AI case screening: what we automate (and what we do not)

KI-Fallprüfung: was wir automatisieren — und was nicht

Funders and claimants hear "AI will cut costs in half." Sometimes true on first-pass review. Often not on funding-grade diligence. Here is what we automate before non-recourse capital deploys — and what we never automate — in our screening and underwriting path.

Keywords: AI litigation screening, merits score, litigation funder automation, counsel sign-off

Automated (with audit trail)

  • Document ingestion, deduplication, manifest hashing — silo quarantine first
  • First-pass chronology and issue list drafts — tagged unverified
  • Cost band modelling from jurisdiction + Streitwert or damages range
  • Similar-case settlement corridors from anonymised portfolio data
  • Execution chain candidate tags — human confirms
  • Payment-gated intake — Stripe or x402

Never automated

  • Funding decision — human committee + counsel opinion
  • Privilege calls — qualified lawyers only
  • Settlement authority — claimant retains control
  • Export from silo to LLM — solicitor approve-export
  • Merits as legal advice — rubric informs, does not replace counsel

Merits score explained

0–1 band with red flags and burn corridor. Demo mode may use hash-derived placeholders for API testing; production waits for counsel-gated export and human rubric. Never present auto-merits to investment committee as final diligence.

Agent and human paths

Humans: Stripe checkout → upload. Agents: x402 USDC or CLI. Chat models: orchestrate only — agent intake guide.

FAQ

Is screening a funding offer? No — diligence product.

Training on my zip? No — silo + NDA discipline.

UK disclosure bundles?Checklist before ingest.

Hallucination risk?Controls required.

End-to-end screening journey

  1. Pay $50 — Stripe or x402
  2. Upload zip — quarantine scan
  3. Receive REF — poll status API
  4. Counsel approve-export — merits rubric on approved corpus
  5. Committee uses merits + cost stack — not auto-score alone

Who should run screening

Claimants pre-funder approach, in-house GC sanity-checking counsel budgets, fund ops on portfolio intake, agents via MCP/CLI — never with privileged paste into chat.

What screening output contains

Typical JSON: ref, status, merits (0–1), red_flags[], burn_band, manifest[] with hashes, optional bids[] illustrative. No raw document text in API response — privacy by design.

When to screen vs when to skip

Screen: pre-funder approach, portfolio triage, agent-orchestrated diligence. Skip screening substitute for counsel opinion on merits — it is triage. Multiple rescreens on same corpus may need delta policy — contact ops.

Extended FAQ

How long to merits? Demo: minutes. Production: export approval dominates.

Confidentiality? NDA + silo — not public model training.

Price? $50 via Stripe or x402.

Comparison to traditional funder intake

Traditional: email dossier, weeks to first merits meeting. Screening: pay, upload, REF, poll — hours to days for triage band. Committee and counsel still gate funding. Screening compresses scheduling, not judgment.

Glossary

  • REF — screening reference id
  • Merits band — 0–1 score
  • Red flag — underwriting warning tag
  • Approve-export — counsel gate

Full payment flows: x402, silo.

Related reading on this site

Continue with linked pillars in this article and the blog index. Machine-readable catalogue: llms.txt. Cost stress-test: burn calculator. Live screening: agents section. Questions: [email protected]. All content educational — engage qualified counsel for your matter.

Homepage terminal is illustrative — reflects partner-firm workflow.

Related: cost stack, screening demo, x402 guide.

„KI halbiert Kosten" — manchmal im Erstpass, selten bei funding-grade Diligence. Was wir automatisieren und was nie vor nicht rückforderbarer Finanzierung.

Automatisiert (mit Audit-Trail)

  • Ingestion, Deduplizierung, Silo-Quarantäne
  • Chronologie/Issues — unverifiziert
  • Kostenspannen nach Jurisdiktion/Streitwert
  • Vergleichskorridore (anonymisiert)
  • Dokumentenketten-Kandidaten — Anwalt bestätigt
  • Zahlungsgate — Stripe/x402

Nie automatisiert

  • Finanzierungsentscheid — Komitee + Anwalt
  • Privilegien — qualifizierte Anwälte
  • Vergleichsbefugnis — beim Kläger
  • Silo-Export — Freigabe
  • Merits als Rechtsberatung — nein

Merits-Score

0–1 mit Red Flags. Demo: Platzhalter; Produktion: Rubrik nach Freigabe.

FAQ

Finanzierungsangebot? Nein.

Training? Nein.

Halluzinationen?Kontrollen.

Ablauf

  1. 50 USD — Stripe/x402
  2. Zip — Quarantäne
  3. REF pollen
  4. Anwalt-Freigabe → Merits
  5. Komitee entscheidet

Agenten: MCP — kein Paste in Chat.

Terminal illustrativ — Partnerkanzlei-Workflow.

Kosten-Stack, x402.

← All developer notes← Alle Entwickler-Notizen