An AI operating layer for solo & small plaintiff personal-injury firms — what it takes to build it, validate it, and get customers, with the honest math.
Prepared July 1, 2026 · Blue Bridge Logic · researched & costed
1 · The verdict, up front
Run it as paid R&D that must earn the right to be a product.
This is a real opportunity with a genuinely open seam — but it is not a raise-money startup today, and treating it like one is the fastest way to lose it. The winning shape: a 6-month, 3-firm, customer-funded sprint. Your design partner pays for the build. Two referral firms validate (or kill) the pattern. Hard gates decide — on evidence, not vibes — whether Erin becomes a product, stays a high-margin boutique consulting line, or stops.
The research supports moving now: margins are extraordinary (~$40–75/firm/month in infra against a $1,500–2,500 price), a compliant AI path exists off the shelf, year-one setup is under ~$11k, and the funded giants are visibly moving upmarket, leaving the 1–5-person firm unserved. The constraint isn't money or tech — it's your hours, and whether the daily-habit wedge actually sticks at firm #1.
~$40–75
marginal infra cost per firm / month
2–4 wks
to a working version in a paying user's hands
~$7–11k
year-one compliance + setup budget (likely case)
3 firms
cap for the 6-month validation sprint
Why this framing and not "launch a SaaS": your own adversarial review flagged that the 1–5-person firm segment might be a poverty trap — low budgets, messy data, high support needs. The counter-evidence is real (one firm already pays ~$1.5k/month happily; legal-tech churn is the lowest in SaaS) but thin (N=1). Paid R&D lets you buy the answer with customer money instead of your savings. The downside scenario isn't zero — it's a healthy consulting boutique with reusable assets and a trust network.
2 · The opportunity
The buyer
Solo to ~5-person plaintiff-side personal-injury firms on Clio: 50–250 active matters, owner is the bottleneck, tech-curious but not technical, referral-driven. The defining pain isn't clock time — it's mental load: "always scanning what needs to be done and not feeling confident about what you need to do next." The dream they articulate unprompted: their accumulated litigation knowledge working for them like an experienced partner.
Market size signal: Clio's 2025 Solo & Small Firm report found 72% of solos already use AI somewhere — but only 8% use it widely. That's a large aware-but-not-operationalized pool, and "operationalize it for me" is exactly Erin's pitch.
The competitive picture (July 2026)
Player
Position
What it means for Erin
Eve ($1B val, ~$164M raised)
"AI-native OS for plaintiff firms," now claims 1,200+ firms; per-seat, quote-only, multi-seat platform adoption
Owns the concept at mid-market. Went 100→450 firms in 8 months — proof the PI bar moves in packs once references exist
EvenUp ($2B val, ~$370M)
Pivoted to "Claims Intelligence Platform"; per-case; favors volume firms
Demand letters fully commoditized — never build the moat there
Supio (~$91M)
ARR 4× since Series A; Thomson Reuters partnership
Moving upmarket with TR distribution
Clio ($5B)
Manage AI + vLex; generalist, not PI-process-deep
Both the substrate and the "why not wait for Clio?" objection
Point tools
Precedent $275/demand, Briefpoint, Smith.ai
Transparent pricing works on solos — copy that posture
The seam
Investors now describe the plaintiff-side AI market (~$682M absorbed) as crowded and are eyeing defense-side — and every funded player is moving upmarket. Nobody verified is selling an operating layer priced per-firm for 1–5-person Clio shops. Quote-only pricing repels solos; Erin publishing a flat price is itself a differentiator. And the least-served, most-defensible job remains the firm's own knowledge: nobody does "push this case through our playbook" for small firms.
Why you, why now
You already have the wedge asset: a paying design partner, a working relationship, his Clio rebuilt by you, a live interactive demo, and two warm referrals from him — in a market where "who else has this?" is the buying question. ABA's 2025 survey: 43% choose AI tools for integration with software they already trust; 33% for the vendor understanding their workflows. Both are literally your story.
Timing: LawPro.ai's Feb 2026 report (300+ PI firms) says adoption has shifted "from experimentation to execution" — buyers now choose on accuracy, transparency, and workflow integration, not novelty.
