Cada has written 100+ grant proposals across 30+ agencies, including NIH, NSF, NASA, ARPA-H, and AFWERX.
About 1 in 5 companies Cada builds grant roadmaps for get flagged for having no peer-reviewed publications (internal Cada pipeline data, 2026). Most of them treat it as disqualifying. It usually is not. If you came out of industry rather than academia, you probably have more usable evidence than you think.
The short answer: you do not need peer-reviewed publications to win an SBIR. Reviewers want preliminary data: results your team generated that support the feasibility of your proposed work. Unpublished bench data, pilot deployments, and production metrics all count, if each claim carries a metric, a comparison, and ideally a figure. Publications are one container for preliminary data, not the requirement itself.
The real question is not "do I have publications?" It is "which of four evidence categories does my data fall into, and does each claim pass a three-part test?" This guide walks through both. It is the same methodology Cada uses internally across 100+ proposals and 30+ agencies.
What do NIH reviewers mean by preliminary data?
Preliminary data is any result your team has generated that supports the feasibility of the proposed research: bench measurements, prototype benchmarks, pilot deployments, retrospective analyses of existing datasets. It does not need to be published. It needs a number, a comparison, and a form your reviewers can evaluate.
Here is the uncomfortable part. NIH does not technically require preliminary data for an SBIR Phase I. In practice, NIH study sections almost always penalize Phase I proposals that lack it. Even small-scale proof-of-concept results measurably improve scores.
An NIH study section is a panel of 15 to 20 scientists scoring your application 1 to 9 on five criteria. These are people who read hundreds of applications. A feasibility story with no numbers behind it does not survive that room, whatever the official guidance says about Phase I expectations.
NSF frames it differently but lands in the same place. If your project pitch or proposal claims prior results, those claims need numbers attached. A feasibility assertion without a measurement reads as a weakness, not as evidence.
So the bar is real. But "preliminary data" and "publications" are different things, and conflating them is where founders talk themselves out of applying.
The four evidence categories reviewers accept
Cada tags every preliminary-data claim in every proposal we write with a source category before drafting. The taxonomy exists because conflating literature results with team-side results is the second most common reviewer-rejection pattern we see, after undefended novelty claims (internal Cada review data, 100+ proposals).
Here are the four categories, strongest to weakest.
| Category | Whose data | Counts as preliminary data? | Typical phrasing |
|---|---|---|---|
| A: Team-side primary | Your company, or a named co-investigator on this proposal, generated it | Yes, strongest form | "In March 2026 we measured X = Y under Z conditions (Figure 2)" |
| B: Team-side adjacent | Your team generated it, but under different conditions than the claim requires | Yes, as supporting evidence with an honest caveat | "Demonstrated at room temperature; the first cryogenic test is a Phase I objective" |
| C: Attributable prior work | A named PI or subcontractor's published work from before they joined your proposal | Only if they join the team; otherwise supporting evidence with explicit attribution | "Dr. X demonstrated Y in prior work at [university] [citation]" |
| D: Literature only | Third parties with no team involvement | No. Belongs in your background or significance sections, never in a preliminary data section | "Prior studies have shown Y [citation]" |
Category A: team-side primary data
Your lab ran the experiment, or a named co-investigator or subcontractor on this proposal did. A figure or numeric result exists. The work happened within the last 5 years.
This is the strongest evidence you can put on the page. A fictional example: a sepsis-prediction startup ran its model on 3 years of de-identified retrospective records from a partner health system and measured sensitivity against the incumbent early-warning score. That is Category A, publication or not.
Category B: team-side adjacent data
Your team generated real data on the same system, materials, or method, but under different conditions than the proposed work requires.
Category B is legitimate supporting evidence if you caveat it honestly. A fictional example: a membrane materials company has 40 hours of performance data at room temperature and is proposing to validate at cryogenic temperatures. The honest framing is "demonstrated at 293 K; the first cryogenic test is the Phase I objective." Reviewers respect that framing. They punish the version that blurs the conditions.
