Every marketing organization asks two questions: which channels drove last year’s sales, and what will next year’s budget allocation return? They sound the same. They’re not. The first is measurement: it looks backward and assigns credit. The second is planning: it looks forward and needs a return on each decision you are about to make.
The popular open-source marketing mix modeling (MMM) tools, Google’s Meridian, Meta’s Robyn, and PyMC-Marketing, were built to answer the first question. KeenOS was built to answer the second. The rest of this post explains the difference and why it matters.
Key highlights:
- Meridian, Robyn, and PyMC-Marketing measure what happened: they assign credit to the week a sale occurred.
- None of them report by when the money was spent, so they can’t tell you what a specific budget decision returned, now or in the future.
- KeenOS tracks every dollar as an investment with a lifetime return, the way an investor values a portfolio instead of an accountant closing the books.
- Marginal ROI, what your next dollar returns, is not the same as average ROI, what your typical dollar has returned. Most tools report the second when you need the first.
- rands running on KeenOS plans see an average 25% incremental revenue lift, validated against $45B in marketing investment optimized across 450+ models.
What the open-source leaders get right
If you’re evaluating MMM options, these three meet the bar:
- They measure impact. Meridian’s core idea is a clean thought experiment: simulate your sales with a channel’s spending, simulate them without it, and call the difference that channel’s contribution. It’s a transparent definition of impact.
- They model carryover. All three know advertising doesn’t pay back in the week you spend it, and they build “adstock” (the slow decay of an ad’s effect over the following weeks and months) directly into the model.
- They check themselves against experiments. Meridian and PyMC-Marketing can compare the model’s answers with real-world tests, which keeps the model honest.
- They’re free, open, and well documented. A capable data science team can stand one up and get credible answers about the past.
That’s the code, and the code is good. Getting a robust answer out of it is a different story. Each setup asks the team for dozens of judgment calls: how fast each channel’s effect fades, what shape diminishing returns take, which of hundreds of candidate models to keep. Turn those dials differently and the same data gives different answers. And even with every dial set well, there’s a question these tools never ask.
The question they don’t ask
All three report contribution by when the sales happened: “this week’s revenue came 12% from search, 8% from TV,” and so on. None of them report by when the money was spent: “the budget you invested in March returned 2.1x over its lifetime.”
The whole difference between measurement and planning lives in that gap. When you plan a budget you aren’t allocating credit to weeks on a calendar. You’re making a series of spending decisions, and each one deserves its own answer: what did it return? Not “what happened in March” but “what did March’s money do,” in March and in every month after.
| Meridian, Robyn, PyMC-Marketing | KeenOS | |
| Credits assigned by | The week the sale happened | The week the money was spent |
| Answers | What drove this week’s sales | What this dollar returns, now and later |
| Carryover (adstock) | Modeled, but credit stays on the calendar week | Modeled, and credit follows the money across its full payback |
| Budget optimizer | Splits one budget across channels | Compares full multi-period plans against each other |
| Team required to run it | In-house data science team, ongoing | Ships with the model built and maintained |
Put simply:
- Measurement asks: where did this week’s sales come from?
- Planning asks: what is this week’s spending worth to the future?
The open-source tools answer the first question and stop. KeenOS answers the second, and that requires a different accounting system, not just an extra report.
An analogy we find useful: an accountant and an investor can look at the same portfolio. The accountant tells you what income arrived this month. The investor tells you what each purchase has earned since you made it, and what it’s still likely to earn. Both are correct. Only one of them helps you decide what to buy next.
Three Things KeenOS Does Differently
1. Built for planning, not just reporting.
Every unit of spend is tracked as an investment. We trace each week’s future sales to the causal spending decisions, so every period of spend in every channel gets its own lifetime return. That makes plans comparable. You can lay out two budget scenarios with different channels and different timing, and compare them on the same footing: projected return per decision.
The open-source tools do offer budget optimizers, and they’re useful. But those optimizers answer one question: given a budget, how should it split across channels? They don’t evaluate a plan, meaning a sequence of spending decisions over time, where timing matters and this quarter’s spend changes what next quarter’s is worth.
2. Impact measured the way markets actually behave.
Out of the box, most tools treat marketing as additive: an ad adds a fixed number of sales on top of everything else. Same ad, same lift, whether it runs in December or in the dead of summer.
