Most marketing measurement tools are built on descriptive and diagnostic analytics - they explain what happened and, sometimes, why. They were not designed to answer the question marketers need answered: where should the next dollar go? That gap is structural. It exists because turning measurement into a budget decision requires causal modeling, saturation curves, and scenario planning that attribution reports, traditional MMMs, and incrementality tests were each built to avoid.
You have a 5:1 ROAS on Meta. You have an incrementality test that confirmed TikTok is working. You have a quarterly MMM deck sitting in your inbox from last week. You have more measurement than you've ever had.
And yet, you're sitting in the quarterly planning meeting, and nobody can answer the question finance just asked: "If we add $500k to the budget next quarter, where does it go?"
The room goes quiet. Someone opens a spreadsheet. Someone else suggests waiting for the next test cycle. The meeting ends without a decision, and the budget gets allocated roughly where it was last year, with a small increment toward whatever showed the best ROAS in platform last month.
This is what happens when a room full of measurement outputs collide with a question none of them were designed to answer.
Why don't attribution reports, MMMs, and incrementality tests answer the budget question on their own?
The measurement industry has given us three methodologies, each genuinely useful, each answering a different backward-looking question - and none of them answering the forward-looking one.
Attribution tells you which touchpoints appeared on the path to conversion. It's fast, granular, and gives you something to show in a Monday morning standup. What it doesn't tell you is whether those touchpoints caused the conversion, or whether the customer was already going to buy. A retargeting ad that shows up two minutes before checkout gets full last-click credit. The YouTube pre-roll that created the intent three weeks earlier gets nothing. Attribution measures correlation, not causation - and the channels that look best in attribution are often the ones closest to conversion, which can mean they're capturing demand that was created elsewhere. Attribution is a fast, useful signal - and it works best when read alongside full-funnel, impression-led data that shows what created that demand in the first place.
Marketing Mix Modelling solves some of attribution's structural bias. It operates on aggregated time-series data, incorporates impression-based channels, and captures effects that click-based tools miss entirely. But a traditional MMM runs on historical data, typically two or three years of weekly observations, and produces channel-level coefficients that reflect the past, not a forecast for next quarter. By the time a quarterly MMM report lands in your inbox, the media environment it was calibrated against may already be several algorithm updates out of date. And even when the outputs are accurate, they describe average historical contribution. They don't tell you whether additional budget in that channel will return $5 or $0.50 on the next increment of spend. Traditional MMMs are a strong foundation for strategic planning - and daily, granular measurement makes their insights more actionable between planning cycles.
Incrementality testing is the closest thing to ground truth. A well-designed geo holdout or conversion lift study gives you genuine causal evidence that a channel drove purchases that wouldn't have happened otherwise. But an incrementality test is a snapshot. It tells you that TikTok was incremental at the spend level you tested, in that window, against that audience. It doesn't tell you what happens when you double the budget, or whether the creative that was working six weeks ago is still working today. Tests typically take four to six weeks to read out - in a media environment where trends go viral overnight and algorithms shift weekly, that latency is a practical constraint for teams making daily decisions. Incrementality remains an essential validation tool; daily, always-on measurement is what makes it more usable between test cycles.
The common thread across all three: they were each built to measure what happened, and the forward-looking question requires something additional.
What's the difference between measurement data and a budget decision?
This is the question the industry mostly sidesteps, and it's worth naming precisely.
Measurement outputs - attribution credit, MMM coefficients, incrementality lift points - are inputs to a decision. They tell you what happened. The decision itself requires something different: a view of what will happen if you move budget, and by how much.
That requires knowing the shape of the diminishing-returns curve for each channel. A channel might show 5:1 average ROAS over the past quarter and still return less than $1 on the next dollar if you're already near saturation. Average ROAS and marginal ROAS are completely different numbers, and it's the marginal number that should drive allocation decisions. Most standard measurement outputs show average contribution rather than the marginal curve.
It also requires scenario planning: if we shift $200k from Google to TikTok, what's the modelled outcome? If we hold Meta flat and scale YouTube by 30%, what does blended ROAS look like? None of the standard methodology outputs answer that question directly. They describe what was, not what would be.
