AI Was Supposed to End Document Review. Reviewers Have Never Been Busier.

Cheaper AI review multiplies matters, so buy for validation and volatile staffing, not for headcount cuts

In a November 2026 opinion piece on EDRM, Rachi Messing of Altorney argues that generative AI has not ended human document review. His company sells a GenAI review engine called MARC and also runs a platform for staffing human reviewers. He says the staffing side had a strong 2025 and is on track to double in 2026, and that MARC is growing too. His explanation is that each matter needs fewer reviewers, but cheaper review makes many more matters worth reviewing, so total demand for people rises. He describes his headcount examples as illustrative, and EDRM notes the views are his own.

Is AI actually shrinking review teams?

Per matter, yes. Across a whole practice, Messing says no. Each case needs fewer reviewers, but cheaper review brings in matters that were previously settled early, resolved on search terms, or skipped as too expensive.

It is a rebound-effect argument: lower cost per document expands the number of matters where review makes sense. It is plausible, but the evidence is thin. The headcount figures are hypothetical by the author's own admission. The only real data point is his own company's growth, reported without a base, so "double" tells you little about market-wide reviewer demand. A vendor with a foot in both camps will see both lines rise, which is not independent confirmation.

What work is left for human reviewers?

Validating AI output, quality-controlling productions, and standing behind the process when opposing counsel or a regulator asks how it was done, plus anything outside what the technology currently handles.

Messing says images, handwritten notes, poor scans and privilege were once reliable reasons to route documents to people, and that today's AI handles them well. He gives no error rates, no test sets and no measurer for "well". That is the claim to push on. If people are now the defensibility layer, the tooling has to make their checking cheap: sampling, audit trails, and answers that cite the source exhibit so a reviewer can verify them quickly. A platform that produces fluent summaries without that traceability moves the cost to the validators.

Should we keep an in-house review team?

Messing reports that several organisations that once ruled out outside platforms are reconsidering, because smaller, more numerous and less predictable matters make a large standing team hard to justify.

Treat that as a hypothesis to test against your own matter history, not a finding. "Several" is anecdotal, and the author sells flexible reviewer staffing, so he gains if you conclude that fixed overhead is the problem. The useful part is the volatility point: the real risk is a workload that swings from a handful of reviewers to a large crew for a short burst. Check how variable your own caseload is before deciding either way.

What should I ask my review vendor?

Ask who measures validation results and how, what human work remains per matter in practice, and whether pricing copes with erratic matter sizes rather than assuming a steady pipeline.

Also ask where the per-document savings go. More matters in aggregate does not help a client paying for one of them. This site has covered the push for transparency with clients on AI's effect on fees, and review is where that question lands first. Ask for protocols, sample sizes and validation reports from matters like yours, not a market-level narrative.

Frequently asked questions

Is this independent evidence that review demand is rising?

No. It is an opinion piece by an executive whose company sells both AI review and reviewer staffing, built on illustrative numbers and self-reported growth.

Has AI solved scans, handwriting and privilege?

Messing says today's AI handles them well, but offers no measurements. Test on your own data and require documented validation before relying on the claim.

Will reviewers per matter keep falling?

His pattern suggests so. He says each technology improvement has moved reviewers to a different part of the workflow rather than out of it, with more weight on validation and ownership of the result.

Sources: EDRM, "AI Was Supposed to End Document Review. Reviewers Have Never Been Busier." by Rachi Messing.