ClearFirst — SBA eligibility pre-screen for loan packagers

1. One-liner

Catches the deal-killing SBA eligibility flaw before a packager spends a dollar on appraisals or weeks of underwriting.

2. Trend signal — why now?

The SBA rulebook (SOP 50 10 8) governing who can get a 7(a) or 504 loan has whiplashed three times in twelve months:

• June 1, 2025 — SOP 50 10 8 took effect, rewriting eligibility, underwriting standards, and procedures.
• January 1, 2026 — Procedural Notice 5000-872050 extended eligibility to businesses with ≤5% foreign/LPR ownership held outside the US.
• March/April 1, 2026 — that carve-out was eliminated. Now 100% of direct and indirect owners must be US citizens or US nationals residing in the US. Green-card holders can no longer hold any ownership interest in an applicant, operating company, or eligible passive company.

That’s a goalpost that moved twice in one quarter. The consequence is brutal and asymmetric: an eligibility error makes a loan ineligible for the SBA guaranty from day one. When the SBA later catches it at guaranty-purchase time, the lender eats a “repair” (monetary clawback) or a full denial of liability. Meanwhile the borrower has already paid for an appraisal ($2–5K), a Phase I environmental ($1.5–6K), and the packager has burned weeks of work — all torched the moment someone notices the affiliation chain or an LPR on the cap table.

No off-the-shelf tool screens for this before the spend. The loan-origination platforms (Centrex, Zeitro, LendingPad) are CRMs and document pipelines — they assemble the package, they don’t rule on eligibility. Eligibility analysis today is a human reading a 400-page SOP from memory, or a $400/hr SBA attorney.

Provenance:

• Signal 1 (demand): “The most common reason for an early denial of liability involves eligibility errors — if the borrower doesn’t qualify for the program, the loan is ineligible for the guaranty from day one.” Affiliation/size-standard mapping is a known, repeated failure mode. — https://michaelcarbonara.com/sba-loans-explained-common-denial-reasons/ and https://windsoradvantage.com/news/sba-loan-default-reasons-repair-denial — 2026-06-07
• Signal 2 (feasibility): The SOP rules are now public, structured, and frequently revised — a perfect target for an LLM that ingests the current SOP and a borrower intake and returns a structured eligibility verdict. The Jan→Mar 2026 citizenship flip is fully documented in SBA procedural notices. — https://www.sba.gov/document/procedural-notice-5000-872050-procedural-notice-5000-872050-update-sop-50-10-8-citizenship-residency-requirements — 2026-06-07
• Signal 3 (economic): Packaging fees run $2,500–$5,000 per loan and broker referral/packaging fees are capped at 0–1% of loan amount — there is already real money changing hands per deal, and a dead deal is pure loss. 1,306 active SBA lenders + 800+ Lender Match brokers + thousands of independent packagers. — https://starfieldsmith.com/2025/12/understanding-sba-7a-loan-fees-and-costs/ and https://www.sba.gov/partners/lenders/lender-reports — 2026-06-07 Category: Regulatory arbitrage

3. The opportunity

The gap is a pre-flight check that doesn’t exist. Today the eligibility decision happens implicitly — buried inside a human packager’s head, surfaced too late, or paid for at attorney rates. The pain isn’t “I can’t find the SOP”; it’s “I didn’t apply the current version of three intersecting rules (citizenship, affiliation/size standard, ineligible-use) to this specific cap table and ownership chain, fast, on day one.”

The incumbents to disrupt aren’t software — they’re (a) the SBA attorney doing $1,200 eligibility memos, and (b) the status quo of finding out at guaranty purchase. A focused AI-first tool wins because it does the boring, repeatable part — map the ownership/affiliation graph, check it against the currently effective SOP, flag the citizenship and size-standard landmines — in two minutes instead of two days, and re-runs it free every time the SOP changes (which it now does constantly).

This is regulatory arbitrage in its cleanest form: the regulation itself is the moat. The rules are too fluid for a generic tool and too narrow for a big LOS vendor to bother with, but they’re life-or-death for the people who package these loans.

4. Target market

• Primary customer: Independent SBA loan packagers / consultants and small-to-mid SBA lending shops (the BDO + credit-analyst pods inside community banks, CDCs, and non-bank 7(a) lenders). Think 1–20 person operations that touch 5–50 SBA files a month.
• Why they buy: In their world, a killed-late deal is lost fees + a burned borrower relationship + reputational hit; for the lender, an eligibility miss is a guaranty repair that can cost six figures on one loan. They will pay to never be the person who told a borrower to order a $4K appraisal on a deal that was dead on arrival.
• Rough TAM reasoning: 1,306 active SBA lenders, 800+ Lender Match brokers, and an estimated several thousand independent packagers/consultants. Even 2,000 paying seats at ~$200/mo = ~$4.8M ARR. The addressable population is countable and reachable.
• Why now for them: The Jan→Mar 2026 citizenship reversal just nuked deals that were eligible 60 days earlier. Every packager who got burned in Q1 2026 is now paranoid and actively looking for a way to not get burned again.