The honest caveats: the wedge rests on one firm; the "wait for Clio's AI" objection is real; and Eve's Auditor already does nightly caseload scanning inside its own platform. Erin wins only by being radically simpler, transparently priced, and on top of the Clio they already have.
3 · What you're selling
Not "AI." An operating layer with one promise: the repetitive operational work of your practice runs itself; you get told what needs you. Modules map 1:1 to observed jobs (from the JTBD framework):
Phase
Modules
Status
V1 — "Win the morning"
Clio Foundation (fields + doc gen) · Daily Triage ("8 of your 120 need you") · Records & Providers tracker with automated collection follow-up
Foundation live · triage is the committed next deliverable
Knowledge funnel: the firm's 13 years → cited, case-specific strategy memos
Prototype, never promise
Architecture decisions already made (and validated by this research):
Single-tenant per firm — in a privilege-sensitive domain, isolation is a feature you sell. It's also the direct answer to the #1 vendor-vetting question ("does my data mingle?").
Sit on Clio, never replace it. Design around the hard constraint: Clio's API allows 3 requests/second per app shared across ALL your customers — nightly scans must be a slow-drip queue against a local cache from day one.
PHI-touching AI runs on AWS Bedrock, not the first-party API and not Cloudflare storage (see §9 — this is the single biggest architectural finding of the research).
Assistive-only; Erin is never the deadline system of record. No "we catch everything" promise — that's uninsurable. Coverage boundaries shown in-product.
Integrity rule
Sell what's shipped. The knowledge funnel is the vision you share honestly as roadmap — the paid promise is only ever what's live. Anchor the buyer on triage-relief today, not memos tomorrow.
4 · Time to a working version in users' hands
The unusual answer: 2–4 weeks — because a paying user already exists and the next deliverable is already agreed. "Launch" isn't a cliff; it's hardening a live engagement into a product surface.
Weeks 1–2
Triage digest v0 — live at firm #1
The daily case-summary scan (already the committed deliverable): nightly Clio read → ranked "what needs you" digest by email/text. Scripted and manual-ish is fine. This is the moment Erin exists — a real user starts their morning on it.
Weeks 3–6
Product surface v1
The Matters app (already designed and deployed as a mockup) backed by real Clio data for firm #1: triage list, records/providers tracker, collection follow-ups, doc generation. Audit trail + one-click override everywhere. Start measuring the habit (digest opens, actions taken).
Weeks 7–10
Firm #2 — the convergence test begins
Onboard the first referral firm by configuration: their field schema, their letter templates, their knowledge folder — same engines. Track % net-new code honestly. Target: live in ≤2 weeks from signed order.
Weeks 11–16
Firm #3 + knowledge-funnel prototype
Third firm onboards config-only. In parallel: index firm #1's corpus and produce the first cited strategy memos, watermarked as draft leverage — prototype, not promise.
Month 6
The gate decision
Three firms paying, metrics in hand → productize harder, run as consulting line, or stop (see §7).
Capacity honesty
This timeline assumes ~25–30 focused hours/week on Erin. If Blue Bridge horizontal work eats the calendar, multiply by ~2×. The single biggest schedule risk isn't engineering — it's support interruptions: a tech-wary attorney calls, he doesn't file tickets. Budget 4–8 hrs/firm/month of high-touch support in the early phase and treat it as the product's real COGS.
The real cost is you. Infra + compliance are noise next to founder time: build hours, onboarding + per-firm data cleanup (messy Dropboxes are real labor every time), support interruptions, and being effectively on-call for a "did-I-miss-something" product. Price for the support load, not the server bill.
6 · Revenue & unit economics
Pricing (locked by ethics + strategy)
One-time foundation/setup: $2,500–$7,500 (scales with Clio mess + corpus size) + flat monthly platform fee: $1,500–$2,500, tiered by active caseload. This is the shape firm #1 already accepted.
Publish the price. Every funded rival is quote-only; transparency is a weapon with this buyer.
Never % of settlement or anything keyed to the lawyer's fee — a Rule 5.4 fee-sharing violation the lawyer can't sign (NYC Bar Op. 2018-5). Never market "bill your clients for it" (ABA 512 forbids billing for time AI saved).
Monthly, or annual with a 90-day out, for the first ten customers — removes the biggest objection a risk-trained buyer has.
