ClerkLane

1. One-liner

Finds the records that respond to a public-records request, drafts the exemption log, and logs a defensible search trail.

2. Trend signal — why now?

Three things are converging in 2025–2026.

Request volume is climbing double digits at exactly the agencies that can’t absorb it. Ann Arbor’s deputy clerk handled 640 requests in 2023, 740 in 2024, 860 in 2025 — as the sole person responsible. Jacksonville, IL took 40% more requests in 2025 than 2024. The redaction and search work that large metros split across teams “ends up in one office at smaller departments, with one person often handling intake, scoping, redaction, review, release, and reporting on the same desk with no dedicated discovery support” (Michigan Advance, CivicPlus, Feb 2026).

The hard part has no tool. The incumbent stack — Granicus GovQA, CivicPlus/NextRequest, JustFOIA — is request ticketing: a portal, a clock, a status field. They explicitly do not go into the mailbox and find what’s responsive. Redaction tools (Redactable, CaseGuard, SafeRedact, Vidizmo) redact documents you already found. The middle step — which of these 4,000 emails and 200 files actually respond to “all communications about the Elm Street rezoning, Jan–Jun” — is still a human reading evenings.

AI-for-public-records just got legitimized. Long-context, cheap inference now reads an entire mailbox for plain-English responsiveness; a law professor testified to Congress in 2026 that AI could change public records processing (StateScoop, 2026). GovTech money is moving (Starbridge raised a $42M Series A led by Craft Ventures, Oct 2025).

Provenance:

• Signal 1 (Demand): Single-person clerk offices drowning — Ann Arbor 640→740→860 requests/yr one person; redaction/scoping lands on one desk at small agencies — https://michiganadvance.com/2026/02/21/process-of-foia-the-unseen-work-of-providing-government-transparency/ — 2026-02-21
• Signal 2 (Economic): Missed-deadline fee-shifting + $1,000 punitive awards; school districts paying “hundreds of thousands” in OT/new hires; ~$1,524 cost per manual DSAR — https://www.thrunlaw.com/news/clock-ticking-avoid-financial-liability-missing-foia-deadline + https://www.civicplus.com/blog/rr/real-cost-of-public-records-requests/ — 2026-05
• Signal 3 (Feasibility): Incumbent tools are ticket trackers, not record finders; fragmented email/drive/RMS systems make manual search the bottleneck; long-context LLMs now make responsiveness scoring viable — https://www.civicplus.com/blog/rr/current-issues-affecting-public-records-request-management/ — 2026-05 Category: Underserved niche

3. The opportunity

The public-records workflow has three steps: (1) intake/track the request, (2) find the responsive records and justify exemptions, (3) redact and release. Step 1 is a solved, crowded market (ticketing portals). Step 3 is a solved, crowded market (redaction software). Step 2 — the one that actually consumes the clerk’s evenings and creates the lawsuit risk — is unowned for any agency too small for an e-discovery platform.

The incumbents don’t do Step 2 because their business model is procurement-locked ticketing seats, and searching a customer’s live mailbox to make a legal-responsiveness call is scary (hallucination on a statutory duty). That fear is exactly the moat: the product has to be built as a defensible-trail generator a clerk reviews and signs, not an autonomous decider. Generic AI players won’t touch the liability; e-discovery vendors (Relativity, Logikcull) price for litigation, not a 6,000-person town. ClerkLane is the focused tool that does only Step 2, for the smallest custodians, and produces an audit trail a clerk can defend in front of a judge.

4. Target market

• Primary customer: The single records custodian at a small US public agency — town/city clerk (population <50K), school district administrative secretary, or special-district administrator (water, fire, library, transit, parks). No dedicated FOIA officer, no e-discovery, no in-house IT.
• Why they buy (in their words): “they have boxes and boxes of years-old records and may not even know what’s in them”; emails, shared drives, RMS, video and social “all operate independently”; one person owns “intake, scoping, redaction, review, release, and reporting” while already doing payroll and council minutes (Michigan Advance / CivicPlus, 2026). They buy because a missed deadline or a missed responsive email is a fee-shifting lawsuit and a front-page story.
• Rough TAM reasoning: ~19,000 municipal governments, ~13,000 school districts, ~38,000 special-purpose districts in the US. Conservatively 30,000+ are small enough to have a one-person records function and big enough to face recurring requests. At $250–600/mo that’s a $90M–$200M serviceable line even before counties and state sub-agencies.
• Why now for them: Request volume up 10–40% YoY at exactly these agencies; the clerk who used to absorb it on evenings is now refusing or burning out; AI-FOIA is no longer a career-risk word to say out loud after 2026 congressional testimony.

