From SBIR Award to a Ranked Deep-Tech Deal List

Federal non-dilutive grant awards are a public, dated, first-party feed of deep-tech companies, and most of them are structurally pre-venture-round by rule. This guide is for early-stage investors, platform and talent partners at funds, and angels who want to read the SBIR and NSF award stream backward into a timed deal pipeline. I carry one worked example all the way through: an energy-storage thesis, the real queries, the intermediate counts, and the wrong turns, so you can follow along on your own case.

Every ranking page on the open web teaches founders how to win an SBIR. This does the opposite. It teaches you how to read the award feed as a sourcing channel, tell a venture-scalable awardee from a perpetual grant-dependent shop, and reach the founder before the priced round.

Why the award feed beats headline search

The award database is a non-obvious sourcing edge because founders advertise their domain, not their grants. In Refolk's index of professional profiles, only 3 US founders name "SBIR Phase II" in their headline, while 22 name "robotics autonomy" and 3 name "quantum computing hardware." Grant status is nearly invisible to keyword prospecting on profiles, which means the people working from LinkedIn headlines and press releases never see this signal at all.

The federal award feed carries what the profile hides: a firm name, a phase, a dated award, an obligated amount, and a named principal investigator. Two feeds matter. The SBIR.gov Awards API returns firm, agency, branch, phase, program, proposal award date, award amount, and UEI or DUNS, among other fields. The NSF Awards API on research.gov covers NSF research awards from 2007 and exposes funds obligated, PI name, start date, title, and abstract text, paginated at 25 awards per page.

There is a second structural reason to work this channel. The eligibility rule is itself a pre-round filter. Standard SBIR requires the firm not be majority owned by any single venture, hedge fund, or private equity firm. If one VC owns 51% or more, the firm is not SBIR-eligible at most agencies. So a standard-track DoD or NSF awardee is, by the rule that let it win the award, structurally pre-majority-raise. You did not have to build that filter. The program did.

The eligibility rule that lets a firm win the award is the same rule that keeps it pre-round.

Map your thesis to the right agency feed

Start by translating your technology area into the agencies that fund it, because pulling the wrong feed wastes the whole exercise. Eleven federal agencies participate in SBIR, and the Department of Defense accounts for more than half of the total program. The domains map cleanly enough to shortlist in an hour.

My worked case is an energy-storage and battery-materials thesis. That points at DOE first, NSF second for the broad materials science slice, and DoD third for power and energy autonomy topics. Note the STTR wrinkle: STTR runs at only six agencies (DoD, USDA, HHS, NSF, DOE, NASA), so a thesis built on university spinouts narrows the feed set. If I filter only SBIR, I miss STTR-only spinouts entirely. That is failure mode number four, and I nearly made it here.

Pull the tech-area slice

A tech-area slice is a keyword search on title and abstract, bounded by agency and year, run against both feeds. On SBIR.gov, query the Awards API or the bulk download by keyword plus agency plus year. On NSF, query the Awards API by keyword on title and abstract. The output you want is a raw award table with one row per award: firm, phase, date, amount, and PI.

Two mechanical limits will bite you. First, the SBIR.gov download page caps you at 10,000 records per pull; the full award file is 290 MB with abstracts and 65 MB without. For a narrow thesis you will stay well under the cap, but a broad keyword like "energy" will blow past it, so tighten the terms or slice by year. Second, the SBIR.gov site warns its APIs are undergoing maintenance. That means a slice can come back partial. Read an empty or thin result as "verify," not as "no awardees."

For my energy-storage slice I ran the DOE and NSF feeds with "energy storage" and "battery materials" on title and abstract, bounded to the last several fiscal years. A single reference point on volume: a June 2026 pull of the NSF feed found 142 Phase I or STTR Phase I title records for 2025 with a median obligated amount of $304,998. NSF funds roughly 400 companies each year across nearly all technology areas, so any single-domain slice is small enough to read by hand once collapsed.

Read the award amounts so a number never surprises you

Knowing the award ceilings lets you sanity-check every row and spot the Phase II milestone that matters for timing. NSF Phase I is up to $305,000 for 6 to 18 months, and Phase II up to $1,250,000, typically over 24 months. Statutory policy caps, adjusted annually, sit higher because agencies can issue larger awards without a separate SBA waiver.

The practical read: a Phase I row near $305,000 is a normal first grant, and a Phase II row is your timing beacon. Phase II is where a firm begins gathering commercialization evidence, which is why a fresh Phase II date is the natural pre-raise outreach trigger. NSF ties its own portfolio to $36 billion in follow-on private investment and 380 exits over fiscal years 2015 to 2025, pulled from Pitchbook. A separate NSF infographic reports $32B in follow-on capital, 420 exits, and 62 firms valued over $250M over the last ten years. Those are aggregate totals. A precise median lag from Phase II to priced round is not publicly established, so rank by Phase II recency rather than a fabricated day count.

Collapse awards to companies, then screen out the mills

Group the raw award rows to one company per row by UEI or firm name, count Phase I and Phase II awards per firm, and note first and most-recent award dates. Only now can you screen, because the mill signal lives in the count-and-context, not in any single award.

An SBIR mill is a firm that lives on serial non-dilutive awards without converting them into commercial revenue. The federal benchmarks make this measurable and, better, public. SBA calculates transition rates daily across all agencies and posts them on the SBIR.gov Company Registry for firms over the threshold, so the mill screen is free.

