No Single Metric Tells You What's Working

Everything below is free and useful before you read a word about who wrote it.

The executive problem

Walk into most quarterly marketing reviews and the pattern repeats: a dense deck of campaign metrics (click-through rates, cost per lead, pipeline sourced, last-touch conversion) sitting alongside sales results, with brand health and organizational capability mentioned, if at all, in a single slide near the end. Paul Magill and Christine Moorman, writing in Harvard Business Review, observed exactly this imbalance and proposed a "marketing road map" to correct it: campaign and sales metrics get measured constantly because they are easy to instrument, while brand building and capability building get measured rarely because they are harder to attribute and slower to show up in a dashboard. The result is a reporting culture that mistakes what is countable for what is important.

That imbalance would be a manageable problem on its own. It has been compounded by a second, more structural shift. For a decade, marketing organizations built their measurement stacks around deterministic, user-level tracking, a cookie or device identifier followed from ad impression to click to purchase, stitched into a tidy attribution path. Privacy changes have eroded that foundation. Apple's tracking limits, alongside browser-level restrictions on third-party cookies, have made it substantially harder to observe a single user's journey end to end. Julian Runge, along with Harpreet Patter and Igor Skokan of Meta Marketing Science, writing in HBR, describe the consequence plainly: single-method, user-level attribution is less trustworthy than it used to look, precisely because the data it depends on is disappearing. Executives who built their intuition on ten-year-old attribution reports are now making decisions on a measurement instrument that quietly stopped working the way they think it does.

The combined effect is a leadership team that is over-informed about campaign efficiency and under-informed about everything attribution cannot see, brand equity, capability building, and the growing share of the customer journey that no longer leaves a clean, user-level trail.

The contrarian thesis

There is no one-size-fits-all measurement solution. This is not a hedge or a caveat. It is the operating premise Google's modern-measurement guidance states outright, and it should reframe how executives think about measurement debates inside their own organizations. Attribution, incrementality testing, and marketing mix modeling each answer a genuinely different question. Attribution answers "which touchpoints did this converted customer pass through." Incrementality answers "how much of this outcome would not have happened without this specific spend." MMM answers "how do all our channels, together with external factors, explain aggregate sales over time."

Because these are different questions, using only one method does not just give you an incomplete picture. It gives you a systematically biased one. A marketing organization that reports only on attribution will consistently overstate the contribution of easily-tracked digital channels relative to brand and offline investment. An organization that reports only on incrementality tests will underinvest in channels that are hard or expensive to test but still contribute. An organization that reports only on MMM will have a directionally sound view of the whole but will lack the granularity to optimize an individual campaign next week. Each method, used alone, is not neutral. It has a lean, and that lean compounds every time it is used to make a budget decision.

The operating model

The practical fix is not to pick a favorite method and defend it in the boardroom. It is to understand the structural relationship between the three and use that relationship as a built-in sanity check. Google's measurement guidance is specific on this point: in-house attribution typically represents an upper-bound, generous view of marketing's impact, because it tends to credit the last or most visible touchpoint with outcomes that would often have happened anyway. Incrementality testing sits at the other end, representing a lower-bound, conservative view, because a well-designed holdout or lift test isolates only the causal effect of the specific campaign under test, stripping out everything that correlation-based methods leave in. MMM, calibrated correctly, falls within the range those two bookend.

That range is the operating model. Rather than treating attribution, incrementality, and MMM as three competing answers to be reconciled into a single false-precision number, treat them as a triangulation: attribution tells you the generous ceiling, incrementality tells you the conservative floor, and MMM tells you where, in aggregate, the truth is likely to sit. A leadership team that understands this range stops asking "which number is right" and starts asking "does our current spend make sense given the range these three methods describe." That is a fundamentally more honest question, and it is the one Runge, Patter, and Skokan point to when they describe MMM calibrated with ad experiments, lift tests, as an emerging gold standard: not MMM alone, but MMM checked against incrementality.

