When to Retain, and What to Send First

Retention here is a preservation decision, not a trial decision

In most commercial matters an expert is a trial-phase problem. Liability gets framed, discovery closes, and someone is retained to opine on what the documents show. That sequence works because the documents sit still.

In a digital marketing dispute they do not. The advertising and analytics record is generated by platforms, held by platforms, and deleted by platforms on published schedules that run whether or not a complaint has been filed, whether or not a hold letter has gone out, and whether or not anyone on the file has yet realized the case turns on it. None of the major products exposes a litigation-hold control. The retention windows are short relative to the pace of litigation, they are rolling rather than fixed, and the most granular data — the layer that would answer the hardest questions — expires first.

So the timing question is not "when will an expert opinion be needed?" It is "what will still exist when it is?" A claim that was straightforward at week three is often unprovable at month nine, not because the facts changed but because the record aged out underneath them.

What too late actually looks like

The failure has a recognizable shape and it is almost never dramatic. Nobody deletes anything. A protective order takes eleven weeks to negotiate because one side wants an attorneys'-eyes-only tier. Production is staged. The first tranche arrives as summary reports because that is what the client's marketing coordinator knew how to export. A supplemental request for the underlying data goes out four months in. By then the granular layer has rolled off, and what comes back is an aggregate that cannot answer the question the aggregate was requested to answer.

Two consequences follow, and they are asymmetric. For the party that needed the granular record, the analysis it would have supported is simply unavailable, and no motion recovers it. For the party defending, the absence is a live argument in both directions — it may be a gap in the opponent's proof, or it may be a preservation problem attributed to whoever had practical control of the account.

An expert brought in at that stage can still describe what is gone and when it went, which has value. It has far less value than having been there in week two to say which objects to preserve.

What an early expert changes about the discovery itself

The concrete argument for early retention is not that the analysis gets done sooner. It is that the discovery goes out asking for things that exist, by the names the systems use for them.

Requests drafted without that input tend to ask for "all advertising data" and receive a folder of screenshots. Requests drafted with it name the object: the change history for a specified account over a specified range, the billing or transaction record as distinct from the reporting interface's cost column, the conversion-action configuration including the attribution model and conversion window in force, the account and access grant records showing who held what permission and which payment profile paid, the tag or container configuration and its version history, and the native export rather than a rendering of it.

Early involvement also changes the preservation letter, which in this field has to name accounts and properties rather than custodians, because the data is not on anyone's laptop. And it lets counsel test a theory before it is committed to a pleading. Finding out in week two that the disputed channel and every undisputed channel fell together is unwelcome. Finding it out in an opposing expert's report is worse.

The first email: what to send, and in what order

This is the list, ordered so that useful work can start at the lowest cost.

1. The operative pleadings. So the analysis can be aimed at the claims it has to serve rather than at whatever is interesting in the data.
2. The contract, every statement of work, and all amendments. In an agency or vendor matter, no scope or performance opinion is possible at all without these. They are the strongest source of duty by a distance, because they are what the parties actually agreed.
3. The scheduling order, with the expert disclosure and rebuttal dates. These drive everything downstream, and the rule's defaults are usually displaced by the order.
4. The protective order, before any confidential production moves. If it is not yet entered, say so — that fact is itself a preservation risk worth tracking.
5. What has actually been produced, in native format with metadata intact, plus the production log. The log matters as much as the documents, because it shows what was requested and what came back.
6. The other side's expert report, if this is a rebuttal engagement, together with that expert's reliance materials — which Rule 26(a)(2)(B)(ii) required them to identify.
7. Any preservation letters and litigation-hold notices relating to ad accounts and analytics properties, so the gap between what was held and what the platforms were deleting anyway can be established while it is still establishable.

A short note on what is missing is as useful as the material itself. "There is no written statement of work; the scope was set in a deck" changes the shape of the engagement immediately, and it is better said in the first email than discovered in month three.

Native format, and why a PDF of an export is not the export

Exports converted to PDF are frequently useless for this work, and the reason is not fastidiousness.

A native export carries the column headers that identify which metric definition was used, the date range and time zone, the segments and filters applied, and often the extract timestamp. Those fields are the evidence. Whether a figure is "conversions" or "all conversions," whether it sits in the account time zone or the reporting time zone, whether a filter excluded internal traffic — these determine what the number means, and they are exactly what a print-to-PDF discards or buries in a header nobody transcribes.

Native format also preserves the ability to reproduce the analysis, which matters because FRCP 26(a)(2)(B) requires a report to identify the facts or data considered with enough specificity that the other side can follow it. An analysis performed on a flattened rendering of a report cannot be checked by anyone, including the expert who performed it.

Where the producing party genuinely cannot export natively, the fallback is to document what was requested, what was received, and what could not be verified as a result. That statement belongs in the report rather than in a footnote.

What counsel should be deliberate about sending

Two protections in FRCP 26(b)(4) shape the file, and they shape it in a way that rewards a little structure at the start.

Drafts of a report or disclosure required under Rule 26(a)(2) are protected "regardless of the form in which the draft is recorded." Communications between counsel and a reporting expert are also protected — with three exceptions. Communications relating to compensation are discoverable. So are those identifying facts or data counsel provided that the expert considered, and those identifying assumptions counsel provided that the expert relied on.

The practical consequence is not that counsel should send less. It is that counsel-supplied facts, data and assumptions should live in an identifiable, dated list from day one, so that the discoverable category is a document rather than a search through months of correspondence. That list is also most of what the report's own statement of facts and data considered needs to say, which means the discipline pays for itself twice.

The related point is about the theory. Sending the case theory is fine and often necessary. Sending it as an assumption the expert is to adopt is a different thing, and an expert who adopts a client's causal narrative without independently testing it is vulnerable on exactly that point.

Scope, stated so the answer is not built into the question

The scope statement in the engagement letter is worth ten minutes of attention because it will be read aloud.

"Reconcile agency-invoiced spend against platform-recorded spend for the period X to Y, and identify and characterize any divergence" is a scope. "Determine whether the agency overbilled" presupposes the answer, and it hands opposing counsel a clean line: the expert was hired to reach a conclusion and reached it. The same applies to causation assignments. "Analyze the traffic and revenue series for the period and identify which alternative explanations the available data exclude, are consistent with, or cannot address" is an assignment that can come back either way, which is precisely why an opinion produced under it carries weight.

Where the assignment genuinely is narrow — reviewing another expert's method, for example — say so, and say what was not examined. A scope that is candidly limited is far easier to defend than one that implies a comprehensive review nobody performed.

What early retention does not fix

Early retention is a preservation and framing advantage. It is not a cure, and three things stay broken regardless of timing.

Data that expired before anyone was retained is gone. If the conduct is two years old when the complaint is filed, the granular layer has already rolled off several platforms, and nothing about being retained early recovers it. The honest deliverable in that situation is a statement of what could not be checked.

A causal measurement cannot be created after the fact. The only methods that measure the incremental effect of advertising require a holdout built while the campaign runs. Retaining an expert early in the litigation is not early enough; the relevant "early" was during the campaign, years before.

The production is still the production. Communications live on personal devices and in messaging apps, ad accounts get deleted at the end of a vendor relationship, and a party that does not have a record cannot produce one. An expert must state what was unavailable and what could not be verified — which is also the answer to "did you consider X?" when it arrives at deposition.

Frequently Asked Questions

Published 2026-08-12