Profession Packs: Official Forms by Field

Trackmint research doc 04 — July 18, 2026

1. The pattern, and why packs matter

Trackmint’s core loop is simple: the professional tracks work items (issues, matters, cases), logs time against them, and bills the client from that same record. On top of that sits the AI module: scan the client’s source documents (W-2s, pay slips, tax returns, bank statements) and auto-fill the official forms the profession lives on.

This is the playbook Glade AI proved in consumer bankruptcy. A debtor hands over six months of pay stubs, tax returns, bank statements, and a credit report. Glade’s AI extracts the data, fills the official petition package (Forms B101, B106 schedules, B107, the B122A-2 means test, the creditor matrix), validates it, and e-files through CM/ECF (glade.ai/bankruptcy). Firms like Van Horn Law Group report 16+ filings per week. The insight is that bankruptcy is not special — it is just the first vertical where three conditions line up:

1. Official, standardized forms. Courts or agencies define the output. The target schema is fixed and public, so the AI has a well-defined destination for every extracted field.
2. Predictable source documents. Clients always bring the same kinds of paperwork — pay stubs, tax returns, bank statements, IDs, deeds, medical records. Extraction models can be trained once and reused.
3. Painful manual re-keying today. Incumbent software is questionnaire- or template-driven: a human still reads each document and types the numbers in.

Wherever those three conditions hold, the same engine — document intake, OCR/extraction, conflict detection, form generation, filing — can be repackaged for a new profession. That repackaging is what we call a profession pack: a bundle of (a) form templates and filing logic for that field, (b) document-type extractors tuned to its source docs, (c) intake checklists and client-portal flows, and (d) the matter/board/billing defaults that make Trackmint feel native to that profession on day one.

Packs matter for three reasons:

• Reuse. The extraction layer is largely shared. Pay stubs, W-2s, tax returns, and bank statements appear as inputs in bankruptcy, family law, immigration (I-864), SSDI, mortgage, and SBA lending. Each new pack that reuses existing extractors is mostly “new output templates,” not new AI.
• Go-to-market focus. “Case management with AI” is generic. “Fills your FL-150 from the client’s pay stubs” sells itself to a family-law attorney in one sentence.
• Defensibility against horizontal incumbents. PracticePanther, Clio, and the AMS/HRIS/ERP incumbents can’t credibly build fourteen vertical form engines. A pack roadmap lets Trackmint out-vertical them one profession at a time.

One important asymmetry, covered in section 5: not every profession has official forms. Software developers and contractors — Barry’s other core persona — have none. Their pack is not a forms pack at all; it is the “Jira × QuickBooks” crossbreed (issues + hours + expenses + invoices). That is fine, and in fact useful: it defines the floor of what every pack contains before any forms are added.

2. The profession map (14 fields)

Compiled July 2026. “AI?” refers to whether incumbents do AI extraction from source docs, not AI in general.

Reading the table, the “fit” logic is consistent: skip anywhere AI extraction is already the incumbent’s product (tax prep, bookkeeping, medical billing), be cautious where a funded AI player is consolidating the exact niche (title/Qualia, enterprise mortgage), and lean in where forms are standardized, source docs match our existing extractors, and incumbents are still questionnaire- or template-driven.

3. Additional discovered verticals

Beyond the core 14, six more verticals surfaced that fit the scan-to-form pattern:

• Family law financial disclosures — STRONG. California’s FL-150 Income & Expense Declaration plus FL-142 (Schedule of Assets and Debts); every state has analogs (e.g., New York’s Statement of Net Worth). The source documents — pay stubs, tax returns, bank statements — are identical to bankruptcy inputs, so the existing extraction engine carries over almost unchanged. Existing software (Gavel, Clio forms) is template-fill only. ~670K divorces per year, and no AI-extraction incumbent found.
• Veterans disability — STRONG, with regulatory landmines. VA Forms 21-526EZ and 21-0995/0996, filled from DD-214s and service treatment records. Tens of thousands of VA-accredited reps. Fee rules under 38 USC 5904 constrain how reps charge, so the play is selling software to the reps, not touching the claimant fee.
• SBA lending (sell to lenders). SBA Forms 1919, 413, and 1920, generated from tax returns and bank statements. Casca ($29M Series A) is already here. The 7(a) program alone is ~$31B+/yr.
• Customs brokerage. CBP Forms 7501/3461 and ISF filings from commercial invoices and bills of lading. DocUnlock (a16z-backed) and KlearNow are contesting the space.
• Financial advisor onboarding. ACAT transfer paperwork filled from brokerage statements. Powder ($5M seed) is early; ~300K+ advisors in the market.
• Medicaid / long-term-care planning — whitespace. State LTC applications require up to 60 months of bank statements — a brutal manual review with no dominant AI extractor. The catch is 50-state fragmentation of forms and rules.