Category C: attributable prior work
A named scientist published relevant results at a university or another company before joining your effort. Their work counts as supporting evidence if you attribute it explicitly, and it can anchor your preliminary data section if they join the proposal as a co-investigator or subcontractor.
This category is where non-academic founders leave the most value on the table. If you licensed technology from a professor, or your technical advisor published the foundational work, that evidence is attributable. The upgrade path is concrete: bring them onto the proposal.
Category D: literature only
Published work by third parties with no involvement from anyone on your team. It belongs in your significance and background sections, where it does real work.
What it cannot do is serve as your preliminary data. A preliminary data section that contains only Category D citations has a structural problem reviewers spot in seconds.
A recurring reviewer question on NASA panels makes the point: were the researchers on this proposal actually involved in the investigations being cited? If the answer is no and the section says "preliminary data," you have already lost credibility.
The metric-comparison-figure test: is it evidence or an assertion?
Source category answers whose data this is. There is a second, equally decisive question: is the claim stated as evidence, or merely asserted? Even a Category A claim can fail here.
A preliminary-data claim passes the test only if it carries all three of these, framed like an experiment: a metric (a number with a unit), a comparison (versus a baseline, the standard of care, or the prior method), and ideally a figure or table the reviewer can look at.
Two fictional versions of the same claim:
Fails the test: "Our triage model is deployed at a 400-bed hospital and performs well in production."
Passes the test: "In a 6-month production deployment at a 400-bed hospital (n = 1,850 encounters), the model flagged deteriorating patients with 84% sensitivity versus 62% for the incumbent early-warning score (Figure 2)."
The first version is an assertion of activity. The second is a result. Reviewers only score results.
The rule Cada enforces internally: every claim in a preliminary data section must pass. If a claim fails, either supply the metric and comparison, or demote the claim out of preliminary data and into the work plan as a Phase I objective. Never leave a bare deployment or validation assertion standing in a section reviewers read as evidence.
What if all your evidence is from the literature?
If everything you have is Category D, you have three structural fixes. All three are legitimate. Papering over the gap is not.
Fix 1: reframe the section. Retitle it as scientific premise or background and remove the preliminary data heading. Third-party literature does strong work under the right heading and damages you under the wrong one.
Fix 2: elevate Category C. If any of the work you are citing was done by someone you can name and recruit, bring them onto the proposal as a co-investigator or subcontractor. Their prior results become attributable evidence, with explicit attribution language.
Fix 3: own the gap. Acknowledge that no team prior data exists, which is defensible for a Phase I in a genuinely new application space. This only works if two things are in place: a sharply defended novelty claim (what specifically is new, and why nobody has measured it before) and a named first-time-execution risk in your risk discussion. Reviewers accept "nobody has done this, and here is how we de-risk the first attempt"; they do not accept silence.
Frankly, fix 3 is the hardest to pull off, and its success depends on how genuinely new your application space is. If three published groups have already measured what you are proposing to measure, the absence of your own data is a real weakness, not a framing problem. In that situation, run the cheapest experiment instead.
The cheapest experiment that moves you up a category
Most founders who think they have nothing are actually holding Category B or C evidence without recognizing it. The rest can usually generate Category A evidence in weeks, not years.
The comparison founders get wrong: a peer-reviewed publication typically takes 12 to 18 months from experiment to journal acceptance. A category upgrade takes 2 to 6 weeks. Reviewers do not require the journal's stamp. They require the number, the comparison, and the figure.
Cheap-experiment patterns that produce Category A data:
- Retrospective analysis of public or licensed datasets. If your technology is computational, benchmark it on a public dataset against the best published method. Cost: staff time, typically 2 to 4 weeks.
- Benchtop pilot on surrogate samples. A materials or device claim tested at reduced scale, with honest scale caveats, converts a literature argument into team-side data for a few thousand dollars of consumables.
