f your model ends in a deck, it is a report card. It tells you how you did. It does not tell you what to do Monday morning, and it never answers for what it recommended. A decision system does both. It builds a weekly buying plan, forecasts your probability to goal, and then shows what changed and why.
At CIMM Summit XV, I joined a panel on “The Future of MMM and Agentic AI,” moderated by Angelina Eng, former head of measurement at IAB, with leaders from across the martech space. We disagreed on plenty, but the one thing nobody pushed back on was that most measurement still ends up as a report when it should be driving a decision.
What the panel agreed on
The old model was a static report card. A team built it over months, a partner delivered a long deck, and it sat on a shelf until the next refresh. The panel agreed that is too slow and too expensive to be the only use case. We also agreed that faster runs and more transparency are making MMM practical for more brands, and that getting people to trust a model depends as much on the humans as on the math.
Nobody on stage disagreed about the problem. We differed on what actionable means. My answer is simple. A model is actionable when a team can act on it this week, and when you can later see whether acting on it worked.
Why smart teams stay stuck
This is not a talent problem. In a Harvard Business Review Analytic Services report on MMM actionability, 87% of marketers said MMM matters, yet only 28% could turn it into timely action. Our own survey found that about four in ten media plans get redrawn to match leadership expectations rather than the evidence.
The model says one thing. The room wants another. The room wins, and nobody finds out who was right. That happens because the model has no memory and no consequences. It publishes and moves on.
Accountability comes from reconciliation
A forecast you never check is a guess with a nice chart. Every plan should be reconciled against actual results, the real test of forecast accuracy versus a causal MMM. When the two differ, you should know why. Maybe the team did not execute the plan. Maybe the market shifted. Maybe an assumption was wrong. Each answer points to a different fix.
This is also how a model improves. Priors built from real transactions and test results give it a sound starting point. Each week of forecast versus actual tightens it. A brand does not have to go dark to learn something. It learns from the decisions it is already making.
It matters most for upper-funnel channels like CTV, audio, and podcast, where effects build and decay over time. You cannot validate long-term decay with a one-time study. You validate it by forecasting, reconciling, and doing it again.
Reconciliation is also how trust gets built. You make a forecast, you check it, and you say what you found. Do that every week and the argument over whether to trust the model mostly goes away. It gives marketing and finance a shared language too: here is what we forecast, here is what happened, and here is why.
The leak nobody tracks
Teams argue about model accuracy. Few track adoption, the share of recommendations that actually get executed.
Picture a team that follows 40% of the plan. A strong model followed 40% of the time loses to a good model followed 80% of the time. The loss is not in the math. It is in the handoff between the system that builds the plan and the system that buys the media.
Closing that gap is the real work. The plan has to be executable, not just insightful, and adoption deserves the same scrutiny as any metric you care about.
What it costs to ignore the plan
In our 2026 benchmarks, drawn from more than 400 brands and $42 billion in media spend, flighted media returned less than a dollar at the margin. With optimized timing and allocation, the same media returned $5.84. Linear TV moved from $1.26 to $6.64.
That is not a modeling gap. That is a recommendation that got seen and set aside.
Brand, creative, and performance in one view
The panel also raised creative, which can account for a large share of ROI and still goes unseparated from media in most analyses. That is a decision-quality problem. If you cannot see which part of the result came from the message and which came from the buy, you cannot fix the right thing.
The same goes for brand and performance in another. Many teams track both. Fewer can connect them. When brand lives in one tool and performance in another, brand becomes the line item you defend and performance becomes the one you cut to. Measured together, brand shows up as the multiplier on everything below it.
What humans still own
People ask whether the human leaves the loop. The better question is which decisions a human still needs to make. Strategy, brand direction, and risk tolerance stay with people. Routine allocation across channels and weeks can move toward the system as its forecasts keep landing. Autonomy should be earned, and reconciliation is how it gets earned.
A test for your own team
Pick one decision you made last quarter. What did the model recommend? What did you do? What happened? What will you check next month to know you were right?
If you cannot answer all four, you have a report card. If you can, you are running a decision system.
Thank you to Angelina and the CIMM team for hosting, and to Claire, Josh, Marc, and Mike for a debate that was honest about what is still hard.
Which decision would you want reconciled first?