The gap between "here is what happened" and "here is what you should do" has a name in analytics: the gap between descriptive analytics and prescriptive analytics. Most measurement tools live on the descriptive and diagnostic rungs of that ladder. Prescriptive analytics, the kind that actually outputs a recommended action, requires causal-economic modelling, saturation curves, and a decision interface that translates all of it into something a team can act on. That combination is rare.
Why has the industry been slow to close this gap?
Three structural reasons, and they're worth understanding because they explain why the gap is unlikely to close on its own.
First, measurement systems were built for what's observable. Clicks, impressions, and conversions are loggable. Marginal returns and saturation curves require modeling forward-looking causal structure, which is a substantially higher technical bar. Reporting is easier to build, sell, and maintain than optimization.
Second, there's a commercial incentive misalignment. Ad platforms report attribution in ways that maximize their apparent contribution to conversions. Last-click models systematically over-credit the channels that sit closest to purchase - which tends to be the platforms running those models. The more accurate the measurement, the more it threatens the reported ROAS that justifies the spend. It reflects how platform-native models are designed to operate - measuring within their own data, without visibility into the full customer journey.
Third, even when better measurement exists, the interface problem is real. Research in decision-support systems has consistently found that models don't get used unless they're simple, transparent, and directly actionable. A quarterly MMM report that requires a data scientist to interpret rarely connects to what a performance marketer needs to decide on a Tuesday morning. The measurement has to be fast enough, granular enough, and clear enough that it connects to the decision being made - not just to the quarterly planning deck.
What does measurement that drives decisions look like?
The organizations making better budget decisions consistently share three characteristics.
They operate on marginal data, not average data. They know not just what their ROAS is, but where each channel's returns start declining. Saturation curves, which show how much incremental return each additional dollar produces before diminishing returns set in, are the foundational tool for allocation decisions. Without them, you're flying on averages that can be actively misleading.
Their measurement moves at the speed of their decisions. If your team is making budget calls weekly or reacting to creative performance daily, a monthly or quarterly measurement cadence isn't a slight inconvenience - it's structurally incompatible with how you operate. Daily, ad-level measurement isn't a nice-to-have for DTC and ecommerce brands - it's increasingly the baseline for teams making frequent budget calls in a fast-moving media environment.
They've closed the loop between insight and execution. A dashboard that produces a recommendation is still only half the system. The other half is whether that recommendation changes what happens in the campaign manager. Brands that achieve meaningfully better outcomes tend to be the ones where measurement feeds directly into budget planning and automated execution - not the ones where it produces a report that sits between analysis and action.
Gymshark used Fospha's daily measurement and automation with Smartly to rebalance spend toward higher-performing campaigns, driving 39% higher ROAS on TikTok and enabling faster, more confident budget decisions during peak periods. That outcome wasn't driven by a better dashboard. It was driven by measurement trusted enough to act on, fast enough to keep up with daily decisions, and connected directly to execution.
What should performance teams ask their measurement vendor?
Three questions that separate descriptive measurement from prescriptive:
Does it show marginal ROAS, or just average ROAS? If a vendor can only tell you what a channel returned on average last quarter, they're showing you the past. What you need is a view of what the next dollar in that channel will return, and whether you're approaching saturation. If they can't show you the curve, they can't help you make the allocation decision.
How often does it update, and at what level of granularity? Monthly or quarterly cadence is incompatible with weekly or daily budget decisions. Meaningful measurement for ecommerce teams updates daily and goes deep enough to inform decisions at the campaign or ad level - not just the channel.
What does the output connect to? Insight that stops at a dashboard is only half the system. Ask how the measurement connects to planning and execution workflows. Whether it feeds scenario planning. Whether it powers automated budget adjustments. The answer tells you whether the vendor is selling you a reporting tool or a decision system.
And then what? Three implications for how you think about measurement
Rethink what your measurement is optimizing for. If it's optimizing for the most accurate historical report, that's a different product than one optimized for the most useful forward-looking decision. The two aren't the same, and confusing them is expensive. Most teams have very accurate reports of what happened and very little clarity on what to do next. Reframe your measurement requirements around the decisions you need to make, not the reports you need to produce.
