Writing a grant proposal for an AI project intimidates most first-timers, but the structure is learnable and the pitfalls are predictable. Since the vast majority of federal AI grants for businesses are SBIR/STTR awards, this guide focuses on that structure — what to write, what reviewers reward, and where applications go wrong, cited to official sources. (dgm builds the AI; grant-writing is best done with your SBDC — see the end.)

Before you write: is this actually R&D?

The single biggest determinant of success happens before the first sentence. Federal grants like SBIR/STTR fund research and development — projects with genuine technical uncertainty that you resolve through experimentation. They do not fund adoption — deploying existing AI across your operations. If your project is “we’ll implement an AI tool,” no amount of good writing will make it fit. If it’s “we’ll develop a novel AI capability whose feasibility is genuinely uncertain,” you’re in the right program. Be honest here; reviewers will be.

Know the gate: pitch first, or full proposal?

Don’t start with a 15-page proposal until you know the agency’s model. Several agencies — notably NSF’s America’s Seed Fund — require a short Project Pitch first; only if it’s invited do you write the full proposal. Other agencies accept full proposals directly against a solicitation deadline. Identify your target agency’s process before investing the hours.

The structure of an SBIR/STTR proposal

While each agency sets its own sections and page limits in the solicitation, a typical proposal covers:

  1. The innovation and the problem. What are you building, what technical problem does it solve, and why is it new? Lead with the innovation, clearly.
  2. Technical objectives and work plan. Specific, measurable objectives and the tasks, methods, and milestones to reach them. Reviewers want to see how you’ll do the work, not just what you hope to achieve.
  3. Technical risk and feasibility. Name the real uncertainties and explain how your approach addresses them. Acknowledging risk credibly is a strength, not a weakness — it’s what distinguishes R&D from routine work.
  4. Team and facilities. Why is your team capable of doing this? For STTR, this is also where the required nonprofit research-institution partnership appears (the small business does ≥40% of the work, the institution ≥30%).
  5. Commercialization plan. Increasingly decisive: a credible path to market — customers, market size, business model, and how the technology gets from prototype to product. Many agencies score commercial potential heavily.

What reviewers reward

Across agencies, strong proposals tend to share traits:

  • Clarity. Reviewers read many proposals; a clear innovation statement and a legible work plan stand out.
  • Genuine technical merit and feasibility. Ambitious but credible — not so safe it isn’t research, not so speculative it isn’t achievable.
  • A real commercialization story. The science alone rarely wins; the path to market increasingly does.
  • A capable, credible team. Reviewers fund people who can execute, not just ideas.

The pitfalls that sink applications

  • Framing adoption as R&D. The most common fatal error. If your project doesn’t have genuine technical uncertainty, it won’t fit — find a different lever (a tax credit or loan) instead of forcing a proposal.
  • Late or incomplete federal registrations. Federal submission systems require registrations that take time to complete. Many applications fail on administration, not ideas. Start registrations weeks ahead of any deadline.
  • Ignoring the solicitation. Page limits, required sections, and topic alignment are not suggestions. Follow the specific solicitation precisely.
  • A weak or missing commercialization plan. Treating it as an afterthought costs scores at agencies that weight commercial potential.

Get free help — and don’t go it alone

You don’t have to write this in isolation. Your local SBDC offers free advising and frequently helps businesses pursuing SBIR/STTR. Some states also run programs that match or otherwise support federal applications (New York’s NYSTAR matches SBIR/STTR up to $200,000, for example). Using these resources is free and materially improves applications.

How dgm helps

dgm implements osFoundry and other AI for US businesses. Grant-writing itself isn’t our service — for that, lean on your SBDC or a specialist grant writer. What dgm can do is help you judge honestly whether your AI initiative is genuine R&D worth proposing, and then build and deploy the AI. A funded proposal is a means; a working AI system is the end — and the build is where we come in.