How to fund your innovation by recovering ~10% of your development spend
Did you know the code you shipped this year could pay for your next hire? Most AI companies leave massive capital on the table—often hundreds of thousands of dollars in unclaimed credits. This strategic checklist, based on current IRS regulations and expert insights, helps you determine your eligibility and audit-proof your claims.
Whether you're building foundation models, fine-tuning LLMs, or transforming manual services into automated software, understanding R&D tax credits is essential for maximizing your runway and funding growth.
Includes insights from the Runpoint Podcast interview with Ari Salafia
Phase 1: The 4-Part Qualification Test
Before diving into calculations, you must pass the IRS's four-part test. Each component is essential—miss one, and your entire claim could be disqualified. Understanding these criteria helps you structure your development work to maximize legitimate credits.
01
Permitted Purpose
Are you creating new functionality, or improving performance and reliability of an AI system? If yes, you're already aligning with the IRS standard of improving performance, functionality, reliability, or quality.
02
Technological in Nature
Does the work fundamentally rely on hard sciences—computer science, mathematics, or statistics? Your innovation must be grounded in technical principles, not business strategy.
03
Elimination of Uncertainty
Can you prove you faced technical challenges where the solution wasn't obvious at the start? Document the unknowns you encountered and how you resolved them.
04
Process of Experimentation
Did you run simulations, employ trial-and-error, or test multiple alternatives? You must demonstrate systematic evaluation of potential solutions.
Startup Bonus: Immediate Cash Flow
Less than 5 years old and under $5M in revenue? You can apply this credit against Payroll Taxes for immediate cash flow relief. Older, profitable companies apply it against Income Tax. This makes the credit especially powerful for early-stage AI companies burning cash on development.
Phase 2: Eligible Activities & Critical Traps
✅ Eligible AI Activities
Algorithm Development: Creating novel ML/Deep Learning models or NLP pipelines from scratch
Data Engineering: Designing novel pre-processing techniques beyond standard libraries
System Architecture: Optimizing LLMs for edge devices or specific hardware constraints
Model Training: Fine-tuning foundation models (RLHF) on proprietary data
Service-to-SaaS: Transforming manual services (consulting/audit) into automated software products
❌ Red Flags (Ineligible)
Routine Configuration: Drag-and-drop CRM/ERP setup with no custom development
Market Research: Surveys, focus groups, or business analysis activities
Cosmetic Changes: UI/UX updates with no backend innovation or technical complexity
Post-Production: Bug fixing after the model is stable and deployed
⚠️ The "Internal Use" Software Trap
Building internal tools like dashboards, resource planners, or workflow automation? If it's not customer-facing, it must pass a significantly higher bar to qualify for R&D credits.
Three Critical Requirements:
Innovative: The software must result in a substantial and economically significant improvement, such as a reduction in cost or improvement in speed.
Significant Economic Risk: You're committing substantial resources under real technical uncertainty. The question is whether the approach or method will actually work with real resources on the table, and you don't 100% know that upfront.
Not Commercially Available: The software cannot be readily purchased, leased, or licensed from a third party. There's no plug-and-play option that meets your needs without significant buildout.
Many AI companies mistakenly claim internal tools that don't meet these elevated standards, triggering audits and disallowances.
The distinction between eligible and ineligible activities isn't always black and white. When in doubt, focus on documenting the technical uncertainty you faced and the experimentation process you employed. Failed experiments are valuable evidence—they prove you were solving genuinely difficult problems.
Phase 3: Maximizing Claims & Avoiding Audits
Contractor & Vendor Checklist
How you structure your external relationships can make or break your R&D credit claim. The IRS scrutinizes contractor arrangements carefully, and common mistakes here trigger audits more than any other factor.
1
Geographic Location
Are contractors physically located in the United States? This is non-negotiable. Paying a US agency that subcontracts work overseas does NOT qualify—the IRS looks at where the actual work is performed, not who cuts the check.
2
Intellectual Property Rights
Does your contract explicitly state you retain all IP? If contractors own the code they write, the IRS considers them to be bearing the development risk, not you. Your contracts must clearly transfer all IP rights to your company.
3
Risk Structure & Payment Terms
Are you paying Time & Materials (hourly/daily rates)? This is the safest structure. Fixed-fee or flat-rate contracts are risky—the IRS argues the consultant holds the technical risk, not your company, which can disqualify those expenses.
Documentation Essentials
Source Control & Project Management
Tag Git commits and Jira tickets as "New Dev," "Spike," or "Investigation." This creates a contemporaneous record of R&D activities that's nearly impossible to fabricate after the fact.
Failed Projects & Experiments
Keep detailed records of what didn't work. Failed experiments are gold in an audit—they prove you faced genuine technical uncertainty and used systematic experimentation to overcome it.
Executive Time Allocation
Only claim reasonable percentages of C-Suite time. Claiming 10% for technical strategy and architecture is defensible; claiming 100% of a CEO's time on coding is not. Be conservative and honest.
Phase 4: The ROI Calculation
Understanding Your Potential Credit
The math is straightforward, but the impact is substantial. For most AI companies actively developing technology, the credit represents a significant source of non-dilutive capital that directly funds future innovation.
10%
Average Credit Rate
Typical percentage of qualified expenses returned as a tax credit
$250K
Median Claim
Average annual credit for mid-stage AI startups
40%
Miss Out
Portion of eligible companies that fail to claim credits
1
W2 Wages
Salaries paid to employees directly engaged in qualified R&D activities
2
US Contractor Spend
Payments to US-based contractors working on eligible development (65% of total)
3
Cloud Dev Costs
Compute resources used exclusively for development and testing
4
× ~10%
Your Tax Credit
When to Seek Expert Help
Given the complexity of R&D tax laws, engaging a specialist can significantly optimize your claim and reduce audit risk. Look for firms with deep expertise in both tax law and the nuances of the technology sector, especially AI development.
A good specialist will help you identify all qualified expenses, correctly document your R&D activities, and navigate the specific requirements for contractor engagements and intellectual property.
Always consult with a qualified CPA. Tax laws are subject to change. This checklist provides general guidance and does not constitute tax advice.
Your Next Steps: From Insight to Action
You now have the framework to evaluate your R&D tax credit opportunity. The companies that win don't just understand the rules—they proactively structure their development processes, contracts, and documentation to maximize legitimate credits while maintaining bulletproof compliance.
1
Assess Your Eligibility
Review the 4-part test and evaluate your current development activities against the eligible activities checklist
2
Audit Your Contracts
Review contractor agreements for IP rights, payment structure, and geographic location compliance
3
Document Everything
Implement Git/Jira tagging, maintain records of failed experiments, and track time allocation
4
Get Expert Guidance
Partner with specialists who understand both AI development and tax law to maximize your claim
Listen to the full interview with Ari Salafia for deeper insights into R&D tax strategies for AI companies.
The bottom line: Many mid-sized and growth companies are sitting on hundreds of thousands of dollars in unclaimed tax credits. The question isn't whether you qualify—it's whether you're documenting and structuring your work to capture the maximum legitimate benefit. Don't leave money on the table.