First-level review is usually the biggest number in a discovery budget. It’s also the work that least needs a partner’s judgment.
This guide covers where that money goes, why first-level review is the phase to bring in-house first, and what changes when you do.
Where the money goes in a matter
Most litigation matters run through four phases:
- Collection and processing. Gathering the data and making it searchable.
- First-level review. Reading each document to decide whether it’s responsive and flagging possible privilege.
- Second-level and privilege review. Senior lawyers make the close calls.
- Production. Load files, natives, and the privilege log go to the other side.
First-level review is where the hours pile up. It scales with the size of the collection, so a bigger matter means a proportionally bigger bill.
A quick way to size it
Take a collection of 200,000 documents after culling. A common benchmark for linear review is about 50 documents per reviewer hour. That’s 4,000 reviewer hours before anyone makes a hard privilege call.
Multiply 4,000 by what your firm or review vendor bills for first-level work. That’s the number you’re trying to change.
Why start with first-level review
It’s high-volume and repeatable. The same judgment gets applied across a large set of documents, which is exactly the kind of work an agent can take a first pass at while your team checks the calls.
It’s also the phase you control least today. You can’t forecast hours on a collection nobody has read yet, so the budget stays a guess until the invoice arrives.
Bringing it in-house doesn’t mean giving up defensibility. Done right, the output is a cited, logged record your team signed off on. We cover what that takes in What “defensible AI review” actually means.
What agentic review changes
Every option before now forced a tradeoff.
| Approach | What you get | What you give up |
|---|---|---|
| Send it to outside counsel | Defensible work product | Cost control |
| Review it manually in-house | Predictable cost | Scale |
| TAR or generic eDiscovery tools | Speed | A reason behind each call |
| Agentic review in-house | Speed, scale, and a cited reason for every call | Nothing on this list |
Agents read and reason instead of ranking. Each call comes with a written rationale cited to the source text. Your reviewers see why a document was tagged, fix what’s wrong, and approve the rest.
The numbers
On a representative single matter (same collection, same requests for production), moving first-level review in-house on Discovery AI produced:
- About 50% lower cost per matter than outside-counsel review
- About 20× faster review than manual or traditional TAR
- An estimated 99% recall across agentic reviews
These figures are illustrative. Your results will depend on collection size, number of custodians, privilege complexity, and timeline.
The bigger shift isn’t the percentage. Review stops being an open-ended hourly invoice and becomes a cost you know before the matter starts.
What it does to your budget
With hourly review, you learn what a matter cost when it closes. With per-matter pricing, you know before you start. Finance can plan around the second one.
For a department that runs the same kinds of matters every year, that turns review from a surprise into a line item. We break down the pricing models in Per-GB vs. per-matter pricing.
What stays with outside counsel
Bringing review in-house changes which work you send out, not whether you use outside counsel. Strategy, motion practice, and trial stay with your firm. The first-level read only got billed by the hour because there wasn’t a better option. Now there is.
Where to start
Pick one recurring matter type with a well-defined collection and requests you’ve seen before. Run it in-house alongside your current process and compare cost, time, and quality. That gives you your own number instead of ours.
If you’d like to model it on one of your matters, we’ll walk through it with you.