Treat last-click ROAS as a signal, not a source of truth. A 2019 study across 15 large Facebook advertising experiments found that observational attribution methods overstated true ad effectiveness - in half of the studies, the estimated lift was off by a factor of three. eBay's landmark paid search experiment measured a true ROAS of roughly -63%, against what attribution methods had suggested was a working channel. These aren't edge cases. They're a consistent finding that channels sitting close to conversion tend to look better in attribution than independent measurement suggests. Build your budget process accordingly.
Ask what your measurement would tell you to change. This is the practical test. If your measurement outputs are broadly confirming your existing allocation - if they're not pointing to anything you should do differently - they're probably not surfacing the marginal data that drives real decisions. Good measurement should occasionally challenge your assumptions - pointing to the channel you're over-investing in, or the one you're systematically under-crediting because the clicks don't show up.
If your measurement isn't regularly changing what you do, it's reporting cost - not competitive advantage. Understanding what measurement is built to answer, and what it isn't, is the first step toward building a system that drives budget decisions.
To see how Fospha's daily, impression-led measurement and forward-looking forecasting connects insight to execution, explore how the Measurement OS works in practice
Frequently asked questions
Why does attribution show high ROAS for channels that incrementality tests say are barely working?
Attribution measures correlation - it assigns credit to touchpoints that appeared on the path to conversion, not the ones that caused it. Channels like branded search and retargeting sit close to purchase, so they collect a disproportionate share of last-click credit even when many of those conversions would have happened through organic channels anyway. Incrementality testing measures the causal lift, conversions that wouldn't have occurred without the channel, which is why the numbers often diverge significantly.
What is the difference between average ROAS and marginal ROAS?
Average ROAS is what a channel returned on total spend over a given period. Marginal ROAS is what the next additional dollar in that channel will return. A channel can show 5:1 average ROAS while the next dollar returns less than $1, because you're near saturation. Budget allocation decisions should be based on marginal ROAS and saturation curves, not averages - but most measurement tools only surface the average.
Why do traditional MMMs arrive too late to be useful for ecommerce teams?
Traditional MMMs are built on two to three years of historical data and typically refresh on a monthly or quarterly cycle. Ecommerce teams often make budget calls weekly or daily, reacting to creative fatigue, algorithm shifts, and inventory changes. A quarterly model often arrives weeks after the data was collected, measuring a media environment that may no longer exist. For decisions made at that cadence, measurement needs to update daily and be granular enough to inform campaign-level calls.
Can incrementality tests replace ongoing measurement?
Incrementality tests provide valuable causal validation, particularly for major budget decisions in your largest channels. But they're episodic - they describe performance at one spend level, in one time window, and take four to six weeks to read out. They can't run continuously across every channel and every campaign without significant revenue cost and operational complexity. The most effective approach combines daily, always-on measurement for continuous optimization with periodic incrementality tests as strategic validation for significant budget moves.
What does prescriptive analytics mean in a marketing context?
Prescriptive analytics goes beyond describing what happened (descriptive) or explaining why (diagnostic) to recommending what to do next. In a marketing context, it means surfacing where incremental spend will drive returns given current saturation levels, modelling what-if scenarios for budget reallocation, and feeding those outputs directly into planning and automated execution. Most measurement tools stop at descriptive or diagnostic. Prescriptive requires saturation curves, scenario planning, and a decision interface that connects insight to action.
Why do platforms report different ROAS numbers than independent measurement?
Ad platforms use last-click or platform-native attribution models that attribute conversions to their own touchpoints. They're measuring inside their own data - they can't see what happened on other platforms, through organic search, or across marketplaces. Independent measurement, particularly MMM-based approaches, uses aggregated business data to model actual contribution rather than claimed credit. The gap between platform-reported ROAS and independently measured ROAS is often significant, particularly for channels like retargeting and branded search that intercept demand already created elsewhere.
What's the minimum measurement cadence a DTC brand needs to make weekly budget decisions?
Daily updates at the ad level. Weekly budget decisions require measurement that reflects the current state of each channel - not data from three weeks ago. That means daily model refreshes, granularity at the campaign and creative level, and outputs connected to planning so insights can be acted on without a manual translation step between report and decision.
