Aperio.

Aperio

An AI mitigation-investigation copilot for U.S. criminal-defense teams

EvenUp's model, JusticeText's buyer, the mitigation specialist's framework: the records-to-mitigation engine for the defense.

~$8B/yr

Public-defense spend (buyer pool)

$6.5B state/local + ~$1.4B federal

Wiggins / Rompilla

Constitutional floor

Failure to investigate mitigation = leading ineffective-assistance-of-counsel ground

~$40M ARR

Year-5 capture (base)

Software-addressable SOM

scroll ↓

01The Problem/ 13

A constitutionally mandated duty that is chronically under-done

The duty-to-investigate gap

Strickland, Wiggins v. Smith (2003) and Rompilla v. Beard (2005) make failure to investigate mitigation a leading ground for reversal; ABA Guideline 4.1 mandates a mitigation specialist on every capital team
Adequate representation is labor-bound: RAND's 2023 standard is ~286 attorney-hours for an LWOP case, ~248 for murder, ~99 for a high-severity felony, work that is routinely skipped
Capacity, not demand, is the constraint: public defenders carry 300–350 felonies/lawyer in stressed counties, ~3x a sustainable load
The job is drudgery-heavy: collecting and triaging tens of thousands of pages of school / medical / mental-health / CPS / military / prison / court records, tagging to domains, gap-spotting, and first-draft assembly
Mitigation gets skipped not because it isn't required, but because there is no time, no trained staff, and no budget to do it

286 / 248 hrs

LWOP / murder case standard

RAND National Public Defense Workload Study, 2023

300–350 felonies/lawyer

Stressed-county caseload

vs. ~150 NAC max (St. Clair MO, Luzerne PA)

~21–27%

PD offices adequately staffed

directional; older BJS-era data

SourcesRAND RRA2559-1 (2023) workload hours. Caseload outliers: St. Clair County MO / Luzerne County PA. Staffing % is directional (JPI/BJS-era data, dated). Wiggins v. Smith 539 U.S. 510; Rompilla v. Beard 545 U.S. 374; ABA Guidelines 4.1 + 2008 Supplementary Guidelines.

02Why Now/ 13

Why now: three curves intersect

Capability: mitigation is long-context record synthesis, frontier LLMs with 200K–1M context windows now do exactly this work (collate thousands of pages, build sourced chronologies)
Adoption: 79–83% of lawyers report using generative AI in 2026; solo/small-firm adoption nearly doubled in one year (27%→53%)
Funding + reform: indigent-defense spend rose to ~$6.5B (from ~$2.3B at Gideon's 50th; note: nominal, ~1.9x real); RAND triggered caseload-limit reforms (e.g., WA Supreme Court 2025) and Arnold Ventures-backed reform
Trust gap creates the opening: leading legal-AI research tools hallucinate 17–34% of the time, unacceptable here, where a fabricated abuse fact is catastrophic and Strickland-exposed
EvenUp just proved the records-to-work-product model on the plaintiff side ($2B+ valuation in ~3 years), the defense side is unoccupied white space

79–83%

Lawyer gen-AI adoption (2026)

Bloomberg Law 83%; Clio 79%

27% → 53%

Small-firm adoption jump

one-year YoY (Smokeball)

17–34%

Legal-AI hallucination rate

Stanford RegLab, JELS 2025

SourcesAdoption: Bloomberg Law (760 practitioners, 2026), Clio Legal Trends 2025, Smokeball. Hallucination: Stanford RegLab (Lexis+ AI >17%, Westlaw ~34%), JELS 2025. Spend: Sixth Amendment Center, Gideon 60th.