Here is the wrong turn I made. In my energy-storage slice, one firm had 12 Phase II awards and I moved to tag it a mill. That is a false positive. The Commercialization Benchmark only triggers at 16 or more Phase II awards in ten fiscal years, excluding the two most recently completed years, and even then a firm passes if it shows an average of at least $100,000 of sales or investment per Phase II award, or patents equal to 15% of awards. The screen is a high count plus a failed commercialization test, never the count alone. I put the firm back on the list and checked its transition rate on the Company Registry instead.

The concentration data explains why the mill screen matters at all. The top 25 DoD SBIR firms cumulatively took 18% of all DoD SBIR/STTR Phase I and Phase II funding, more than $2.3B from FY2012 to FY2021. And only 4 of those 25 generated more in Phase III contracts than they took in non-dilutive awards. Grant volume alone predicts grant-dependence, not commercial scale. A naive keyword slice is dominated by these incumbents, so ranking by low prior-award count is what isolates the fresh, fundable names.

Screen out the already-funded, then rank by timing

Two firms remain to remove before ranking: those already majority-VC-owned, and those that closed a priced round after the award. Both are "already funded" for your purpose, and both need a specific check.

For majority-VC ownership, use the eligibility rule in reverse. A firm majority-owned by multiple venture, hedge, or PE firms must file a VCOC Certification, and that path is N/A for STTR. Only some agencies opt in under Section 5107: NIH is the most prominent opt-in agency, NSF opts in under specific conditions, and DoD does not allow majority-VC ownership for SBIR awards. So a standard-track DoD or NSF SBIR award is, by rule, evidence the firm is not yet majority-VC-owned. Do not overread this. Minority VC on the cap table is fine and common; the disqualifier is majority ownership, not any VC at all. Dropping a strong standard-track firm because it took a small angel or seed check is failure mode number three.

For a post-award raise, the rule cannot help you, because a firm can close a priced round after its award date. Cross-check external funding databases for each finalist. A Phase II winner from eighteen months ago may already be Series A. Confirm pre-round status against public funding records before you call any name pre-round.

Now rank. Sort by recency of Phase II award and proximity to the typical follow-on window, and prioritise fresh Phase II winners showing commercialization signals such as patents or early revenue.

In my energy-storage case, this collapsed a slice of a few dozen awards to a short list of first-time and low-count Phase II winners, standard-track, with recent award dates. That is the deal list. The example query below is the shape of the ask that produces it directly.

Refolk is where this stops being a spreadsheet exercise. Once you have the ranked firms, the friction is attaching a real, contactable founder to each name and confirming the PI is not a hired technical lead. Asking Refolk in plain English for the founders behind a slice, rather than exporting award rows and hunting each one by hand, is where the timed pipeline actually gets built.

How this goes wrong

The screening is where this method earns its keep, and it is also where it fails quietly. Here are the eight ways a slice lies to you, and what to check.

Two of these deserve extra weight because they waste the most time. The PI-is-not-the-founder trap is specific to this channel: the named principal investigator on a federal award can be a hired technical lead, not an equity founder. Confirm title and ownership before you treat the PI as your outreach target, or you will pitch a term sheet to an employee. And the API-maintenance trap is silent by design: a slice that comes back thin because the API is mid-maintenance looks identical to a real empty result. Always confirm a surprising zero against the bulk download and the NSF feed before you conclude a domain has no awardees.

Founder discoverability: why enrichment is uneven

Once you have a ranked firm list, the last mile is finding the person, and how hard that is varies sharply by domain. The award record gives you a firm and a PI name; turning that into a founder you can reach is where domains diverge. Refolk's index shows the spread directly.

Read this as a warning about which channel to lead with. In a domain like robotics autonomy, 22 founders headline the domain, so profile search and the award feed reinforce each other. In energy storage and battery materials, only 1 US founder headlines the domain in Refolk's index, so the award feed is not just an edge, it is nearly the only structured way in. And across every domain, only 3 US founders headline SBIR Phase II at all, which is the whole reason the award database outperforms headline prospecting: the grant is invisible on the profile, so the award record is your primary key.

The practical consequence: in headline-thin domains, do not expect to confirm a founder from a profile search. Work from the firm and PI, then resolve identity by asking for the person behind the company rather than searching for the credential.

The procedure, start to finish

Here is the full method as a checklist you can run against your own thesis. It maps one-to-one to the steps above.

Keeping the pipeline current

This is a feed, not a one-time pull, so the value is in re-running it on a cadence tied to award cycles. New Phase II awards post continuously, and each fresh Phase II date is a new timing beacon. Re-run the slice on a schedule, diff against your last pull, and route new low-count, standard-track Phase II winners straight to the top of the queue.

Three things will change under you and should be re-checked rather than memorized. First, the benchmark thresholds are in flux, with a proposed rule targeting firms with more than 25 Phase II awards in five fiscal years, so re-confirm the current trigger before each screening pass. Second, the award ceilings adjust annually, so treat the statutory caps as a moving figure and re-read them each year. Third, the SBIR.gov API status changes, so verify the feed is live before trusting a thin result. The program itself lapsed on 30 September 2025 and was reauthorized in April 2026 through 30 September 2031, a reminder that even the feed's existence is worth confirming on a cadence rather than assuming. Build the re-check into the workflow, and the award stream becomes a standing, first-party deal pipeline instead of a project you do once.