A Litigation Hold You Manage by Email Is a Sanctions Motion Waiting to Be Filed.

The duty to preserve evidence starts the moment litigation is reasonably anticipated, not the day the complaint is served. A hold sent as a one-off email, with no tracking of who acknowledged it or what got deleted anyway, is exactly the gap opposing counsel builds a spoliation motion on. Rule 37(e) gives a court real teeth, up to telling the jury to assume the missing documents were bad for you.

I have come to believe that most mid-size litigation practices and in-house legal departments in South Florida are one bad custodian away from that instruction. The problem is not that people do not know the rules. The problem is that the workflow for executing those rules was designed around paper and fax and never got rebuilt for a world where a single custodian might touch Slack, Gmail, Salesforce, and three shared drives in one afternoon.

Identifying custodians at the moment the hold triggers

The first failure point in a manual hold process is custodian identification. A partner sends an email to the five people they know are involved. Three weeks later, discovery reveals two more custodians who had relevant communications from day one. The opposing party files a motion. The court looks at the timeline.

Under FRCP 26(b)(1), proportionality in discovery requires the party resisting production to show burden relative to likely benefit, but you cannot make that argument credibly if your custodian list was assembled by memory at 11pm.

An AI agent working against your matter management system, whether that is Clio Manage, Relativity, or a practice-specific instance of Salesforce Legal, can pull organizational hierarchy data, communication metadata, and prior matter history to surface custodians the hold-issuing attorney did not name. That output is a documented artifact, timestamped, tied to a matter number. When opposing counsel asks "how did you identify custodians," you have an answer that is not "we asked around."

The agent does not make the legal judgment about who must be preserved. That stays with counsel. What it does is surface the organizational graph so the judgment is made with complete information.

Hold issuance and the acknowledgment record

A hold email with no read receipt and no tracking mechanism is an assertion, not a record. In Pension Committee of the University of Montreal Pension Plan v. Banc of America Securities, 685 F. Supp. 2d 456 (S.D.N.Y. 2010), Judge Scheindlin laid out a framework for gross negligence in preservation that shaped hold practice for years. The 2015 amendments to FRCP 37(e) specifically rejected the negligence-based adverse-inference trigger from that line of cases — Rule 37(e)(2) now requires a finding of intent to deprive. Scheindlin's diagnostic of what gross preservation failures look like in practice still tracks how courts read the record, even though the sanctions calculus has tightened.

An AI agent handling hold issuance sends the notice, tracks whether each custodian opened it, records when they acknowledged it, and flags non-responses after a configurable window, say 48 or 72 hours, for attorney follow-up.

That is not complicated logic. What makes it defensible is that the log is immutable and exportable. When the court asks for documentation of your preservation efforts, you produce a custodian-by-custodian acknowledgment ledger, not a forwarded email chain.

Platforms like Exterro and Zapproved have offered hold management modules for years. The shift now is that the same orchestration layer that tracks acknowledgment can also trigger downstream actions: applying a preservation hold in Microsoft 365 Purview that overrides retention policies for named custodians, placing a litigation hold flag on a Salesforce record, or notifying IT to snapshot a shared drive. The coordination that used to require three phone calls and a follow-up ticket now runs as a single triggered workflow.

Monitoring custodian compliance after the hold issues

Issuing the hold is the beginning, not the end. Custodians change roles. People leave the firm or the company. New custodians become relevant as the case develops. A hold issued in January and never revisited is a standing invitation for a spoliation argument by March.

The workflow problem here is the cost of re-review. A litigator working a docket of forty active matters cannot manually re-audit custodian status every thirty days. What an agent can do is check against HR system updates (terminations, transfers), flag new names appearing in produced communications who were not on the hold list, and push a weekly status digest to the responsible attorney with exceptions highlighted, not a list of everything, just the open items that require a human decision.

This is where the proportionality argument under FRCP 26(b)(1) starts to work in your favor. If you can show the court a documented cadence of compliance monitoring, with dates and custodian-level status, you have a better record on the intent-to-deprive question under Rule 37(e)(2). The qualifier matters: monitoring is protective only when it is paired with documented response to exceptions. A log that shows you saw a custodian go dark and did nothing is worse than no log at all. The discipline is response, not just observation.