The framework

Attribution maps the customer journey and assigns credit to touchpoints along it. Its unique value is granularity: it shows the sequence of channels and content a customer interacted with before converting, which is invaluable for optimizing the mechanics of a funnel (creative, sequencing, channel mix within a campaign) even though the credit it assigns is correlational rather than causal, and increasingly incomplete as user-level tracking degrades.

Incrementality testing offers precise, causal data for specific campaigns and channels. By holding out a control group or randomizing exposure, it answers the question attribution cannot: would this outcome have happened anyway. Its unique value is causal certainty at the level of an individual test, at the cost of scope: a single incrementality test tells you about one channel or campaign, not your entire marketing system, and running enough tests to cover the whole portfolio is expensive and slow.

Marketing mix modeling provides a holistic overview of all media channels, sales, and external factors (seasonality, pricing, competitor activity, macroeconomic conditions) modeled together. Its unique value, as Runge, Patter, and Skokan emphasize, is that it works from aggregate, natural variation in the data and does not require user-level tracking at all, which makes it durable against the privacy changes that are degrading attribution. What it gives up in exchange is real-time granularity: MMM is typically refreshed periodically, not continuously, and works best for portfolio-level allocation rather than day-to-day campaign optimization.

No one of these three does what the other two do. That is the whole point.

The steps

Building a real marketing road map (one that shows effectiveness and efficiency together with brand and capability building, not campaign metrics alone) takes deliberate sequencing:

1. Inventory what you currently measure. List every recurring marketing report and tag each metric as campaign/sales, brand, or capability. Most organizations discover the brand and capability columns are nearly empty.
2. Separate outcome layers explicitly. Build the road map with at least three tiers: campaign efficiency (cost per outcome, conversion rates), campaign effectiveness (contribution to revenue and pipeline), and brand/capability building (awareness, consideration, team and process maturity). Report all three together, not the first tier alone.
3. Assign a measurement method to each layer. Campaign efficiency is well served by attribution. Campaign effectiveness needs incrementality testing to confirm causal impact. Brand and capability building are usually better served by MMM and periodic survey-based tracking than by either attribution or single-test incrementality.
4. Establish a testing cadence for incrementality. Pick a rolling set of channels and campaigns to holdout-test each quarter, prioritizing the channels attribution credits most heavily. Those are exactly the ones most likely to be overstated.
5. Refresh the MMM on a fixed schedule and calibrate it against incrementality results. Use lift-test outcomes as ground truth to adjust the model's channel coefficients, closing the loop between the aggregate view and the causal one.
6. Present the triangulated range, not a single number. In executive reporting, show the attribution ceiling, the incrementality floor, and the MMM estimate side by side for major channels, so the leadership team internalizes the range rather than anchoring on whichever number is most convenient.

Roles

Triangulated measurement fails when everyone assumes someone else owns the reconciliation. A workable ownership model looks like this:

• Analytics or marketing science owns MMM design, refresh cadence, and the incrementality testing program, the causal and aggregate methods that require statistical rigor and are least suited to being run informally inside a campaign team.
• Campaign and channel teams own attribution-level reporting for their own channels: the day-to-day, touchpoint-level view they need to optimize creative, targeting, and sequencing in near real time.
• A shared owner, typically a CMO, VP of marketing operations, or head of marketing analytics, reconciles the three views into one exec-facing report, is accountable for explaining discrepancies between them, and has the authority to adjust budget recommendations when attribution and incrementality disagree sharply on a given channel.

Without a named reconciler, the three methods sit in three separate reports, each team defends its own number, and the executive team is left to referee a methodology dispute it is not equipped to judge.

Metrics

Whichever method produced a number, the number itself has to clear a bar before it belongs in front of executives. Linda Popky, writing in HBR in 2015, set out three criteria that have not aged out of relevance even as the tools around them have changed: a good metric must be easy to understand and use, easily replicated, and actionable. Popky's warning is the durable part of her argument, worth repeating word for word:

Executives "rely on metrics that look good in a report, but in reality don't affect the organization's goals."