5. Product sketch (MVP)

• Connect read-only to the agency’s mailbox (Microsoft 365 / Google Workspace) and a shared drive; no migration, no RMS replacement.
• Paste the plain-English request; ClerkLane returns a ranked list of candidate responsive documents with a one-line “why this matched” for each.
• Custodian triages with two buttons (Responsive / Not) — the model re-ranks live as they correct it.
• Auto-drafted exemption log: for each withheld/redacted item, the suggested statutory exemption (state PRA/FOIA citation), with the clerk editing before it’s final.
• Defensible search trail: an auto-generated record of the custodians, date ranges, search terms, and systems searched — the document you hand a judge to show the search was “reasonably calculated.”
• Statutory clock per request with the agency’s state deadline rules baked in, plus a “you are at risk” escalation.
• Export the responsive set straight into whatever they already use to redact and release (or a built-in basic redact pass for SSNs/DOBs).

6. AI angle — what’s load-bearing

Remove the AI and there is no product — it would just be another ticketing portal, of which there are five. The load-bearing work is semantic responsiveness scoring against a vague natural-language request across a large, messy, multi-format corpus (threaded email, attachments, scanned PDFs, drive files) — the exact task that was infeasible before long-context models and is still infeasible with keyword search, which is why clerks read everything by hand. The secondary AI job is mapping each item to a defensible exemption citation given the agency’s state statute. The human stays in the loop by design: the model proposes, the clerk disposes, and the trail records both — which is what makes it legally usable rather than reckless.

7. Localization angle (if any)

N/A — this is a US-only play, and that is the wedge. Public-records law is 50 state statutes plus federal FOIA, each with its own exemptions, deadlines, and fee rules. The defensibility comes from encoding that state-by-state nuance correctly, which is the opposite of a generic global tool. International is not the expansion path; more US agency types (counties, state sub-agencies) is.

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

• Pricing: $300/mo base for small agencies (under ~25 staff mailboxes) on annual purchase order; $600/mo mid tier (school districts / multi-department). Add-on usage block for unusually large single requests.
• ACV: ~$4,800 blended (skewed by mid-tier school districts at $7,200).
• Rough math to $1M ARR: ~210 agencies at $4,800 ACV. That is a rounding error against 30,000+ eligible agencies.
• Rough math to $5M ARR: ~1,000 agencies, or ~700 plus a county/state-sub-agency tier at higher ACV. Achievable inside the small-agency segment alone without moving upmarket into enterprise procurement.
• Expansion path: start with the clerk’s office, expand to other departments in the same agency (police records, HR), then to neighboring agencies via the same state-association channel. ACV grows by department count and by request-volume tier.

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

• State municipal-clerk and school-board associations. Every state has one (e.g., state Municipal Clerks Association, state School Boards Association) with conferences, listservs, and certification programs. Sponsor one state’s clerk conference, run a live “watch it scope a real request in 4 minutes” demo. These associations are tight referral networks — clerks call each other.
• The lawsuit/news list. Public-records denial and delay lawsuits are public. Scrape the last 12 months of state-court PRA/FOIA suits and local news of agencies that blew a deadline; that agency’s clerk is a warm lead who just lived the pain. ~Send a one-page “here’s what would have caught this” with a 3-minute Loom.
• Pilot-to-PO motion. Offer a free scope of one real backlogged request the clerk picks. They see it find an email they’d have missed. Convert on a single-line annual PO under the agency’s no-bid threshold (commonly $5–10K) — no RFP, no 6-month procurement.
• MRSC / state research-council and records-management trainers already teach “responding to public records requests” workshops; co-present a session. They are the trusted voice the target customer already listens to.

10. Build complexity — justification

Medium. Off-the-shelf: M365/Google read-only Graph/Workspace APIs, document parsing, long-context LLM for responsiveness and exemption mapping, standard web stack. Custom work: a reliable ingest pipeline over messy real mailboxes/drives, the human-in-the-loop triage+re-rank UX, and a correctly-encoded state statute/deadline/exemption ruleset (starts with 3–5 states, expands). No novel models, no hardware. A focused pair ships a credible v1 for 3–5 launch states in ~3–4 months; the long pole is statute encoding and trustworthy ingest, not ML.

11. Gating checklist

Gate	Pass?	Note
Legal in target market	✅	Tool assists a lawful disclosure duty; human signs off; read-only access with agency consent.
Ethical — no harm / dark patterns	✅	Increases transparency compliance; human-in-loop prevents wrongful withholding.
Market exists (evidence above)	✅	Rising volumes, fee-shifting liability, incumbent ticketing spend already in budgets.
1–5 person team can build this	✅	Medium build, no research breakthrough, off-the-shelf APIs.
Launchable with <$50K / ₹40L	✅	Inference + dev + one conference sponsorship.

All five pass.