ESI volume triage before the review clock starts

The discovery cost fight in complex commercial litigation in Miami-Dade and Broward courts increasingly centers on ESI volume. A corporate defendant with a custodian pool of twenty people generating two years of email, Slack, and shared drive data can hit several million documents before any first-level review runs. At first-level linear review rates of roughly $1 to $5 per document depending on tier and before any TAR 2.0 workflow, the math gets ugly fast.

An AI agent integrated with a collection tool, Nuix, Relativity, or OpenText Axcelerate, can run pre-review analytics: deduplication, near-duplicate clustering, email thread suppression, and domain filtering (stripping out clearly irrelevant external mail domains before review even starts). These are not new capabilities. What is new is the ability to run them automatically against a defined custodian set the moment collection closes, produce a volume report with category breakdowns, and feed that report directly into the proportionality meet-and-confer with opposing counsel.

On the matters where we have run dedup, thread suppression, and domain filtering before review opens, the population can drop materially before a reviewer touches anything, and most of what comes out is email threading and system-generated noise rather than substantive content. That reduction changes the economics of the case, which changes the settlement calculus.

The other edge of the audit trail

The honest counterargument to everything above is that the same logs that document your compliance also document your failures. An automation that mis-triggers, over-preserves, or fires a preservation hold against the wrong custodian set produces a precise record of that mistake. Over-preservation has its own costs: storage, review burden, privilege exposure on documents that should have aged out. And the agent's own decision logs and audit trail are themselves discoverable under Rule 26 if the other side argues your hold process was unreasonable.

That is not an argument against automation. It is an argument for treating the automation as a regulated piece of legal infrastructure, with attorney sign-off on scope changes, periodic audits of what the agent is doing, and clear written procedures for how exceptions get resolved. The audit trail cuts both ways. You have to be willing to live with what it records.

Rule 37(e) sanctions as a financial argument

Rule 37(e) sanctions are not abstract. In GN Netcom, Inc. v. Plantronics, Inc., 930 F.3d 76 (3d Cir. 2019), affirming the underlying District of Delaware sanctions order, the court engaged squarely with the bad-faith standard under Rule 37(e)(2) after a senior executive deleted relevant emails and instructed others to do the same. The trial court awarded punitive sanctions, permissive adverse-inference instructions, and millions in fees. An adverse-inference instruction in a jury trial is not a fine. It is the court telling the jury to assume the documents you failed to preserve would have proven the other side's case. In a contract dispute or a trade-secret case, that instruction can be outcome-determinative.

The legal and financial exposure here is thus not proportionate to the cost of fixing the workflow. A firm running a proper hold management system, with documented custodian identification, acknowledgment tracking, and compliance monitoring, can show a court exactly what it did and when. That documentation is the difference between a curative measure under Rule 37(e)(1) and an adverse-inference under Rule 37(e)(2).

The workflow investment is a fraction of one sanctions motion, and it is certainly a fraction of what a lost case costs.

Five questions to ask any vendor selling hold management automation

Before you sign a contract for any platform or agent-based workflow in this space, I would put these five questions directly to the sales team:

1. Can your system produce a per-custodian acknowledgment log, with timestamps and method of delivery, that is exportable in a format I can attach to a court filing or a declaration?

2. How does the preservation hold work across Microsoft 365 Purview and Google Vault simultaneously, and what happens when a custodian's mailbox is in one environment and their file storage is in another?

3. If a new custodian becomes relevant after the initial hold issues, how does the system surface that name, and how does it document the decision to add or exclude them?

4. What is the audit trail for hold modifications, specifically, if I expand or narrow the scope of a hold six weeks after it issues, is that change logged with the identity of the attorney who made it and the reason?

5. Where does the analysis run, and what is your written representation that custodian content, including privileged communications and work product, is not used to train any model and is not retained outside our tenant?

If the vendor cannot answer all five questions with a product demonstration, not a slide deck, keep looking.

FloridAI Agency works with litigation practices and in-house legal departments in South Florida to build and deploy workflow automation across matter management, discovery operations, and compliance. pablo@floridai.agency