In 2016, Facebook told advertisers it had been reporting average video view time incorrectly — overstated by 60 to 80 percent — for roughly two years. That number had helped justify a genuine, industry-wide shift of advertising budgets toward video content. By Facebook’s own account, the error wasn’t deliberate manipulation. It was a miscalculation in methodology. It went uncorrected for two years anyway, because almost nobody outside the platform was in a position to independently check it. A class-action lawsuit followed. Facebook eventually settled for a reported $40 million, without admitting wrongdoing — and without most of the advertisers who had shifted strategy on that number ever receiving a clear account of how much of their two years of decisions had rested on it.
Around the same period, the Association of National Advertisers commissioned an independent investigation into U.S. media-buying practices. What it found was less about a single error and more about a structural pattern: widespread, non-transparent rebate arrangements between media agencies and media suppliers — arrangements clients were frequently never told about, and that shaped which channels and placements those clients were being advised to buy.
Neither of these stories is really a story about fraud, and treating them that way misses the more useful lesson inside both of them. What they actually reveal is a structural condition economists have a name for: the principal-agent problem. It shows up any time one party — an agent — is hired to act on behalf of another — a principal — whose interests aren’t perfectly aligned, and who can’t fully observe what the agent is actually doing on their behalf. A marketing relationship is a textbook example of this condition, whether anyone in the relationship uses that language or not.
In marketing specifically, this rarely looks like lying. It looks like convenient metric selection, which is a much harder thing to catch and a much easier thing to justify to yourself. An agency reporting on a campaign where conversions came in soft is unlikely to fabricate a number. It’s far more likely to lead the report with impressions, reach, or engagement instead — all genuinely true, all measured correctly, and all quietly chosen because they happen to be the numbers that look best that month. This isn’t a one-time distortion. Repeated across many campaigns, this habit doesn’t just shape a single report — it gradually shapes what gets tested, recommended, and built going forward, tilting toward whatever is easiest to report favorably, regardless of whether it’s actually the thing moving the business.
The newer wave of AI-generated reporting dashboards adds a layer to this problem rather than solving it, which is worth naming plainly because it’s often assumed to do the opposite. A dashboard that auto-generates a polished, narrative summary from raw platform data inserts an additional layer of abstraction between a business owner and the underlying numbers. A business owner reading a clean, well-formatted AI summary often has less visibility into which numbers were selected and why than someone looking at a raw, unfiltered export — the summary just feels more authoritative because it’s better formatted, not because it’s more transparent.
None of this is an argument for treating every agency or marketing partner with suspicion. That would be its own kind of mistake, and an expensive one — most of these dynamics happen without anyone intending to mislead anyone. It’s an argument for structure instead of trust as the primary safeguard, because structure works regardless of anyone’s intentions. A few practical habits meaningfully close this gap without requiring a business owner to become a data analyst: requesting direct, read-only access to raw platform data rather than only ever accepting a summarized report; where budget allows, separating the party buying media from the party independently evaluating how well it performed; and anchoring reporting conversations around outcomes tied to revenue — qualified leads, booked conversations, closed deals — rather than metrics that are comparatively easy to make look favorable regardless of the underlying result.
The simplest test to run this week doesn’t require new software or a new vendor relationship — call it the Source Check. Pull up the last marketing report received and ask, plainly: where did these specific numbers actually come from — pulled directly from the platform, or summarized first by the same team being measured by them? If that answer isn’t immediately obvious from the report itself, that’s the gap worth closing before the next one arrives — not the report after that, this one.