03Solution/ 13

An agentic loop that builds the social history

From a disorganized records dump to evidence-grounded work product

1) PLAN: generates a records-collection plan across school, medical, mental-health, CPS, military, employment, prison, and court custodians, with request templates
2) INGEST & TAG: builds a sourced, Bates-linked chronology and tags every fact to ~10 mitigation domains and to ACEs / ABA frameworks
3) SCORE & FLAG: produces a coverage scorecard against recognized frameworks and a gap report ('no CPS or prenatal records yet')
4) DRAFT: assembles first-draft mitigation report / sentencing-memo work product, every assertion linked to a source page
Extractive-only and citation-locked: 'no source = no claim.' Confidence flags, full audit trail, human-in-the-loop verification by design

Records-to-domain auto-tagger + coverage scorecard

Wedge feature

The daily-pain killer; drafting is the upsell

Citation-locked, no-fabrication

Output posture

Bates-linked; descriptive, never predictive

SourcesFrameworks: ABA Guidelines + 2008 Supplementary Guidelines; Adverse Childhood Experiences (ACEs). Descriptive (organizes/sources evidence), not predictive (no dangerousness scores), outputs are discoverable and cross-examinable.

04Product/ 13

Domain depth horizontal tools won't build

Proprietary mitigation-domain ontology mapped to ABA Guidelines, Supplementary Guidelines, and ACEs, the scoring rubric is the product
~10 mitigation domains with coverage-scoring and gap-flagging that no horizontal legal-AI tool has
Records-collection orchestration across 8+ custodian categories; every custodian template, redaction quirk, and request pattern learned compounds
Chronology / source-of-truth becomes the team's case file, workflow lock-in
Trust layer is the moat: extractive-only outputs, confidence flags, audit log, zero-retention / no-training data handling for acutely sensitive PII (CPS, medical, psychiatric)
Defender-only by license: confidentiality and work-product control the defense can rely on

~10

Mitigation domains

Each scored for coverage + gaps

8+

Records custodian categories

school/medical/mental-health/CPS/military/employment/prison/court

SourcesACEs framework relevance: 4+ ACEs is associated with ~6x higher odds of incarceration (UK household survey); ~45% of one incarcerated-male cohort had 4+ ACEs. Used as a descriptive evidentiary rubric, not a predictive score.

05Market/ 13

Market: a $3–4B labor-value pool, a software-addressable core

Sized bottom-up; soft inputs flagged and ranged

TAM: records/investigation/sentencing work-product value across all U.S. criminal defense: ~$3.0–4.4B/yr (base ~$3.8B). This is a labor-value pool, not a near-term software wallet (all legal AI across every practice area is only ~$1.45B today)
SAM: defense-side sentencing/mitigation workflow: software-addressable floor ~$110M (≈50,000 serviceable seats × ~$2,200 ACV); full displaceable mitigation-services value ~$0.5–1.5B
SOM: realistic 3–5 yr capture: ~$25–55M ARR by year 5 (base ~$40M); year 3 ~$6–12M (reflecting slow, RFP-driven government procurement)
Capital is small and shrinking (~26 new death sentences/yr), the credibility beachhead, NOT the revenue base. Volume lives in federal + non-capital felony sentencing
Honest read: the two softest load-bearing inputs are the private criminal-defense market (~$10–18B est.) and the % of non-capital felonies that get paid mitigation, SAM swings on both

~$3.0–4.4B/yr

TAM (labor-value pool)

$22B defense pool × ~15–20% workflow share

~$110M floor

SAM (software-addressable)

broader workflow value ~$0.5–1.5B

~$40M / ~$6–12M

SOM yr5 / yr3

~4–6% of workflow SAM, ~1% of TAM

SourcesDefense pool ≈ $8B public (6AC $6.5B + U.S. Courts ~$1.4B) + ~$14B private (ESTIMATE: 3–4% of $417.9B U.S. law-firm revenue, IBISWorld 2024; range $10–18B). Workflow share inferred from RAND hours. Seat SAM: ~15,000 PDs (BJS FY25) + ~30,000 serious-felony/federal private defenders + ~2,000–5,000 mitigation specialists (NASAMS, no public count) + capital teams. Legal-AI market $1.45B 2024 → $3.9B 2030 @17.3% (Grand View). TAM/SAM are estimate stacks, read as a ceiling, not a measured market.