Explicitly examined and skipped: workers’ comp (AI is insurer-side), guardianship accountings (tiny market), evictions (fragmented, reputational risk), and FAFSA (free by design).

4. Ranked Top 5 profession packs

Ranking criteria: extractor reuse from the bankruptcy-style engine, form standardization, absence of a funded AI incumbent, market volume, and clean per-case pricing.

1. Family law / divorce financial disclosures (FL-150 + state analogs). The cheapest possible reuse of the engine. The inputs are the same pay stubs, tax returns, and bank statements the bankruptcy pipeline already parses — only the output template changes. Forms are court-standardized per state (start with California’s FL-150/FL-142, expand state by state). Volume is large and recurring: ~670K divorces per year plus support-modification filings, each requiring updated disclosures. Incumbents (Gavel, Clio’s form tools) are questionnaire template-fill with no extraction, and no funded AI competitor was found. This is the fastest path from “bankruptcy engine” to “second revenue line.”

2. SSDI / disability representatives (SSA-3368/3369/3373 + appeals). Federal forms, standardized nationwide — one template set covers the whole country, unlike state-fragmented verticals. ~1.94M SSDI applications in 2024. The PI-AI wave (EvenUp, Supio, Eve) went after demand letters and skipped form-fill entirely; the SSDI-specific incumbent (Prevail) has zero AI extraction. The rep bar is small (~4K NOSSCR firms) and concentrated, which makes it reachable, and per-case pricing maps cleanly onto how reps get paid. Adjacent expansion into VA disability claims (section 3) shares the disability-evidence extraction work.

3. Probate, starting California (DE-111, DE-160/161 inventory, plus IRS 706/1041). ~2.6M probate cases per year and $18.2B in estate-law revenue, yet the AI money in this space (Wealth.com, Vanilla) points the opposite direction — summarizing documents for planning, not filling court forms for administration. Scanning bank/brokerage statements, deeds, and insurance policies to build the DE-160 inventory and the 706/1041 tax forms is greenfield. Starting with California’s standardized statewide Judicial Council forms (~300K cases/yr) sidesteps national fragmentation while proving the pack.

4. Independent insurance agencies (ACORD forms + carrier supplementals). ACORD 125/126/127/130/140 and the ACORD 25 COI are nationally standardized — the dream target schema. Re-keying from expiring policies, loss runs, and SOVs is an acute, daily pain across 39,000 agencies. The funded AI wave (Bevaya, Kalepa, Cytora) sells to carriers; the agency side is underserved, and the AMS incumbents (Applied Epic, AMS360) still do manual intake. Applied’s acquisition spree validates both demand and an exit path. Main risk: those incumbents build it natively — which argues for moving before they do.

5. Immigration — financial/civil document pack (I-864 and friends). A wedge, not a full suite. Docketwise’s AI (“IQ”) extracts identity documents only; nobody at scale extracts tax returns, W-2s, and pay stubs into the I-864 Affidavit of Support, or handles foreign-language civil certificates. With 18K+ AILA members and ~10M+ USCIS filings a year, even the financial-doc slice is a real market — and it reuses, again, the exact bankruptcy-input extractors. Glade itself lists immigration as a secondary practice area, so speed matters here.

Honorable mentions: small-broker scan-to-1003 mortgage, construction WH-347 certified payroll + COI tracking for small subs, and Medicaid LTC. Deprioritized: tax prep, bookkeeping, medical billing, PI demand letters, HR/I-9, FAFSA — commoditized, crowded, or structurally unattractive.

5. The Software Development pack: the “no-forms” reference pack

Row 14 of the table is deliberately different. Freelance developers, contractors, and small dev agencies have no official forms. Their paperwork is SOWs, contracts, invoices, and W-9/1099s — and the 1099 burden sits with the payer’s accounting stack, not the contractor. There is no government-defined output schema to auto-fill, so the scan-source-docs pattern simply does not apply. Bonsai already dominates the contract/SOW template niche, and templates are not where Trackmint wins anyway.

