AI eDiscovery: What In-House Counsel Should Ask Vendors

Legal departments are using legal AI for eDiscovery to move early case assessment and review oversight closer to the start of a matter. The vendor may run the platform, but in-house counsel still needs the review method, validation sample, data controls, and cost model in writing.

Jimmy Toy, Chief Legal Officer at Articore Group, managed 40 to 50 active IP cases across the United States, Australia, Europe, and Asia from an in-house seat. In a conversation with GC AI CEO Cecilia Ziniti, he described how class-action discovery can pull the CTO, product, engineering, marketing, and finance teams into the matter:

It's time consuming, especially e-discovery. In class actions it gets very time consuming and disruptive. You reach out to the CTO to understand how part of the platform works, they might get deposed, and it extends across teams, product and engineering, marketing, even finance.

That cross-functional burden keeps the in-house role focused on oversight. Outside counsel and the eDiscovery vendor manage the AI-assisted review, including the model and platform. In-house counsel still answers for how the review happened when a court, a regulator, or the CFO asks.

GC AI is an enterprise legal AI platform built for in-house teams. Founded by three-time general counsel Cecilia Ziniti, GC AI helps legal teams research the law, work with matter documents, and verify the source language behind decisions. As of October 2026, 2,200+ legal teams including 300+ law departments use GC AI. In this workflow, in-house counsel uses it before and after the vendor call:

Before the vendor call: Use Research and US Case Law to establish the current defensibility standard and identify the cases to raise.
After the call: Upload the vendor's statement of work and your notes to Files, then use Exact Quote to compare the written commitments with what the vendor promised.

Why Move eDiscovery Oversight In-House With AI?

Moving eDiscovery oversight in-house gives legal a direct view of the method, timeline, budget, and data controls. The vendor runs the review while legal owns the oversight.

Cost control: Separate standard culling from work that requires legal judgment. Ask what the vendor fee includes, what outside counsel bills separately, and what triggers a change order.
Speed: Use early case assessment (ECA) to understand custodians, themes, likely volume, and risk sooner. That helps the business decide whether to settle or litigate before full review.
Security and privacy: Keep corporate data, personally identifiable information (PII), and privileged communications in a controlled matter environment. Put access, retention, deletion, and any model-training terms in the engagement.

What In-House Counsel Owns in an AI eDiscovery Review

Outside counsel and the vendor run the review platform, code privilege calls, and produce the responsive set. In-house counsel decides whether the AI eDiscovery process holds up, sizes the engagement, and signs off on the methodology. Three duties sit inside that role:

Approve the review methodology before the vendor starts, and confirm which standard it has to satisfy: opposing counsel's agreement, a judge's order, or a regulator's expectations.
Confirm the validation and sampling plan up front, while there is still time to fix a problem before the final production.
Track cost against the litigation budget monthly, which catches scope creep months before the final invoice does.

Six AI eDiscovery Questions to Ask Vendors

Ask each question before a single document is processed:

What AI or TAR method is the review running? Traditional technology-assisted review trains on a matter-specific seed set a human reviewer codes by hand. Newer LLM-based review starts from written instructions with little or no coded training set. Ask which one you are paying for and why the vendor picked it for this matter.
How was the model trained or seeded, and on whose data? If an inexperienced contract reviewer codes the seed set, the model inherits those mistakes at scale. Ask who coded it, how many documents it covered, and whether outside counsel reviewed the coding before training started.
What is the validation and sampling protocol, and who reviews the sample? “95% accurate” without recall and precision figures from a named sample is a marketing line. Ask for the sample size, the recall and precision it produced, and whether opposing counsel gets to see them.
Who owns and can access your documents during the engagement? Confirm where the data lives, who at the vendor can see it, and whether it sits in a segregated environment or a shared one. Put the answer in the engagement letter.
How is the work priced, and what triggers a change order? Per-gigabyte pricing, per-hour review rates, and flat engagement fees each shift the incentives. Get the pricing model and the change-order trigger in writing while the document count is still an estimate.
What happens to your data and any derived model when the matter closes? Ask whether your documents, or any model trained on them, get deleted, retained, or reused on a future matter. Put that answer in the engagement letter too.

How Courts Evaluate AI eDiscovery Review

Courts have accepted technology-assisted review (TAR) when the process is reasonable, transparent, and supported by quality-control testing and validation.