5. Product sketch (MVP)

• Paste or upload a borrower intake (entity structure, owners + citizenship status, NAICS, affiliates, use of proceeds) → get a GO / FLAG / KILL eligibility verdict in under two minutes.
• Ownership & affiliation mapping: builds the ownership graph, applies the >50% affiliation tests and the applicant-plus-affiliates size-standard check against the borrower’s NAICS.
• Citizenship/residency screen: flags any owner who is an LPR or non-citizen, or any US-citizen owner whose principal residence is outside the US — the exact landmines that changed in 2026.
• Use-of-proceeds & ineligible-business check against the current SOP’s ineligible categories.
• Versioned rulebook: every verdict is stamped with which SOP/procedural-notice version it was run against, so a verdict run in February vs. April is auditable and re-runnable.
• Plain-English “why” memo for each flag, with the SOP citation — drop it in the file as the eligibility work product (replaces the attorney memo for the common cases).
• Change alerts: when the SOP changes, auto re-screens the customer’s open pipeline and flags newly-dead deals.

6. AI angle — what’s load-bearing

Remove the AI and you have a checklist PDF — which already exists and which nobody reads correctly under deadline pressure. The load-bearing work is: (1) parsing messy, unstructured borrower intake (org charts, cap tables, narrative descriptions) into a structured ownership/affiliation graph, and (2) reasoning over the intersection of three rule families against the current SOP text and returning a cited verdict. That’s exactly the unstructured-input → structured-rule-application task LLMs got good at. The SOP is the ground truth; the model’s job is faithful application + citation, not invention. This is RAG-over-regulation with a structured-output verdict — load-bearing, not decorative.

7. Localization angle (if any)

N/A — this is a US-only play by definition. SBA loans are a US federal program; the regulation is the product. There is no India/SEA cut. The “localization” here is regulatory-vintage localization: being pinned to the exact currently-effective SOP version is the whole game.

8. Business model — path to $1M–$5M ARR

• Pricing: Tiered SaaS. Solo packager: $99/mo (up to ~15 screens). Small shop: $299/mo (unlimited screens, 3 seats). Lending team: $499–$999/mo (more seats, pipeline auto-re-screen, audit export). Optional per-screen overage for high-volume.
• ACV: ~$2,400–$6,000 blended.
• Rough math to $1M ARR: ~350 customers × ~$240/mo × 12 ≈ $1.0M. Very reachable against a base of 1,300+ lenders and thousands of packagers.
• Rough math to $5M ARR: ~1,500–1,800 paying accounts, or fewer accounts trending toward team tiers ($499+). Requires landing inside lenders, not just solo packagers.
• Expansion path: seat expansion within lending teams; pipeline auto-re-screen as a premium add-on; adjacent rule engines (504 vs 7(a) specifics, franchise eligibility, change-of-ownership rules); a white-label API so an LOS vendor embeds the eligibility verdict.

9. Go-to-market wedge — first 100 customers

• The Q1-2026 burn list: Mine SBA-lending LinkedIn, the NAGGL community, and SBA-attorney blog comment sections for packagers/BDOs who publicly griped about the citizenship reversal. Cold-DM each a 90-second Loom showing their own style of deal run through ClearFirst flagging the LPR landmine. Expect 5–8% reply on a hyper-relevant pain.
• Ride the SOP-change news cycle: Every time the SBA issues a procedural notice, publish a same-day “what changed and who’s now ineligible” breakdown + a free single-deal screen. The audience is already searching that exact query that week.
• Partner with SBA-lending trainers/consultants (CapitalAx-style broker-training programs, NAGGL course instructors) — they have the email lists of new packagers who are most scared of making an eligibility mistake. Rev-share or affiliate.
• Land-and-expand via one lender pod: sell one community-bank SBA team, get the case study (“caught 3 dead deals in month one”), use it to walk into the next 20 PLP lenders.

10. Build complexity — justification

Medium. Off-the-shelf: LLM with structured output, RAG over the public SOP/procedural notices, standard web stack. Custom work: encoding the affiliation/size-standard logic as a deterministic rule layer the LLM feeds into (you do not want the model free-styling a >50% ownership test), building the versioned-rulebook ingestion so each SOP revision is a dated dataset, and an eval harness against known-good/known-bad fact patterns. A technical founder with an SBA-eligibility advisor ships a credible v1 in ~3–4 months. The hard part is correctness and trust, not infrastructure.

11. Gating checklist

Gate	Pass?	Note
Legal in target market	✅	Advisory/screening tool; it informs an eligibility decision, doesn’t make the loan. Disclaim “not legal advice.”
Ethical — no harm / dark patterns	✅	Reduces wasted borrower spend and bad-faith approvals. Net-positive.
Market exists (evidence above)	✅	Eligibility errors are the most common early denial reason; real money per deal.
1–5 person team can build this	✅	Technical founder + domain advisor, ~3–4 months.
Launchable with <$50K / ₹40L	✅	Inference + hosting + advisor time. Well under $50K.

