How I Work on a Paid Search Matter

The first call, and what it is for

The first conversation is short and deliberately general. Until a conflict check clears I cannot take case material, so what is useful is the shape of the matter rather than its contents: what kind of dispute it is, which platforms the spend ran on, roughly what date range is in question, who holds account access today, and what dates are already fixed.

The most valuable answer that comes out of it is usually to a question attorneys rarely ask first: how much of the record still exists.

The conflict check comes before anything else

A conflict check compares the parties in your matter against every party I have worked for or against — in testimony and in ordinary consulting work, because a prior operating engagement is as much a conflict as a prior retention. What I need is names: the parties, their parent and affiliated entities, the agencies and consultants involved, the platform if a platform is a party, and counsel on both sides.

If a conflict exists, the correct outcome is that I never saw your case material, and the only way to ensure that is for it never to have been sent. Names alone disclose nothing about your theory.

Scoping the question in writing

An engagement is scoped as a question, not a subject. “Whether the account record shows spend continuing after the pause directive of a given date, and in what amount” is a scope. “Review the Google Ads account” is not.

Scope goes in writing before work starts: the question, the materials I expect to rely on, what I am expressly not opining on, and whether the role is testifying or consulting-only. When scope changes it changes in writing: the list of materials relied upon has to be true when the report is signed.

What access or data I need

Paid search is not one system, and a request naming only “the Google Ads account” will miss most of the record. What I normally ask for:

Read-only account access, or complete exports. Access beats a produced spreadsheet, because it lets me segment the data rather than accept somebody else's segmentation.
Change history from both surfaces. The interface export and, where the window still allows it, the programmatic pull. They are not equivalent records.
Manager account structure and link history. Which manager account owns the advertiser's account, when the link was created, and who was granted or removed as a user.
Billing from both sides. Platform transaction records and the agency's invoices. Reconciling the two is the entire case in an overbilling matter.
Tag and analytics configuration. Container version history, conversion and key-event definitions, and in writing the analytics property's current and historical retention setting.
The contract stack. Master agreement, statements of work, insertion orders, media plans, and any performance target referenced in them.
Directives. The correspondence in which budgets were approved, campaigns paused and targeting changed. The account shows what happened; this is where you find what was asked for.
Third-party records. Call tracking, lead capture, the customer database, reporting tools, and server logs wherever invalid traffic is at issue.

One thing to know before a preservation demand goes out: an agency that created the account under its own manager account owns it by default, and either side of a manager link can sever the link.

The retention clock sets the order of the work

Paid-search evidence ages out on published schedules that run from today, not from the date of the alleged wrong. That is why preservation and export come before analysis. These figures were read from platform documentation on August 14, 2026, and they change.

Google Ads change history in the interface: two years. Google states that data older than two years will not be available.
The field-level change record through the Google Ads API: thirty days, and ninety days for the coarser record of which resources changed.
Google Ads hourly, daily and weekly reporting data: thirty-seven months, effective June 1, 2026. Past that line a day's spend can no longer be tied to a specific change.
Microsoft Advertising change history: six months at every granularity, against thirty-six months of performance data.
Meta advertising insights: thirty-seven months for aggregate totals, thirteen for unique-count fields and hourly breakdowns, six for frequency breakdowns, effective January 12, 2026.
Analytics retention is a setting a party controls. Shortening it deletes the affected granular data in the next monthly process.

A preservation letter does not change a retention setting or extend a platform window, and no platform's documentation says a hold pauses its clock. Once a window closes there is no backup to request.

What the analysis actually involves

Reconstruction means building a dated timeline of what the account did — from the change record, the campaign settings and the day-by-day spend — and setting it against the directives and the contract. It either puts a specific act next to a specific instruction or shows that no such act is recorded.

Reconciliation means testing whether the numbers different systems report about the same events agree, and explaining every place they do not. Google publishes its own list of reasons its click count diverges from third-party tracking: repeat clicks, stripped referrer headers, browsers without JavaScript, redirect implementations, and its own filtration of clicks it judged invalid. A discrepancy with a documented cause is not a finding.

Then the questions that decide whether a comparison is legitimate at all. Did the conversion action definition change mid-period? Did the attribution model change? Did modeled conversions begin? If so, the figure before and the figure after are not the same measurement.

What gets delivered

Usually two things. First, a short written read-out to counsel: what the record contains, what it appears to show, what is missing, and what would have to be obtained to go further. It is not written to be filed; it exists so you can decide whether a report is worth commissioning.

Second, if the matter goes that way, a written opinion: the question, the materials relied on and how each was obtained, the method in enough detail that another examiner could repeat it, the findings, and the limits. The limits section is not a formality — it names what the record cannot establish, and it is drafted before the findings are finalized. Exhibits are the underlying exports rather than pictures of them, because a screenshot preserves a rendering and not the record.

Findings are stated as what the data shows or is consistent with. They may be challenged, and no opinion I write asserts an outcome.

When the data does not support the theory

Sometimes it does not, and I say so in the read-out, before a report exists. There are three usual versions: the record is gone because the window closed while the matter was worked up; the record shows something other than what the theory requires; or the dispute was never a data question, because the argument is about what was promised or what somebody intended.

The last one deserves spelling out, because there is no published standard of care in this field to measure conduct against. A claim framed as departure from professional norms has to be rebuilt on the contract, the scope of work and the platform's own documentation, or it has no anchor — and an examination of the account will not supply one.

What drives cost, and how engagement models differ

I do not publish figures; any offered before the shape of a matter is known would be fiction. What is worth understanding is what moves the number.

How many accounts and platforms. One search account over one date range is a different exercise from four platforms and two manager accounts.
Whether access exists. Read-only access is faster than a produced export, and far faster than a record that has to come through discovery first.
Whether the record has to be rebuilt. Once a native window closes, the record gets reassembled from billing, third-party tools, logs and reporting artifacts — the most expensive work in this field, and caused almost entirely by delay.
Whether the role is testifying. Deposition and trial time, with the preparation attached, is the largest variable after the data itself.
Timing. A compressed schedule costs more than an ordinary one.

The models differ in what is produced and who sees it. A preliminary assessment ends at the read-out memo. A consulting engagement is non-testifying work supporting counsel's own analysis and discovery. A testifying engagement produces a signed report and carries deposition and trial exposure. A rebuttal engagement starts from another expert's report. Billing is hourly against recorded time, never contingent on the result, and agreed in writing with counsel.

What I will not agree to

Not a fee contingent on the result. Not an opinion drafted toward a conclusion supplied in advance. Not testimony on the ultimate legal question, which belongs to the trier of fact and not to an examiner of advertising records.