PPC Damages and Lost Profits

Two claims that keep getting merged, and should not be

Damages in a paid-search dispute usually arrive as one number when they are really two claims. Wasted spend asks how much money went into the account that should not have gone in: a misconfigured campaign, spend after a directive to pause, invalid traffic, spend outside the intended targeting, fees charged on work not performed. Lost profit asks what the business would have earned had the program run properly — which requires a counterfactual, and the counterfactual is where these cases are decided.

This page carries the verdict it does because the second column is modeled and never measured. Nothing in an ad account records what would have happened. A lost-profits figure is an argument built on assumptions, and the useful work is naming them first.

What the account record establishes on its own

Before any model is built, separate what the platform record shows from what has to be inferred. Within its retention window, an ad account can establish:

• Spend. Cost by day, campaign, ad group, keyword and ad, at daily granularity for 37 months as of June 1, 2026, and as monthly and coarser aggregates for much longer.
• Delivery. Impressions, clicks, cost, where ads served, on what devices, against queries that cleared Google's volume threshold.
• Configuration. What was live: targeting, match types, budgets, bidding, negative keywords, exclusions.
• Change. Who made which change and when, under the email attached to the account, for as long as the window holds it.
• Platform adjustments. Invalid-click counts and the credits issued.

That list is the strong part of a case. It shows that spend continued after an instruction to stop, that budget went somewhere the agreement did not contemplate, that a conversion tag was broken, that fees were charged in a month the account shows no work. What it does not contain is margin, profit, or any fact about what a customer would have done differently. Spend and delivery are in the account. Profit is not.

The three lost-profits methods, and what each one assumes

Commercial damages practice recognizes three classical methods of proving lost profits. A paid-search expert normally supports these rather than performs them, and each rests on an assumption the account record can support or undermine.

Before-and-after

Compare the plaintiff's performance before the wrongful act to its performance during and after, and treat the shortfall as the loss. The assumption: that the earlier period is a legitimate control — comparable season, auction, site, offer and measurement.

Yardstick, or benchmark

Compare the plaintiff to a comparable but unaffected entity and treat the divergence as the loss. The assumption: that the comparator is genuinely comparable, which in paid search means similar auctions at similar volume with similar measurement in place.

Market model, or market share

Model the total market and the plaintiff's expected share of it absent the wrongful act. The assumption: that both can be modeled from evidence rather than asserted.

A fourth appears in contract cases: the contract-terms method, where the agreement specifies what the plaintiff was entitled to — a cost-per-acquisition target, a minimum lead volume, an allocation of spend — and damages flow from the gap between promised and delivered. Often the most tractable in a paid-search matter, because the promised figure is written down. Its weak point is next.

The measurement change hiding inside a before-and-after model

This is the failure I find most often and the one least often pleaded around. A before-and-after model, and a contract target keyed to conversions, both assume the two ends of the comparison are the same measurement. Frequently they are not.

• The conversion action may have been redefined. Counting can move from one conversion per click to every conversion, and an action can be added to or removed from the primary column. Either moves the number without anything changing in the business.
• The attribution model may have switched. Four rules-based models were retired in 2023 and affected conversion actions were upgraded to data-driven attribution, which Google is reported as saying typically produces about a 6% increase in reported conversions.
• Modeling may have begun or stopped. Modeled conversions flow into every report using a conversion column, and whether an account receives them turns on conditions the advertiser does not control.

Where any of that happened mid-period, part of the delta between the before number and the after number is an artifact of how the numbers were produced. A model assigning the whole delta to the defendant is measuring two different things and calling the difference a loss. Whether the measurement held constant is squarely a paid-search question, and one of the few places the account record can dismantle a damages theory outright.

The but-for problem, specifically in paid search

Even with the measurement held constant, a paid-search program is not a closed system. Five forces move performance without any party doing anything:

• Seasonality. Search demand in most categories is strongly seasonal. Comparing a peak quarter to a trough quarter unadjusted produces a number that is mostly calendar.
• The auction. Cost per click is an auction outcome; the advertiser does not set the price. A competitor raising bids moves cost per click and impression share independently of anything the defendant did. Auction Insights is one of the few native ways to test whether the competitive set changed, though it never shows a competitor's bid or spend.
• Entry and exit. A well-funded new competitor in the same auctions changes the achievable cost per acquisition. A market fact, not a breach.
• The platform's own changes. Match-type behavior, campaign type rollouts, automated bidding defaults, auto-apply recommendations and attribution changes shift measured performance on Google's schedule.
• The advertiser's own changes. Price changes, stock-outs, a site redesign or a change to the offer all move conversion rate, and none appear in the ad account.