06Competition/ 13

Competition: white space the analog already validated

Vertical × defense-side is unoccupied

EvenUp is the proof: records→evidence-grounded narrative + managed human review, on the plaintiff side, $2B+ valuation, $385M raised, 2,000+ firms, ~10,000 cases/week in ~3 years. Aperio is structurally 'EvenUp for the defense'
Horizontal legal AI (Harvey $11B / ~$300M ARR; CoCounsel, $650M acquisition / 1M+ users; Clio $3B / $200M+ ARR) is civil/BigLaw, not built for biopsychosocial mitigation
Nearest horizontal-defense threat: Paxton AI ($28M raised, $2,999/user/yr) drafts sentencing memos from prompts, but does NOT run multi-domain social-history investigation or framework coverage-scoring
Nearest defense-side player: JusticeText (~$4M ARR, 100+ PD agencies, 4,100+ attorneys), does audiovisual discovery (body-cam, jail calls), not mitigation narrative. Complementary, and proof a defense-only AI tool monetizes
The real competitor is the manual status quo, constitutionally required, routinely skipped
EvenUp's lesson cuts both ways: its large human-review team shows anti-fabrication + human QA IS the product (and bounds gross margin)

$2B+ val / $385M raised

EvenUp (the analog)

Series E led by Bessemer; RELX/LexisNexis participated

$2,999/user/yr

Paxton AI (horizontal-defense)

Drafts memos; no investigation/scoring

~$4M ARR

JusticeText (defense-side comp)

100+ PD agencies; AV evidence, not mitigation

SourcesEvenUp Series E: BusinessWire / LawSites / Fortune (Oct 2025). Harvey: CNBC (Mar 2026) / Sacra. CoCounsel: Thomson Reuters (Aug 2023). Clio: LawNext (Jul 2024). Paxton: LawSites (Jan 2025). JusticeText: Thomson Reuters Institute, $4.0M ARR (Dec 2025). Incumbent-entry risk is real: RELX (EvenUp investor) and Thomson Reuters have distribution to bolt on a module.

07Business Model/ 13

Business model: per-seat SaaS + per-case expansion

Matches how legal buyers buy; aligns price to records-heavy value

Primary: per-seat SaaS sold in org/enterprise tiers (annual, PO- and grant-friendly), how PD offices already buy software
Expansion engine: a per-case 'capital pack' usage overlay that aligns incremental price to the records-heavy cases that drive value and inference cost
Value anchor that survives scrutiny: specialist-hour leverage: Aperio recovers an estimated 50–150 hrs/capital case (~$4k–$11k of billable/staff capacity at $75–100/hr), making a low-thousands seat price trivial
Land-and-expand: land one capital unit or pilot → expand by seats, per-case packs, and adjacent county/statewide offices
Plan for a human-in-the-loop QA layer (EvenUp's playbook), a services component, not 100% automation, given the no-fabrication bar

~$4k–$11k

Recoverable capacity / capital case

50–150 hrs saved × $75–100/hr (estimate)

~110–120%

Target NRR

enterprise benchmark 118%; bias to org/enterprise mix

SourcesPer-case value is an estimate grounded in ABA Supplementary Guidelines workload + $50–100/hr mitigation-specialist billing rates (capital statutory caps ~$75/hr). NRR benchmark: enterprise ~118% / SMB ~97% (Benchmarkit/KeyBanc). Value anchored to specialist-hour leverage and more complete, better-sourced mitigation records, not to reversal-cost 'insurance.'

08Pricing/ 13

Pricing: well below enterprise legal AI

Priced for the most budget-constrained buyer in law

Independent specialist / solo defender: ~$249/seat/mo (~$3,000/yr), self-serve, bundled case allotment
Small nonprofit / firm tier: ~$199/seat/mo at 3+ seats
PD office / enterprise org: negotiated site license, target $1,500–3,000/seat/yr at 10–50 seats with a ~$25–50K org floor
Capital case pack (usage overlay): $1,500–5,000/case by records volume; à-la-carte report drafting $300–800/report
Comps frame the ceiling: Harvey (~$288K/yr, 20-seat floor) is far above this segment; CoCounsel ~$225/seat/mo; Paxton $2,999/user/yr. JusticeText's realized ~$1,000/seat/yr is the real-world floor for this exact buyer, expect grant subsidy or discounting at the high end

