Marketing Efficiency Ratio: A Defense of the Blunt Metric

What the marketing efficiency ratio actually measures

The marketing efficiency ratio is one of the simplest metrics in demand generation. Take all the revenue a business earned in a period. Divide it by everything spent to earn it, across paid search, paid social, retail media, and every other line. The quotient is MER, sometimes called blended ROAS. If a business earned two hundred thousand in revenue against fifty thousand in total marketing spend, its MER is four. There is no model inside that number, no window, no credit rule, no assumption about which ad a buyer saw first.

That absence is exactly what distinguishes it from channel-level ROAS. A platform-reported ROAS answers a narrower and more flattering question: of the conversions this channel can associate with an ad it served, what was their value against this channel spend. Every channel in the account answers that question about itself, in isolation, and the answers routinely sum to more revenue than the business actually earned, because the same sale is claimed by more than one channel. MER cannot double-count, because it never counts by channel at all.

What is MER in marketing, and why it exists

MER emerged from practice, not from the academic literature. It was popularized by measurement vendors and operators who watched channel dashboards report healthy returns while the business bank balance told a different story. The gap between a portfolio of green channel scorecards and a flat or shrinking contribution line is the specific problem MER was built to expose. Industry explainers from measurement platforms describe it in the same terms: an anti-vanity anchor that a marketer cannot game by rearranging attribution, because it sits above attribution entirely.

It should be tiered by what the evidence actually supports. MER is an industry-originated metric, not a peer-reviewed construct, and its status here is emerging rather than established. What makes it defensible is not its provenance but its logic, and that logic is directly supported by the established experimental literature on how correlational, channel-level advertising metrics overstate what advertising actually caused. The rest of this piece grounds the blunt metric in that literature.

The gaming that channel-level ROAS invites

A metric is gameable when the party being measured can improve the number without improving the outcome. Channel-level ROAS fails this test in a structural way. Because attribution assigns a single sale to a channel according to a credit rule, and because most default reporting still leans on last-click or platform-native attribution, revenue can be moved between channel reports by changing which channel is positioned to catch the final click. Spend can be steered toward the audiences and search terms a channel would have won regardless, harvesting conversions that were already going to happen and booking them as return.

This is not a hypothetical failure mode. It follows from the mechanism design of the auction the spend flows through. The generalized second-price auction that underlies sponsored search has no dominant-strategy equilibrium, which means optimal bidding requires modeling competitors rather than declaring true value, and the platform optimizing your bids is optimizing toward the reported conversion, not toward incremental profit. When the scorekeeper and the player are the same automated system, the score drifts toward whatever is cheapest to book. A blended metric removes the board on which that game is played.

The evidence that channel metrics overstate: activity bias

The strongest academic foundation under the case for a blended metric is the study of activity bias. In a set of three controlled experiments, Lewis, Rao, and Reiley showed that online behaviors are far more correlated over time than intuition suggests: a person who is browsing at a given moment is also more likely to be searching, clicking, and buying at that same moment, advertisement or no advertisement. An observational metric that credits an ad for a conversion cannot distinguish the ad from this background correlation, and so it overstates the ad effect, sometimes dramatically.

This is the mechanism that makes channel-level ROAS untrustworthy as a causal number, and it is established, not emerging. Every channel report is an observational estimate of exactly the kind activity bias inflates. MER does not solve activity bias at the level of the business as a whole, and no simple ratio can. What it does is refuse to launder the bias into channel-by-channel decisions, because it never expresses a per-channel causal claim in the first place. A business steering by MER is at least steering by a number that reconciles to its own revenue.

The eBay experiment: when a channel takes credit it did not earn

The most direct evidence that a channel can report strong returns while causing almost nothing comes from a large-scale randomized field experiment at eBay. Blake, Nosko, and Tadelis found that paid search advertising on the company own branded and trademark keywords produced no measurable short-term incremental benefit: the buyers who clicked those ads would overwhelmingly have arrived through organic listings anyway. For non-brand keywords the picture was split by consumer type, where infrequent and new users were positively influenced but frequent users, whose purchases were unaffected by the ads, absorbed most of the spend, driving the average return negative.