That warning applies with particular force to triangulated measurement, precisely because it introduces more numbers, not fewer. A road map with attribution, incrementality, and MMM results side by side is only useful if each number is legible to a non-specialist, can be reproduced by someone else running the same method, and points to a decision someone is actually prepared to make. A metric that is precise but unreplicable, or replicable but disconnected from any budget or strategy decision, is a vanity metric wearing a lab coat, and three methods make it three times as easy to hide one in the mix.

The checklist

Use this to audit your current measurement stack this week:

• Can you name, for each major channel, whether its performance number comes from attribution, incrementality, MMM, or an untested assumption?
• Does your reporting show brand and capability building at all, or only campaign and sales metrics?
• When was your last incrementality test run, and on which channel, and was it the channel attribution credits most generously?
• Is your MMM calibrated against real lift-test results, or running on its own assumptions unchecked?
• Does any single metric in your top-line report fail Popky's three-part test, understandable, replicable, actionable?
• Is there a named individual accountable for reconciling attribution, incrementality, and MMM into one view, or does each team defend its own number independently?
• If attribution and incrementality disagree materially on a channel's impact, does anyone notice, and does it change the budget?

The 30/60/90

Days 1 to 30, Inventory and baseline. Catalog every current measurement report and method in use. Identify which channels have no incrementality testing history and which metrics in current reporting fail the understandable/replicable/actionable test. Build the three-tier road map skeleton (efficiency, effectiveness, brand/capability) even if some tiers are initially empty.

Days 31 to 60, Launch the first triangulation. Run incrementality tests on the one or two channels attribution currently credits most heavily. Stand up or refresh an MMM using existing data, explicitly including brand and capability variables alongside media channels. Name the individual who will own reconciliation across methods.

Days 61 to 90, Institutionalize the range. Calibrate the MMM against the incrementality results from days 31 to 60. Present the first side-by-side exec report showing the attribution ceiling, incrementality floor, and MMM estimate for major channels, alongside the brand/capability tier of the road map. Set the recurring cadence (quarterly incrementality tests, periodic MMM refresh) that keeps the triangulation current rather than a one-time exercise.

The technology layer

Triangulated measurement depends on a precondition that is easy to overlook: knowing exactly what changed, when, and why, across every campaign and system involved. Incrementality testing in particular is only as trustworthy as the record of what was actually live during the test window (which creative ran, which audience was targeted, what changed mid-flight) because a causal claim collapses if the underlying campaign configuration shifted without anyone noticing. Software that keeps a reliable record of what changed across a campaign, the way Launch does, is what makes causal, incrementality-style measurement possible after the fact, since a test is only as good as the certainty that nothing else moved during it.

Frequently asked questions

Q: If I can only invest in one measurement method right now, which should it be? There is no correct single answer, which is the article's central point, but if forced to sequence, start incrementality testing on your most heavily-attributed channel, since it directly tests the number you are most likely over-trusting today.

Q: Why have privacy changes made attribution specifically less reliable, rather than all methods equally? Attribution depends on deterministic, user-level tracking to stitch a journey together; Apple's tracking limits and similar browser restrictions remove much of that user-level signal, while MMM works from aggregate data and incrementality works from randomized holdouts, so neither depends on the same signal being intact.

Q: How often should an MMM be refreshed? There is no universal cadence in the sourcing here, but the operating principle is consistency: refresh on a fixed schedule and recalibrate against incrementality test results each time, rather than treating the model as a one-time build.

Q: Does triangulating three methods mean tripling the measurement budget? Not necessarily: the goal is reallocating existing measurement effort across three methods in proportion to where each is strongest, rather than funding three complete, independent measurement programs.

Q: Is there evidence that this triangulated approach actually correlates with business results? Data cited by Google, drawing on Kantar and Measure Protocol research, found that C-level leaders who placed high importance on marketing mix models were over twice as likely to exceed revenue goals by 10 percent or more: one data point, not a guarantee, but directionally consistent with treating MMM as a core method rather than an afterthought.