~$3,000/yr

Independent / solo

~$249/seat/mo, self-serve

$1,500–3,000/seat/yr

Enterprise PD org

~$25–50K org floor

$1,500–5,000/case

Capital case pack

usage overlay; à-la-carte report $300–800

SourcesComps: Harvey ~$1,200+/seat/mo, ~$288K floor (leaked/third-party); CoCounsel ~$225/seat/mo (Westlaw bundle higher); Paxton $2,999/user/yr (LawSites). JusticeText realized ~$1,000/seat/yr (Thomson Reuters Institute, 2025). Prices are hypotheses to validate with 3–5 design-partner pilots before a public price book.

09Go-to-Market/ 13

Go-to-market: capital wins credibility, federal/felony wins TAM

Beachhead: capital + appellate/habeas defender orgs and independent capital mitigation specialists, sharpest duty (ABA Guideline 4.1), highest willingness to pay (court-authorized expert funds), concentrated and reachable (NASAMS-networked), exact product-method fit
Clean second beachhead: federal sentencing, 64,124 cases/yr nearly all needing a §3553(a) narrative, served by 83 federal defender orgs + ~13,000 CJA panel attorneys
Real market = expansion: non-capital felony sentencing + holistic-defense social-work units in large PD offices
Hybrid motion: enterprise/public-sector sales to the ~50–100 largest defender systems (sticky, slow) + product-led growth (PLG) bottoms-up to ~9,000 NACDL direct members + independent mitigation consultants who buy with a card
Channels: NLADA, NACDL (~90 affiliates / ~40K members), NAPD, capital-defense networks, state PD associations, and CLE curricula (mitigation is taught, embed in training)
Public-sector playbook: cooperative purchasing vehicles (GSA/NASPO/Sourcewell), grant alignment (BJA/Byrne JAG/state commissions), CJIS + work-product trust posture, grant-funded pilots

64,124 cases/yr

Federal beachhead

83 FDOs + ~13,000 CJA panel attorneys (USSC FY2023)

~9,000 + ~40K

PLG channel reach

NACDL direct members + affiliate members

3–5 capital + 2–3 holistic orgs

Design partners

framework co-development under DHAs

SourcesFederal cases: USSC 2023 Sourcebook. Defender system: U.S. Courts Defender Services (83 FDOs, ~4,200 staff). NACDL membership: NACDL. Public-sector cycles run 6–18 months; cooperative-purchasing vehicles shorten RFP friction.

10Moat · Why Us/ 13

Moat / why us

Proprietary mitigation-domain ontology (ABA Guidelines + Supplementary Guidelines + ACEs), the coverage-scoring rubric is hard to copy credibly and is the product
Workflow lock-in: the sourced chronology becomes the team's case file / source of truth
Records-integration network effects: every custodian template, redaction quirk, and request pattern compounds (JusticeText's proven playbook in an adjacent niche)
Trust/defensibility: in a domain where generalist tools hallucinate 17–34%, a verifiable, citation-locked, Bates-linked record is the differentiator they won't build for this niche
Anti-fabrication + human-QA discipline (EvenUp's hard-won lesson) is existential here, first-mover depth in frameworks and records-orchestration is the durable edge
Defender-only licensing builds trust and category ownership with a mission-aligned buyer

SourcesDefensibility caution: white space is unoccupied today but the moat must compound fast, Paxton already surfaces mitigating facts, and RELX/Thomson Reuters have distribution to enter.