12. Feasibility score

Axis	Weight	Score	Notes
Problem intensity	20	16/20	Hair-on-fire for the lender (six-figure repair risk) and the packager (lost fees + relationship). Frequency is per-deal, weekly+. Not a daily grind, which caps it below 17.
Demand evidence	15	12/15	Strong: documented as the #1 early-denial cause, real per-deal spend, 3 rule flips in a year. Docked for lack of verbatim customer quotes — pain is sourced from legal/industry analysis, not yet from interviews.
Build feasibility	15	11/15	Doable in 3–4 months. Correctness bar is high; the deterministic affiliation logic + eval harness is real engineering, not a weekend.
Distribution clarity	15	12/15	Countable, reachable audience; a clear “burn list” + SOP-news-cycle wedge. Not a 2-week sprint, hence not 13+.
Revenue mechanics	15	12/15	Pricing benchmarked against $2.5–5K packaging fees and $400/hr attorneys — easy to justify $99–499/mo. ARR math is reasonable; team-tier penetration is the swing assumption.
Time to first revenue	10	7/10	A scared packager can be sold in weeks once the tool demonstrably catches a real landmine. Needs a trust-building pilot, so not sub-4-weeks.
Defensibility	10	5/10	Moat is regulatory currency + accumulated eval set + workflow lock-in, not tech. A well-funded LOS could bolt this on — but the rules are too narrow/fluid for them to prioritize, and being first and trusted on each SOP change compounds.
Total	100	75/100

13. Qualitative modifiers

Founder-fit tags

domain-expertise-required · technical-heavy

You need a real SBA-eligibility brain in the building (or as a co-founder/advisor). Get the affiliation or citizenship logic subtly wrong and the product is worse than useless — it gives false confidence on a six-figure-risk decision. The technical build is tractable; the domain correctness is the bar.

Key assumptions to validate (3–5)

1. Assumption: Packagers/lenders will trust a software verdict enough to act on it (or at least to stop spending before a human confirms). How to test: Run 20 real anonymized historical deals (including known-dead ones) through a manual prototype; ask 15 packagers “would you have ordered the appraisal if you’d seen this?”
2. Assumption: The eligibility verdict can be made accurate enough on the common fact patterns to be relied upon. How to test: Build an eval set of 100 labeled fact patterns from SOP examples + attorney write-ups; target >95% on citizenship/affiliation/use-of-proceeds before charging.
3. Assumption: Willingness to pay $99–499/mo is real, not just nodding. How to test: Pre-sell 10 annual pilots at a launch discount off the cold-outreach burn list before building the full product.
4. Assumption: SOP volatility continues (or at least the fear of it persists). How to test: It already flipped 3× in 12 months; monitor for stabilization — if rules freeze for 18 months the “auto re-screen” value erodes but the core check remains.

Risk flags

1. Liability/trust risk: A wrong “GO” on a deal that’s actually ineligible is a serious failure. Mitigate with conservative FLAG-don’t-GO defaults, mandatory human sign-off, clear “not legal advice” framing, and a citation for every verdict.
2. Regulatory-dependency risk: The product lives and dies by the SBA program existing roughly as-is. A major program restructuring is both a risk (rebuild) and an opportunity (everyone needs re-screening). Net: the volatility that creates the risk is also the demand engine.
3. Platform/incumbent risk: An LOS vendor (Centrex, Zeitro) or an SBA-attorney firm could add an eligibility module. First-mover trust on each SOP change + an accumulating eval set is the defense, not patents.

14. Structured verdict

Score:                  75/100
Verdict:                GO
Confidence:             Medium
Best-fit builder:       Technical founder + SBA-eligibility domain expert (co-founder or close advisor)
Time to revenue:        8–12 weeks (pre-sold pilots possible earlier)
Capital to launch:      $15–30K ($ inference, hosting, advisor time, eval-set build)
Top 3 assumptions to validate first:
  1. Packagers will act on a software eligibility verdict — test with 20 historical deals + 15 interviews
  2. Verdict accuracy >95% on common fact patterns — build a 100-case labeled eval set before charging
  3. WTP $99–499/mo — pre-sell 10 annual pilots off the Q1-2026 burn list
Kill criteria:
  - Abandon if <3 of 15 interviewed packagers say a software verdict would change their pre-spend behavior
  - Abandon if eval accuracy on citizenship/affiliation can't clear 95% within the build window
  - Abandon if the SBA program is restructured so heavily that eligibility becomes trivial/automated upstream

15. Next step — 1-week validation sprint

• Day 1–2: Pull 20 real (anonymized) SBA fact patterns — mix of clean, affiliation-trap, and post-March-2026 citizenship-trap deals. Hand-run each through the current SOP to establish ground-truth verdicts.
• Day 3–4: Build a no-code prototype: paste intake → ClearFirst returns GO/FLAG/KILL + cited reasoning. Run all 20 cases; measure where it agrees with ground truth.
• Day 5: Put it in front of 15 packagers/BDOs from the burn list. Go signal: ≥8 of 15 say “I would have stopped spending / I’d pay for this,” AND the prototype hits ≥90% agreement with ground truth on the citizenship + affiliation cases. No-go: below either bar — the trust or the accuracy isn’t there yet.

The result is falsifiable: a counted agreement rate against ground truth, and a counted yes/no on changed pre-spend behavior. Not “people seemed interested.”