Real markets don’t work that way. Marketing multiplies demand. The same campaign is worth more when demand is high and when the rest of the plan has built momentum, and channels support each other. KeenOS models it that way. The practical consequence: our estimate of what a campaign did, and what it would do if you moved it, reflects the market conditions it actually ran in rather than an average week that never existed.
3. We count the impact you can’t see yet.
This one matters most, and it’s the easiest to miss. Advertising keeps working after the reporting window ends. A brand campaign in November is still generating sales next spring. That is exactly what carryover describes.
The open-source tools stop counting at the edge of the data. Spend near the end of the window looks artificially weak simply because most of its payback hasn’t arrived yet. The money isn’t underperforming; the report is undercounting it.
For planning the open-source tools get it exactly backwards, because when you plan you are always at the edge of the forward-looking window. The decisions you care about most (next month’s, next quarter’s) are precisely the ones whose returns haven’t shown up yet.
KeenOS puts a value on that unseen tail. When the model says an ad’s effect decays over months, our accounting follows through: we project the remaining payback and put today’s value on it, the way an analyst values expected future cash flows. It is an estimate, and we label it as one. But an estimate of the tail beats pretending it is zero.
As time unfolds, KeenOS then uses the actuals to reconcile the forecast. And the system learns over time and improves its estimates of that unseen tail. The result is that spending gets judged on what it will do now and into the future, not on the current calendar window.
Everyone has the physics. We built the ledger.
If you’ve read about MMM you’ve seen “adstock” everywhere; Meridian, Robyn, and PyMC-Marketing all advertise it. So it’s fair to ask what’s actually different if everyone models carryover.
Adstock is physics: it describes how an ad’s effect decays over time. Everyone has it, including us. What you do with that physics is decision-making, and there the paths split. The other tools use carryover to explain where this week’s sales came from, then leave the credit sitting on this week’s report. We accumulate the credit to the spending that earned it both now and into the future.
In one sentence: they use adstock to explain the past; we use it to value your decisions.
A note on marginal vs. average ROI. This distinction is also where the “unseen tail” idea bites hardest in practice: a channel’s average ROI can look flat while its next dollar is still earning well above that average, because so much of that next dollar’s payback is still on its way. If your current reporting shows only the average, you’re planning with half the picture.
We didn’t avoid the hard part
There’s a reason the open-source tools stop where they stop: giving every spending decision its own return is genuinely hard.
Any week’s sales come from many decisions at once: last week’s search ads, last quarter’s brand campaign, and everything in between. The split must assign credit fairly across tactics and time. Done carelessly, the numbers shift for reasons that have nothing to do with your marketing.
The other tools sidestep the question by reporting on the calendar instead, which means they can’t tell you what an individual decision returned.
We chose to do the work. The rule we follow is simple enough to say in one sentence:
The same budget spent in March versus the same budget spent in July earns the same credit at the same age into the future.
In other words, the contribution plays no favorites. When two campaigns show different returns, it’s because they truly performed differently (one ran in a stronger season, say), not because of a quirk in the reporting. We test this on every change we ship, we compared our approach against the established methods for fairly splitting shared results, and we wrote down why we chose it. Numbers you plan with should hold up to questioning, and these do.
What this means for you
If all you need is a backward look at last quarter, an open-source library in the right hands can give you one. One practical note, though: free tools aren’t free to run. Meridian, Robyn, and PyMC-Marketing are code libraries, not products. Standing one up means having data scientists on staff to build the model, tune it, keep it honest as the data changes, and translate its output for the people who actually set budgets. That’s a real team doing real work, every quarter. KeenOS arrives with that work already done, and with people who stand behind the numbers.
If you need to decide what to do next, the questions change, and the tool has to change with them:
- What did each spending decision return? You get a lifetime return for each channel in each period, instead of weekly credit splits.
- Which of these two plans is better? You can compare scenarios on equal footing, with timing accounted for.
- Is my late-quarter spend really underperforming?Before you judge it, we can show how much of its return is still on its way.
- Can I trust these numbers enough to act on them? The contribution behind provides fair credit, tested continuously and documented in detail.
Get in touch
Measurement tells you what happened. KeenOS helps you decide what happens next. If you’d like to see your own plan through this lens, get in touch and we’ll walk through it with your data.