11Roadmap/ 13

Roadmap

Phase 1 (0–12 mo): land 3–5 capital/appellate design partners; ship the records-to-domain auto-tagger + ABA/ACEs coverage scorecard; publish a validated extractive-accuracy methodology
Phase 2 (12–24 mo): federal sentencing beachhead (§3553(a) memos across 83 FDOs + CJA panel); PLG to independent specialists + NACDL members; first statewide PD enterprise contracts
Phase 3 (24–48 mo): down-market into non-capital serious-felony sentencing + holistic-defense social-work units, the volume expansion
Adjacencies (the billion-dollar option): clemency/parole, immigration relief, reentry, and civil-legal-aid social histories, the same records-to-narrative engine
Throughout: deepen the ontology, custodian-integration network, and CJIS/security posture that compound the moat

SourcesStandalone mitigation TAM is modest (~$30–150M seat-based); venture-scale returns require the Phase-3 down-market expansion and adjacency capture, not capital alone.

12Unit Economics/ 13

Unit economics (illustrative, pre-pilot)

Bottom-up estimates to validate with design partners

Mid-size org ACV ~$40–60K (15 seats × ~$2,500 + capital packs); independent specialists ~$3–8K
CAC ~$15–40K/org deal (AE + NLADA/NACDL conference presence + pilot); blended payback ~12–18 months on a 6–18-mo public-sector cycle
Inference is cheap and routable: ~$30–90 LLM COGS per heavy capital case (~5,000 pp), ~$5–20 typical, <5% of revenue with model routing, prompt caching of the record corpus, and batch ingestion
Target gross margin ~75–85% at scale, below classic 90% SaaS because the no-fabrication bar requires a human-QA review layer (the real margin variable, per EvenUp), not inference
Path: year 3 ~$6–12M ARR, year 5 ~$25–55M ARR (base ~$40M), NRR ~110–120% via seat + per-case + adjacent-office expansion

~$40–60K

ACV (mid-org)

independents ~$3–8K

~$15–40K / 12–18 mo

CAC / payback

vs. ~14–18 mo mid-market SaaS median

~75–85%

Gross margin

inference <5% of rev; human QA is the swing

SourcesLLM COGS modeled on Anthropic pricing (Sonnet 4.6 $3/$15, Opus 4.8 $5/$25 per 1M; Batch API 50% off; prompt caching reads ~0.1×/writes 1.25×) at ~600 tok/page. ACV/CAC/margin/NRR are bottom-up estimates, not observed deals, validate with 3–5 pilots.

13The Ask/ 13

The ask

Raising a seed round to fund the extractive, citation-locked engine + mitigation ontology, 3–5 capital design partners, and CJIS/security trust posture
Use of funds: product + framework co-development (45%), GTM into the federal + capital beachhead (30%), security/compliance + data handling (15%), human-QA/services layer (10%)
Milestones to Series A: land 3–5 design-partner orgs + first 200–500 paying seats, reach ~$1–3M ARR, and publish a validated accuracy benchmark for biopsychosocial record synthesis
Why back it now: a constitutionally floored, under-served demand; a model already validated at $2B+ on the plaintiff side; and unoccupied defense-side white space with a fast-compounding ontology + records moat
The honest case: a modest standalone TAM with a credible path to venture scale through down-market expansion and adjacency capture

Seed

Stage

Path to Series A on design partners + ~$1–3M ARR

~$40M ARR (base)

Year-5 target

range ~$25–55M

SourcesTargets reflect software-addressable SOM (~$25–55M ARR yr5), discounted from the broader labor-value pool. Capital is the credibility beachhead; revenue scale depends on federal + non-capital felony expansion.

The bottom line

The records-to-mitigation engine for the defense

EvenUp's model. JusticeText's buyer. The mitigation specialist's framework.

A constitutional duty (Wiggins, Rompilla) meets a labor bottleneck (RAND: 286 hrs/LWOP case) and a now-capable technology (1M-context, citation-locked synthesis)
Purpose-built, no-fabrication defense work product, the trust posture generalist tools won't build for this niche
Capital for credibility; federal + felony sentencing for scale; clemency / immigration / reentry for the option value

Open the app →Research brief Product brief

Sources & methodology

Every figure, sourced

Market sizing is bottom-up; soft inputs are flagged and ranged. Figures were adversarially stress-tested before inclusion, shaky numbers were downgraded or dropped.