The standard traces to Da Silva Moore v. Publicis Groupe, 287 F.R.D. 182 (S.D.N.Y. 2012), the first opinion to approve predictive coding for document review. Judge Andrew Peck approved the method based on the parties' agreement, the volume of documents, the technology's demonstrated accuracy, cost proportionality, and the transparent process both sides could see.

In Rio Tinto v. Vale, 306 F.R.D. 125 (S.D.N.Y. 2015), Judge Peck described TAR as an accepted tool for document review and emphasized that the parties' agreed protocol, transparency, and quality-control testing mattered.

The cases provide a useful diligence frame for evaluating newer LLM-based workflows. Ask for the methodology, the validation sample with real numbers behind it, and the process for transparency and cooperation. A vendor's defensibility claim is as strong as the sampling data it can produce on request.

How to Control AI eDiscovery Vendor Spend

The 2024 ACC Law Department Management Benchmarking Report reports that legal departments allocated roughly 52% of legal spend internally and 48% externally. Use that split to frame a matter-specific budget. Two questions keep AI eDiscovery spend visible beyond the pricing model itself:

What does the vendor's price include, and which review tasks are billed separately?
Can the vendor provide a monthly report tied to review progress, with documents processed, share complete, and hours billed side by side?

How GC AI Supports AI eDiscovery Vendor Oversight

Use GC AI around the vendor-run review: establish the legal standard before the call, then check the statement of work against the vendor's promises after it.

Before the first vendor call, take the defensibility question into Research, which pulls from primary law and returns citations, and US Case Law links each opinion to its full text. Here is the shape of the question:

We are defending a class action in the Southern District of New York and our vendor proposes an LLM-assisted document review. What have courts in this district required to accept a technology-assisted review protocol? Start from Da Silva Moore v. Publicis Groupe and Rio Tinto v. Vale, add anything more recent, and list the validation elements courts have accepted, with citations.

Use those results to turn the six questions above into the vendor-call agenda.

After the call, upload the vendor's statement of work and your call notes to Files and check the promises against the paper:

Attached are the vendor's statement of work and my notes from our call. For each of these six commitments, quote the SOW language that covers it or say it is missing: (1) the named review method, (2) who codes the seed set, (3) the validation sample size and the recall and precision it will report, (4) segregated data storage, (5) the change-order trigger, (6) deletion of our data and any derived model at close.

Exact Quote returns the SOW's own language for each commitment, character for character, so the gaps are visible before anyone signs. The same workflow starts before vendor selection, with the litigation hold notice that begins the preservation clock.

Keep an oversight file with the methodology you approved, the sample you checked, and the statement of work you verified. A judge, a regulator, or your CFO may ask how the review happened.

Start Your 14-Day Trial

Frequently Asked Questions

What Is AI eDiscovery?

For litigation and investigations, AI eDiscovery helps legal teams identify, collect, process, and review electronically stored information. It can automate work that otherwise requires large manual review teams, including document classification and summarization. In-house counsel typically oversees the process while outside counsel or a vendor operates the review platform.

Does AI eDiscovery Create Data Privacy Risks?

Yes, especially when a vendor's data handling is weak. AI eDiscovery can create data privacy risk around storage isolation, encryption, and whether client documents or derived models enter external training. For an eDiscovery vendor, in-house counsel should confirm those safeguards, who can access the matter, and what happens to the data and any derived model at close before documents are uploaded.

Can AI Identify Privileged Documents in eDiscovery?

Yes. AI can flag potentially privileged documents in eDiscovery. A lawyer still needs to review the flags before making a final privilege determination. Ask how the vendor trained the model to identify privilege markers and how it validates the results across the matter.

Is AI eDiscovery Defensible in Court?

AI eDiscovery can be defensible in court when the review follows a reasonable, transparent process and the team validates the results. In Da Silva Moore and Rio Tinto, courts accepted technology-assisted review with attention to methodology, transparency, and quality control. Ask the vendor for the sampling protocol and the recall and precision figures behind any accuracy claim.

How Can In-House Counsel Control AI eDiscovery Costs?

Start with the vendor's pricing model and change-order triggers. Ask which review tasks the fee includes, which work outside counsel bills separately, and whether the vendor will report documents processed, share complete, and hours billed together each month.This is general educational information for in-house legal teams. It is not legal advice for any specific matter.