| Builds and tunes the native apps on phones and tablets. | Direct research | Offline sync, Cross-platform parity +2 more | App store submissions, Asset swaps | |
Typically qualifies
Typically does not
|
||||
| Keeps builds, deploys and production infrastructure running and scaling. | Direct research | CI/CD build-out, Autoscaling and reliability engineering +1 more | Routine deploys, On-call ticket response +1 more | |
Typically qualifies
Typically does not
|
||||
| Moves, models and stores the data the business runs on. | Direct research | Pipeline architecture, Real-time streaming development +1 more | Scheduled report building, Dashboard upkeep | |
Typically qualifies
Typically does not
|
||||
| Designs how the product keeps accounts and data protected. | Direct research | Pen-test remediation redesign, Authentication and authorization architecture +1 more | Compliance questionnaires, Routine patching +1 more | |
Typically qualifies
Typically does not
|
||||
| Builds the tests that tell the team whether new work actually holds up. | Direct support | Building test harnesses and automation for new development, Designing experiments to resolve uncertainty | Manual regression passes on shipped features | |
Typically qualifies
Typically does not
|
||||
| Leads a team of engineers day to day and reviews how they build. | Direct supervision | First-line technical supervision of qualifying work, Design review +1 more | Hiring, Performance reviews +3 more | |
Typically qualifies
Typically does not
|
||||
| Owns the technical direction of the company. | Direct supervision | First-line supervision where they directly direct the technical work | Board meetings, Fundraising +2 more | |
Typically qualifies
Typically does not
|
||||
| Turns product goals into technical requirements the team builds against. | Direct support | Defining technical requirements and acceptance criteria for qualifying work, Participating in design alternatives | Roadmap presentations, Customer calls +1 more | |
Typically qualifies
Typically does not
|
||||
| Keeps the databases fast, consistent and available. | Direct research | Sharding, replication and performance architecture | Routine backups, Access provisioning | |
Typically qualifies
Typically does not
|
||||
| Connects the product to each customer's systems. | Direct support | Custom technical integration development where the approach was uncertain | Standard implementations, Demos +1 more | |
Typically qualifies
Typically does not
|
||||
Wage categories follow IRC Section 41. Only the portion of time spent on qualified research counts, and the percentage must be documented.
Illustrative example using sample figures. Your actual credit depends on your facts; see Form 6765 and consult a tax professional.
The IRS requires qualifying research to satisfy four tests. Here's how they typically map for this industry:
The activity aims at improving a product, process, software, or technique used in the business — not just a business outcome.
The work relies on principles of engineering, computer science, biology, chemistry, or another hard science to resolve uncertainty.
There was genuine uncertainty at the outset about whether — or how — the component could be built or improved.
The team evaluated alternatives: A/B tests, prototypes, benchmarks, simulations, or iterative design-build-test cycles.
Take these into a 15-minute conversation with your team.
The last question matters most. Customer-funded work and work performed outside the US are generally excluded. Funded work is excluded to the extent your business does not bear the financial risk or does not retain substantial rights in the results.
Software built mainly for your own back-office functions, such as finance or HR, faces a higher bar than software you sell or offer to customers, including a heightened innovation test. If one system serves both purposes, the customer-facing portion can often be reviewed separately from the internal portion. Read the IRS rules on internal-use software
Potentially, yes. The uncertainty only has to be new to your business, not new to the industry or the world. You still need to show the work met the four-part test, including a process of experimentation.
It can. Amounts paid to a cloud provider for the right to use computers in qualified research may count as computer rental, subject to the requirements in Section 41. Hosting for production use by customers generally does not count, so separate development and test environments from live workloads.
Generally at 65 percent of the amount paid, if the work was performed in the US, your business bore the economic risk of the outcome, and it retained substantial rights to the results. The contract terms matter, so review them before including the cost.
Possibly. A qualified small business, generally one with gross receipts under $5 million and no gross receipts more than five years back, can elect to apply the credit against its employer payroll taxes instead of income tax. The amount still depends on your documented qualified research expenses.
Records made during the work hold up best: tickets, sprint boards, commit history, design docs and test results. Pair them with support for how each person's time was allocated to qualifying projects.
Most companies underestimate what they can claim. Our study covers every qualifying role, every component, and every year still open for amendment.
Get your free estimate →