Instead, the Software Dev pack is the Jira × QuickBooks crossbreed, built directly on Barry’s core persona insight: the work item you track is the same thing you bill.

• Issue tracking + Kanban + sprint/progress tracking (the Jira half): boards, backlogs, assignments, client-visible progress.
• Time logged per issue → fee → invoice (the QuickBooks half): hours land on the issue, roll up to the project budget, and flow straight into a client invoice — no export, no re-keying between a tracker and a billing tool.
• Contractor expense tracking: billable expenses (software licenses, cloud costs, travel) attach to the same project and appear on the same invoice. This overlaps with what freelancers currently use QuickBooks Self-Employed for; Trackmint doesn’t need to replace their tax accounting, just capture the client-billable expense flow at the point of work.

Today this persona is forced to stitch together Jira/Asana + Harvest/Toggl + an invoicing tool (or bolt Tempo onto Jira). The pack’s value is collapsing that stack into one object model.

Strategically, the Software Dev pack is the reference pack: it defines the universal core every pack contains — work items, boards, time, expenses, invoices — with the forms module count set to zero. Every other pack is “reference pack + a forms engine + profession-specific templates.” Notably, Glade has exactly the opposite shape: it has the forms engine but no time tracking and no hourly-billing invoicing (glade.ai/platform; gap confirmed in our Glade research notes). Lawyers on Glade still need something like PracticePanther for the billing half, and PracticePanther has no boards and no AI forms. Trackmint’s bet is that the union — tracked work + billed time + auto-filled forms — is the product neither side can easily build.

6. Recommended pack sequence

v1 — Software Dev pack + generic client billing (the core). Ship the reference pack first: issues/boards, per-issue time logging, expenses, and invoicing, plus a “generic professional services” skin (matters instead of issues, hourly rates, simple retainers) usable by any client-billing professional. Reasoning: it is the largest reachable audience (~60–70M US freelancers, self-serve, no compliance surface), it requires no forms engine or regulatory review, and it forces us to get the core object model — work item = billable item — right before anything is layered on it. Everything later depends on this spine.

v2 — First legal forms pack: Family Law, with Bankruptcy as the fallback. Add the AI document-scan → official-form engine, targeting family law financial disclosures (FL-150/FL-142, California first). Reasoning: it is the #1-ranked pack — same source-doc extractors as bankruptcy, standardized court forms, ~670K divorces/yr, and no AI incumbent, whereas bankruptcy means displacing Glade in its home vertical. Bankruptcy Ch 7/13 remains the well-mapped fallback (the forms, sources, and Glade’s feature set are fully documented) and a natural fast-follow: family-law and bankruptcy firms overlap heavily in the consumer-debtor bar, so the second forms module is cheap once the first ships. Either way, v2 is where the “Legal pack” from the persona doc becomes real: matters + trust-aware billing + AI forms in one tool — the combination neither Glade nor PracticePanther offers.

v3 — SSDI pack, then Probate-CA. SSDI next: federal standardized forms mean no per-state buildout, the ~4K-firm rep bar is cheap to reach, and no incumbent has AI extraction. Probate-CA follows, reusing the financial-statement extractors for DE-160 inventories and opening the $18.2B estate-administration market. Both extend the same extraction engine with modest new document types (medical/earnings records; death certs and deeds).

v4 and beyond — Insurance ACORD, Immigration I-864, then opportunistic verticals. ACORD is the biggest prize among the remaining packs but the first one whose source docs (loss runs, SOVs, policies) don’t overlap the consumer-financial extractor set, so it costs a real new extraction investment — do it once the engine and GTM are proven. The Immigration financial-doc wedge can slot in anywhere from v3 onward since it reuses existing extractors; timing depends on how fast Docketwise and Glade move. Behind those: veterans disability (with 38 USC 5904 fee-rule care), Medicaid LTC, construction WH-347, and small-broker 1003 as demand appears.

The through-line of the sequence: build the billing spine once (v1), build the forms engine once (v2), then add packs in order of extractor reuse and incumbent weakness (v3+). Each pack should be mostly templates and go-to-market — if a proposed pack requires a brand-new extraction stack and faces a funded AI incumbent, it belongs at the back of the queue.