The math at each stage
Scale
ARR (at $1.5–2.5k/mo)
+ Setup fees
Infra + fixed
Support hrs/mo
Read
3 firms (mo. 6)
$54–90k
$7.5–22k
~$2.5k/yr
12–24
Validation sprint; comfortably self-funding
10 firms (yr. 1–2)
$180–300k
$25–75k
~$6–7k/yr
40–80
Real business; support becomes the job
15–20 firms
$270–600k
—
~$10k/yr
60–160
The solo ceiling. Past here: hire support, raise, or hold
Gross margin on infrastructure runs ~97%. Loaded honestly with support time (at even $100/hr opportunity cost), a firm at $1,500/mo with 6 support hrs/mo still yields ~55–60% true margin early — improving as the product absorbs the support surface. LTV math: legal-tech churn norms (~6.5%/year) imply a retained firm is worth $50k+ over five years. At founder-led CAC (~$0 cash), the economics are excellent if retention lands at legal-tech norms — which is exactly what the validation plan measures.
7 · The validation plan — gates, not vibes
0
Design-partner truth Cleared
A firm pays real money for the underlying work and refers peers unprompted.
1
Pattern replication — weeks, costs a few calls
Run the JTBD interviews with both referral attorneys. Pass: ≥4 of 7 hypotheses hold (same core jobs, same "get my head back" driver, knowledge-bank problem, Clio, referral behavior, comparable team shape).
Both diverge → the "vertical" is one man's workflow. Keep serving firm #1 profitably; do not build multi-firm.
2
The habit — by week 8 at firm #1
Digest opened ≥5 days/week for 4 consecutive weeks; flagged items acted on; the "8–10 minutes finding my last thought" self-report drops; zero missed-critical incidents. (Pilot norms: ≥70% weekly engagement over a ~3-month pilot.)
Habit doesn't stick with the friendliest possible user → the wedge is wrong. Re-scope with him before selling anyone else. Do not onboard firm #2 on a broken wedge.
3
Convergence — firms #2 and #3
Each onboards with ≤20% net-new code, identical triage/knowledge engines byte-for-byte, live within 2 weeks, at full price on the standard agreement.
Firm #3 still needs bespoke engine work → Erin is honestly a consulting line. That's a fine business — price it as one and stop calling it a product.
4
Retention + the channel — months 6–12
Zero logo churn at 3 months; ≥1 customer sourced by customer referral (not your network); tracking toward legal-tech's ~93.5% annual retention; NPS ≥36. This is the profile that reads as vertical-SaaS PMF.
Churn or silence → the poverty-trap thesis was right. Fall back to boutique mode with the assets and relationships intact.
Design-partner mechanics (evidence-backed)
Cap at 3 concurrent firms — more than 3–5 and a solo can't service them. Charge them (paying partners convert; Sierra converted 100% of paid design partners). Contract the exit: feedback cadence, named-reference & logo rights, founding-firm discount, and the milestone where pilot pricing converts to standard. Common Paper publishes a free design-partner agreement.
Work the two warm referrals by hand. The pitch is one sentence: "‹Firm #1› hired me to fix his Clio and build his AI system; he told me to call you." Demo with the live product on firm #1's (sanitized) reality.
Make firm #1 the showcase. Formalize the reference/referral arrangement in the design-partner agreement. His unprompted advocacy in PI Facebook groups and masterminds — where vendor self-promo is banned but member recommendations are the currency — is worth more than any booth.
Concede weaknesses early. Lawyers are trained hole-pokers; the appellate move (concede, then show why it doesn't matter) builds the trust that closes. Precision everywhere — one sloppy term of art costs the deal.
Best cost/density anywhere; local; repeat exposure. Speaking > exhibiting — pitch "AI competence for PI practice" CLE talks
Clio App Directory listing → Certified App track
$0
Table-stakes credibility ("integrates with what I trust" = the 43% factor); Certified tier gets Clio reps recommending you
Podcast guesting — Personal Injury Mastermind, Lawyerist, Legal Talk Network
$0
Founder-story episodes are earned, not bought; Lawyerist's audience: 80% are their firm's tech buyer
ABA TECHSHOW Startup Alley (apply ~Nov)
$1,800
Cheapest credible national stage + press coverage (LawSites)
Phase 3 (12+ months, only after Gate 4): scale spend
AAJ Annual booth (~$4,200) and ClioCon sponsorship once ~10 customers can staff the story. Skip MTMP ($7.5–8.5k booths, mass-tort crowd) — wrong ICP, maybe forever. Every dollar of paid channel before retention proof is burned.