A model that does not address these overstates, because it assigns the entire difference to the defendant. The standard response is to name one confound and argue the model cannot separate it from the alleged conduct — usually available, which is a reason to build expecting it.

What courts have said about damages models that ignore the market

Honesty first. No opinion has been located that admits or excludes damages testimony in a paid-search dispute specifically, and none treating modeled conversions or attribution modeling as a reliability question. This is new ground; what follows are analogues.

In Concord Boat Corp. v. Brunswick Corp. (8th Cir. 2000), an economist's model of an engine market was held wrongly admitted because it did not incorporate all aspects of the economic reality of that market and did not separate lawful from unlawful conduct; the court said it ignored inconvenient evidence, including a recall and a merger both sides conceded were unrelated to the challenged conduct. The but-for problem, stated by a court (opinion text, read August 14, 2026).

In Target Market Publishing, Inc. v. ADVO, Inc. (7th Cir. 1998), a lost-profits projection for a direct-mail advertising venture assumed expansion into dozens of zones where no sales effort had begun, at full price — while the one test market's advertiser base had fallen from fifteen participants to three. The testimony was excluded for too great an analytical gap between the data and the opinion: the template for attacking a model built on assumed scaling (opinion text, read August 14, 2026).

A third line runs through franchise advertising-fund litigation, where a class-wide lost-profits calculation built on fictional average franchisees was rejected as speculative: a model that averages across advertisers, accounts or outlets rather than proving loss for each is vulnerable on the same ground. I read these in online reproductions, and confirm citations against the official reporter.

Benchmark data, and how much weight it can carry

A yardstick model reaches for an external comparator, and in paid search what is on offer is a published benchmark report. I keep these out of tables so they do not acquire unearned authority.

Take the most widely circulated example. WordStream's 2026 benchmark report states its method: a sample of 13,474 search advertising campaigns running between April 1, 2025 and March 31, 2026, each subcategory including at minimum 52 unique active campaigns. Its all-industry averages are a 6.64% clickthrough rate, a $5.42 cost per click, an 8.18% conversion rate and a $66.69 cost per lead (read August 14, 2026). WordStream is a division of LocaliQ, part of Gannett, which sells managed paid-search services, and the sample comes from its own client and partner base rather than a probability sample.

That does not make the figures dishonest. It makes them a convenience sample published by an interested party: useful as a sanity check, weak as the foundation of a damages number. An achievable cost per acquisition is set by the auctions that advertiser competes in, its own offer and its own conversion path, so a shortfall against a heterogeneous average measures nothing anyone lost. A benchmark used in a matter should carry its publisher, sample, date and commercial interest on the face of the report.

Double counting, and why wasted spend is rarely wasted in full

Two arithmetic problems undo otherwise sound claims.

Double counting. Recovering the wasted dollars and also the profit those same dollars would have produced if spent correctly can compensate the same loss twice, depending on how the theory is framed. The two are frequently pleaded together and need reconciling before a number is built, not after an opposing expert points at it.

Gross versus net. Spend on a badly targeted campaign still bought clicks, and some converted. A wasted-spend figure asserted at gross is the easiest number in a case to attack, because the account record usually shows the return. The defensible version is net of the value the spend produced, and the account supports that directly: cost against clicks, conversions and conversion value for the campaigns and dates in issue.

The same trap sits in invalid-traffic claims: adding a detection vendor's flagged percentage to the platform's invalid-click credit can count the same traffic twice, because the vendor measures after the click lands and the platform filters before billing.

Where my work ends and a damages expert's begins

A paid-search expert who volunteers a lost-profits number is working outside the field. Quantifying lost profits belongs to a forensic accountant or an economist, and qualification gets raised before method. The division of labor that holds up:

• Did the account run as alleged? Change history, settings, spend by day, delivery by location and device.
• Was the spend recoverable? Invalid traffic, targeting, match type, brand versus non-brand, and what each returned.
• Is the comparison period a legitimate control? Season, competitive set, site and offer changes.
• Are the conversions the same kind of thing at both ends? Conversion action definitions, attribution model, the onset of modeling.

Those are factual predicates. They make a damages model stronger or take it apart, and they are answerable from records rather than assumptions. The number itself belongs to somebody else, and the admissibility of expert testimony is a separate discipline this site hands off rather than analyzes; Federal Rule of Evidence 702 was amended effective December 1, 2023 and is published in full by the Legal Information Institute (read August 14, 2026). Findings may be challenged like any others. What I can offer is predicates that are documented, sourced and reproducible from the same export anyone else can examine.

Frequently Asked Questions

Published 2026-08-14