12. Feasibility score

Axis	Weight	Score	Notes
Problem intensity	20	16/20	Hair-on-fire near a deadline or after a lawsuit; weekly+ pain for the custodian, but the buying office often tolerates the status quo until something breaks.
Demand evidence	15	12/15	Multiple independent signals: volume growth numbers, fee-shifting liability, existing ticketing/redaction spend. No public competitor revenue for this exact step — both a gap and an unproven-willingness risk.
Build feasibility	15	11/15	No ML novelty, but trustworthy ingest over messy real systems + multi-state statute encoding is real engineering discipline, ~3–4 months.
Distribution clarity	15	11/15	Named channels (state clerk/school-board associations, lawsuit lists, MRSC trainers) with a clean pilot-to-PO motion; gov sales cycles still add friction even under no-bid thresholds.
Revenue mechanics	15	12/15	Pricing benchmarked to incumbent ticketing budgets; $1M ARR needs only ~210 of 30,000+ agencies. Annual PO timing is the soft spot.
Time to first revenue	10	7/10	Pilot can close on a single-line PO in weeks where budget exists, but gov budget cycles can push some deals to the next fiscal year.
Defensibility	10	5/10	Moat is encoded multi-state statute nuance + association trust + workflow lock-in via the audit trail of record. Copyable by a funded GovTech incumbent in ~12 months; head start and focus are the defense.
Total	100	74/100

13. Qualitative modifiers

Founder-fit tags

technical-heavy · domain-expertise-required — needs solid engineering for reliable corpus ingest and a real public-records-law advisor (a records officer or government-transparency attorney) to encode statutes credibly.

Key assumptions to validate (3–5)

1. Assumption: Small-agency custodians will trust an AI-proposed responsive set enough to use it, given the legal stakes. How to test: Run 10 real backlogged requests for 10 clerks; measure whether they accept the ranked set as a starting point and whether it surfaces a document they’d have missed.
2. Assumption: A single-line annual PO under the no-bid threshold avoids RFP procurement. How to test: Confirm no-bid thresholds and purchasing process with 15 target clerks across 5 states before building pricing.
3. Assumption: Responsiveness scoring is accurate enough on messy real mailboxes (not curated demos) to be defensible. How to test: Blind benchmark against a clerk’s own hand-determined responsive set on 20 historical requests; target high recall (missing a responsive record is the fatal error).
4. Assumption: State clerk/school-board associations will give a startup a demo slot. How to test: Pitch 5 state associations for a 2026 conference session; need 2+ yeses.

Risk flags

1. Liability / accuracy risk: A missed responsive record can cause a wrongful-withholding lawsuit. Product must optimize recall and be framed as decision-support with a human signature, never an autodecider. This is the dominant risk and shapes the whole design.
2. Platform dependency: Relies on M365/Google read-only access; tenant admin policies or API changes can block ingest. Mitigate with multiple connector paths and an export-bundle fallback.
3. Procurement timing: Government budget cycles can stretch time-to-revenue past the bootstrapper-friendly window even when intent is high.
4. Incumbent response: Granicus/CivicPlus could bolt a “find responsive records” feature onto existing ticketing seats within ~12 months; defense is depth of statute encoding and the small-agency segment they under-serve.

14. Structured verdict

Score:                  74/100
Verdict:                GO
Confidence:             Medium
Best-fit builder:       Technical founder + a public-records-law/clerk domain advisor
Time to revenue:        8–14 weeks where budget exists; longer across fiscal-year boundaries
Capital to launch:      $15–35K (inference, dev time, one state conference sponsorship)
Top 3 assumptions to validate first:
  1. Clerks trust and act on the AI-proposed responsive set — 10 real-request pilots, measure missed-doc catches
  2. Single-line annual PO clears no-bid thresholds — confirm purchasing process with 15 clerks across 5 states
  3. High recall on messy real mailboxes — blind benchmark vs. 20 hand-determined historical responsive sets
Kill criteria:
  - Abandon if <30% of 20 piloting clerks say the ranked set is a usable starting point they'd pay for
  - Abandon if recall against hand-labeled historical sets is below the level a transparency attorney will defend
  - Abandon if a funded incumbent ships an equivalent responsive-search feature inside ticketing before your v1 launches

15. Next step — 1-week validation sprint

• Day 1–2: Pull 12 months of state-court PRA/FOIA delay/denial suits and local-news “agency missed deadline” stories in 3 target states. Build a list of 40 clerks who recently lived the pain.
• Day 3–4: Cold-call/email 20 of them offering to scope one real backlogged request for free. Get 6 to hand over an anonymized historical request + the documents they eventually deemed responsive.
• Day 5: Run responsiveness scoring against those 6 real corpora; compare to the clerk’s own hand-determined set. Go/no-go on a falsifiable bar: in ≥4 of 6, the tool must surface every record the clerk deemed responsive (recall = 100% on that set) and the clerk must say they would have used it. Anything less means accuracy isn’t defensible yet — fix or kill.