9 · Legal & compliance playbook
The three findings that change the architecture
PI plaintiff firms are generally NOT HIPAA business associates — records arrive via the client's own authorization, and HIPAA doesn't follow the data downstream (HHS FAQ 705). What actually binds Erin: contract terms implementing the firm's confidentiality duties (Rules 1.6/5.3), records-custodian and protective-order terms, and state law — especially Washington's My Health My Data Act, which applies directly to processors and carries a private right of action. But the market demands BAAs anyway (Clio tells firms to require them): sign them willingly, run a HIPAA-aligned program, and never say "HIPAA certified" (no such thing).
The compliant AI path is AWS Bedrock. Anthropic's first-party HIPAA-ready org exists but is sales-gated and excludes the Batch API — doubling scan costs. Bedrock: Claude on the AWS HIPAA-eligible list, self-serve BAA via AWS Artifact, price parity, batch −50% available, embeddings under the same BAA. (Disable Bedrock invocation logging or PHI lands in CloudWatch.)
Cloudflare's BAA is Enterprise-only (~$60k/yr) and its developer platform isn't confirmed BAA-covered at any price. Keep Cloudflare as compute/routing; PHI at rest lives on AWS.
Rules that shape the business
Pricing: subscription / per-matter / flat only. Anything keyed to the lawyer's fee or recovery is a Rule 5.4 violation the buyer cannot sign.
Marketing claims: FTC "Operation AI Comply" fined DoNotPay $193k over "robot lawyer" claims. Keep a substantiation file for every quantitative claim; frame outcomes as what better preparation makes possible, attributed to the lawyer. (Virginia LEO 1901 is the value story: firms keep the same contingent fee when AI makes them faster — the efficiency gain is theirs.)
UPL: stay in the Westlaw posture — sell to firms only, output is drafts/flags for attorney review, no vendor-to-claimant communication, never "AI lawyer/paralegal" branding.
SOC 2: defer. Solo/small firms vet with questionnaires, not audits; even Clio's App Directory wants MVSP hygiene, not SOC 2. Ship a security page + CAIQ self-assessment + pen-test letter now; buy SOC 2 (~$12–25k) only when a specific deal commits in writing.
Tax: charge the Nevada firm no sales tax (destination sourcing; NV doesn't tax SaaS). WA B&O ≈1.5% on services. Get one accountant pass on ESSB 5814 (WA's Oct-2025 sales-tax expansion to IT services, under litigation) before invoicing WA customers.
Entity: WA LLC now (~$500 all-in); Delaware flip (~$3–7k) only if a priced round ever happens. Contracts: Common Paper's free CSA + DPA + BAA + AI Addendum with ~$1k of lawyer review.
Turn the buyer's checklist into your sales kit
Bar guidance (ABA 512, TX Op. 705, NC 2024 FEO 1, WSBA 202505) tells lawyers exactly what to demand from AI vendors. Pre-answer all of it in a one-page vendor due-diligence record you hand every prospect: contractual no-training with written flow-through to the model provider, retention/deletion terms, single-tenant isolation down to the RAG layer, subpoena-notice clause, 72-hour breach notice, subprocessor list — plus a template firm AI policy and a protective-order compliance letter (2025–26 federal POs now require AI tools touching discovery to prohibit training and enable deletion). Nevada has no bar AI opinion yet, so this over-prepares you for your first market. Doing the lawyer's homework for them is a closing asset.
10 · The bear case — what to internalize before betting
N=1. Everything rests on one firm plus two untested referrals. Gate 1 exists precisely because this could be one man's workflow, not a market.
The poverty trap. "Funded players ignore the 1–5-person firm" can mean opportunity — or that the segment can't support software economics: low budgets, messy Dropboxes, high-touch support. The mitigations are the convergence kill-test and pricing for support; the risk doesn't fully retire until Gate 4.
