Document review used to scale linearly with page count. In 2026 the curve bends — then breaks — when cheap automation becomes expensive correction. Litigation funders pricing Tranche 1 need to know which phase a matter is in, or they misread €/page quotes as sustainable burn. This article explains the three phases, variables that move €/page, and red flags in vendor proposals.
Keywords: document review cost, e-discovery pricing, AI legal review, litigation burn
Three phases of the curve
- Phase 1 — Manual dominance — €8–15/page all-in for complex regulatory matters; linear scaling with corpus
- Phase 2 — AI assist — 30–50% first-pass savings if protocol and privilege model are fixed before automation starts
- Phase 3 — Correction tax — full re-review after hallucinated issues, missed families, or bad privilege calls
Variables that move €/page
| Variable | Effect on cost |
|---|---|
| Languages | Multi-lingual corpora need translator QC layer |
| Privilege model | Unset model → Phase 3 rework |
| Corpus cleanliness | Email threads, duplicates inflate volume |
| QC depth | 1% vs 10% sample changes effective €/page |
| Hosting/processing | Fixed platform fees before review hours |
What funders model
We ask for €/page bands with explicit assumptions: corpus size, languages, privilege model, QC depth, and vendor vs counsel split. A low headline rate without QC definition is a red flag in Tranche 1 covenants. Tie review spend to full cost stack — review is rarely the only line item.
Vendor proposal red flags
- "AI will review everything" without privilege protocol owner
- Per-page rate without hosting/processing breakout
- No language for supplemental corpus mid-case
- QC described as "sampling" without percentage and escalation rules
FAQ
Can AI skip Phase 1? Sometimes on clean corpora — rarely on commercial litigation with privilege fights.
How to avoid Phase 3? Fix protocol, silo export, human sign-off — silo.
Waste overlap?Seven useless spends — duplicate review.
Worked example (anonymous)
100k-page corpus: Phase 1 quote €1.2M all-in. Phase 2 AI assist with fixed privilege model: €650k first pass + €120k QC. Phase 3 failure (missed family): €400k rework — total €1.17M with worse counsel morale than Phase 1. Funder memo tagged protocol owner and QC % before approving Tranche 2 review draw.
Integration with cost stack
Review is one layer in full lawsuit stack — never the only line in Tranche 1 model.
Negotiating with review vendors
Require: privilege protocol sign-off before start, QC sample rate in contract, supplemental corpus rate card, exit migration assistance, AI assist transparency (which model, which fields human-reviewed). Funders attach vendor summary to Tranche 1 memo.
Technology change mid-matter
Switching platforms after month 4 without case boundary is item 6 in useless spends. If switch required, budget parallel run overlap — not big-bang migration.
Extended FAQ
€/page vs €/doc? Doc pricing favours email-heavy corpora — disclose assumption.
Hosting separate? Yes — show in burn report distinct from review hours.
Sample funder covenant language
"Review vendor and €/page assumptions fixed in Tranche 1 memo unless corpus grows >20% or committee approves vendor change." Ties curve phase to stack governance.
Glossary
- QC depth — sample rate reviewed by humans
- Privilege model — rules for confidential material
- Phase 3 tax — rework cost after bad AI pass
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.