Read that result against a channel dashboard. Branded paid search would have shown a high ROAS throughout, because it was catching conversions and booking them as its own. The experiment revealed those conversions to be substantially non-incremental. This is the clearest possible illustration of why a channel-level number can be simultaneously accurate as a report and wrong as a decision input. It is a single-firm study and should be treated as a documented mechanism rather than a universal constant, but the mechanism is general: the more correlated a channel exposure is with pre-existing intent, the more its reported return overstates its causal contribution.

Blended ROAS vs ROAS: what the blunt metric buys you

Set the two side by side. Channel-level ROAS is precise, granular, and gameable, and it answers a question about a channel that no owner ultimately cares about in isolation. Blended ROAS, which is MER, is coarse, portfolio-wide, and resistant to reallocation games, and it answers the question an owner actually holds: for every dollar this business put into marketing, how many dollars of revenue came back. The blunt metric buys three things the granular one cannot.

First, it is arithmetic that reconciles to reality: MER cannot claim more revenue than the business booked, so it cannot drift into the fiction that a stack of channel reports can. Second, it captures the halo, the effect where paid social lifts branded search, where a display impression seeds a later organic visit, where the channels help each other in ways no single-channel credit rule can see. Because MER counts total revenue against total spend, those cross-channel effects are inside the number automatically rather than fought over between dashboards. Third, it is cheap and immediate, computable every week from two figures a business already has, with no attribution modeling or tracking reconstruction required.

Incremental ROAS: what the blunt metric cannot do

A defense of MER that oversold it would defeat its own purpose. MER is a guardrail, not a proof. It tells you whether the whole marketing operation is efficient in aggregate. It does not tell you which channel caused what, it does not isolate incrementality, and it moves with everything else in the business, so a strong month of organic demand or a seasonal surge will flatter MER while a soft market will depress it regardless of how well the ads performed. Reading MER as if it were a causal channel verdict repeats, in reverse, the exact error it was built to correct.

The measurement of causation lives in experiments, not ratios. Incremental ROAS, the return measured against a real counterfactual, requires holding something back and comparing. Geo experiments randomize non-overlapping regions into treatment and control to recover true causal lift without individual tracking, a method developed and documented by Vaver and Koehler at Google and since extended. Ghost ads, developed by Johnson, Lewis, and Nubbemeyer, record the would-be ad impressions a control group would have seen, letting an advertiser measure incrementality at a fraction of the cost of older public-service-announcement holdouts, and demonstrated a retargeting lift of 17.2 percent in site visits and 10.5 percent in purchases in the study that introduced them. MER is the standing weekly guardrail. These experiments are how you learn what actually caused the return. The two are complements, and a serious operation runs both.

How to read the marketing efficiency ratio

A blunt metric earns its keep only when it is read with discipline. The following practices keep MER a truth-teller rather than a new vanity number.

• Report MER as the top-line efficiency figure and channel-level ROAS as a diagnostic beneath it, never the reverse. The blended number governs; the channel numbers explain movement within it.
• Anchor a target MER to your own contribution margin, not to a borrowed benchmark. There is no universal good MER, because the efficiency a business needs depends on what it keeps after cost of goods and delivery. A published industry figure is not a target.
• Watch the trend, not a single reading. A one-month MER is noisy and confounded by demand and seasonality. The signal is the direction over a stable window against a recorded baseline.
• Use MER to detect a problem and experiments to diagnose it. When MER falls, a geo holdout or a channel-off test tells you what actually caused the drop; the ratio alone cannot.
• Keep the caveats attached. Never present MER, or any return figure, as a naked number without the spend, the window, and the method behind it. A metric without its provenance is a claim, not evidence.