"I'll wait for Clio." The incumbent doesn't have to come downmarket to hurt you; the buyer just has to believe their existing platform will do it. Your answer is speed, the trust channel, and PI-depth Clio won't prioritize — and honestly not selling prospects for whom the answer is weak.
Platform dependency. Erin lives on Clio's API (3 req/s, no increases, terms can change) while Clio's own AI moves up-stack. Adapter architecture + the knowledge layer living outside Clio is the hedge, not a cure.
Reliance risk is the product risk. Once an attorney runs his morning on Erin's digest, whatever Erin misses is something a human stopped checking — and in PI the misses (SOL, deadlines) are malpractice-adjacent. Never be the deadline system of record; put coverage boundaries in the contract and the UI; carry AI-affirmative E&O; treat any missed-surface as sev-1.
Solo-founder physics. Build + sell + support + comply, alone, for a buyer who works 65-hour weeks and calls instead of filing tickets. The 3-firm cap and buy-don't-build for commodity edges (records OCR, demand generation, intake voice) are load management, not strategy garnish.
The moat isn't the feature. Indexing a Dropbox is replicable. What compounds: the accumulated per-firm knowledge base (switching cost), the referral-graph distribution, and the daily habit. The funnel is a prototype to prove, not a moat to bank on.
Concentration. Firm #1 is simultaneously your revenue, your reference, and your channel. Keep the relationship formal (design-partner agreement), keep delivering the boring stuff excellently, and diversify the moment Gate 3 clears.
The downside is bounded. If every product gate fails, you exit with: a paying consulting line at excellent margins, reusable delivery assets, deep vertical knowledge, and a referral network. That's not zero — it's the reason paid R&D beats betting savings on a SaaS launch.
11 · The first 90 days
Weeks 1–2 — foundation + first live value
Ship triage digest v0 to firm #1 (the committed deliverable) — instrument opens/actions from day one
File WA LLC + business license + EIN (~$500); open business banking; Stripe with ACH default
Sign the AWS BAA in AWS Artifact (self-serve); stand up Bedrock access; decide PHI-on-AWS boundary
Get E&O + cyber quotes (Vouch, Coalition broker) — ask the five AI-exclusion questions in writing
Weeks 3–6 — validate the pattern, formalize the partner
Run the JTBD interviews with both referrals (Gate 1) — before building anything multi-firm
Sign the design-partner agreement with firm #1 (Common Paper template): reference rights, feedback cadence, founding pricing, conversion milestone
Stand up the contract stack: CSA + DPA + BAA + AI addendum, ~$1k lawyer review
HIPAA-aligned program: HHS SRA tool + written policies + training (~$2–3.5k path)
Build the sales kit: one-page due-diligence record, security page, template firm AI policy, protective-order letter
Harden the product surface at firm #1; track habit metrics weekly (Gate 2 opens)
Weeks 7–13 — the convergence test
If Gates 1–2 pass: close firm #2 at full price on the standard agreement; onboard by config; log net-new code % (Gate 3 opens)
Apply to Clio App Directory (free) — start the security-review clock early
Book a WSAJ seminar table or CLE speaking slot (convention is Sep 30–Oct 2 — decide by early August)
Pitch one podcast (Personal Injury Mastermind founder-story angle)
Start the knowledge-funnel prototype on firm #1's corpus — watermarked drafts, never promised
Then hold the line: no firm #4, no paid marketing, no SOC 2, no fundraising conversations until the Month-6 gate review says the pattern converges. The discipline is the plan.
Sources & provenance. Synthesized from the Erin product PRD, competitive research, and JTBD framework (June 2026, Blue Bridge vault) plus three fresh research passes (June 30–July 1, 2026) on operating costs, GTM/validation benchmarks, and compliance/insurance — full citations in research/erin-business-plan-research.md. Key figures verified against primary sources; estimates and unverified items are flagged there. Claude pricing per Anthropic (Opus 4.8 $5/$25, Sonnet 4.6 $3/$15, Haiku 4.5 $1/$5 per MTok; Batch −50%). Insurance and compliance-platform figures are quote-pool medians, not bound quotes.
Confidential — Blue Bridge Logic internal. Not legal, tax, or insurance advice; the flagged items (WA MHMDA posture, ESSB 5814, BAA scope) deserve one pass with counsel and an accountant before firm #2 signs.