- Head-to-head benchmark of your prototype. Run your existing prototype against the incumbent method on 20 to 50 samples. Small n with a clean comparison beats large claims with no comparison.
- Reframe production metrics you already have. Industry founders often sit on deployment data that just needs evidential-form framing: pull the sensitivity, the baseline, and the time window out of your logs and build the figure.
One honest caution: small-n data invites statistical scrutiny. Present it as proof-of-concept supporting feasibility, with n stated plainly, not as definitive validation. Overselling a 20-sample pilot costs more credibility than the pilot earned.
How NIH, NSF, and NASA weigh preliminary data differently
The four categories and the three-part test apply everywhere. How heavily each agency leans on them varies.
| Agency | Expectation | Where it binds |
|---|---|---|
| NIH | Heaviest. Technically not required for Phase I, but study sections routinely penalize its absence | Preliminary Studies subsection of the Research Strategy; feasibility claims in Specific Aims |
| NSF | Conditional. Binds when you claim prior results; feasibility must be shown with numbers, not asserted | Intellectual merit framing in the Project Pitch and Project Description |
| NASA | Strong team-side expectation. Reviewers probe whether cited investigations involved your team | Technical volume preliminary data discussion |
Practical implications:
- Applying to NIH with zero team-side data is the hardest path. NIH Phase I budgets typically run around $300,000, and study section culture expects pilot results at those stakes. If you are pre-data, the cheapest-experiment section above is your pre-application to-do list.
- NSF is the most forgiving venue for a pre-data founder, because the expectation binds only when you claim prior results. A tightly argued approach with a de-risking plan can carry a Phase I pitch. Do not manufacture weak claims to fill the gap; an absent claim is better than a failed one.
- NASA punishes attribution ambiguity hardest. If you cite investigations your team was not part of, say so explicitly and place them in your background discussion. The recurring reviewer question is exactly "was your team involved in this work?"
I should note: expectations shift with solicitation cycles and review panels, and no methodology guarantees an award. What the taxonomy guarantees is that you will not lose points for a structural mistake that was avoidable before you wrote a single section.
Where this fits
Preliminary data is one gate among several. If what you have is customers rather than data, the commercial traction vs publications rubric maps which agencies credit traction instead. Before you invest in either, the 4-question founder intent filter tells you whether SBIR is the right instrument for your company at all, and the 7-point grant readiness assessment places your evidence gap among the other things reviewers score. If publications are on your roadmap anyway, patent vs publication timing covers the sequence that protects both.
Get a straight answer on your preliminary data
Most founders misclassify their own evidence, in both directions. Industry founders undersell production data that would pass the test with better framing. Academic founders oversell literature that reviewers will discount on sight.
Cada does a free preliminary-data audit: we classify your existing evidence into the four categories, run each claim through the metric-comparison-figure test, and identify the cheapest experiment that moves you up a category, before you commit 40+ hours to an SBIR application. It is a 15-minute call with a straight answer at the end.
If your evidence is weaker than you hoped, you will know exactly which experiment fixes it. If it is stronger than you feared, you will know which program to point it at.
No pitch, no obligation, nothing to pay. Book a free preliminary-data audit.
Sources
- NIH SEED -- NIH small business program structure, Phase I application guidance, and budget norms
- NIH Center for Scientific Review -- how study sections are assembled and how the 1-to-9 scoring works
- NSF America's Seed Fund -- Project Pitch mechanism and Phase I proposal requirements
- NASA SBIR/STTR -- solicitation and technical volume requirements
- The four-category taxonomy, the metric-comparison-figure test, the 1-in-5 figure, the reviewer-rejection-pattern observations, and the cheap-experiment cost estimates are Cada's own methodology and internal pipeline data (100+ proposals across 30+ agencies)
Agency expectations shift with solicitation cycles and review panels, so verify requirements against the live solicitation you are applying under. All company examples are fictional and used for illustration only.