Step two · before any modelling
Can this data answer the question at all?
A fit takes minutes and hands back confident-looking numbers whether or not the design could ever have identified them. So we ask whether it could first, and it costs a second. Nothing here has been modelled — this is arithmetic on your file.
The verdict
Caution
The model will run and most of your channels will come back with numbers you can use. The ones named below will not, and they carry that mark through every screen after this one.
4 things to carry forward, all about how wide the answers will be, not whether you get them: Paid Social, Paid Search, TV — Broadcast and Out of Home.
How this verdict is decided
There is no score to improve and no percentage to argue with. A caution about one channel never demotes the whole dataset, because it has somewhere better to live: that channel’s own row, and its trust grade after the fit.
Takes about two minutes, and it is the step that catches impossible assumptions before they become results.
Five questions we ask of every dataset
3900 rows · 11.4× headroom
342 things to estimate
2 channels flagged
Paid Social and Paid Search
2 channels flagged
TV — Broadcast and Out of Home
Present
a demand control is mapped
7.31 · line at 5.00
across the whole dataset
Where channels moved together
- Paid Social moved with Paid Search (r=0.74)
Five things to know, each with what to do about it
Paid Social and Paid Search correlate at r=0.74; their individual ROIs will trade off.
What to do: Vary their budgets independently for a quarter if you can. Until then, trust the pair's combined return more than either individual number.
Two channels move closely together0.740 measured · 0.70 thresholdOpen
Paid Social and Paid Search correlate at r=0.74; their individual ROIs will trade off.
What to do: Vary their budgets independently for a quarter if you can. Until then, trust the pair's combined return more than either individual number.
The largest sits 7 robust standard deviations off the local level, around 2022-03-28. Under a normal likelihood a handful of periods like this can pull the whole fit toward explaining them.
What to do: If these were real events — a promotion, a stockout, a price change — add them as a control column so the model can attribute them instead of absorbing them into media. If they are reporting errors, correct them at source. If they are genuine volatility, no action is needed: a full fit tests a heavy-tailed likelihood on known returns hidden in your own spend, and uses it only if it does better.
It runs in 41% of weeks, in bursts rather than at the start or end of the window. The effect is identified from the handful of on/off transitions rather than from the whole period, so its interval comes back wide even when the fit itself is sound.
What to do: Nothing to fix in the data — this is a media plan, not a defect. Expect a wide range on this channel and weight it accordingly; a lift test on a single burst would pin it faster than more history would.
Out of Home is off air most weeks0.583 measured · 0.50 thresholdOpen
It runs in 38% of weeks, in bursts rather than at the start or end of the window. The effect is identified from the handful of on/off transitions rather than from the whole period, so its interval comes back wide even when the fit itself is sound.
What to do: Nothing to fix in the data — this is a media plan, not a defect. Expect a wide range on this channel and weight it accordingly; a lift test on a single burst would pin it faster than more history would.
What we will do about it
Findings can change the model, never your numbers — we never impute, winsorise or drop a row. A change a finding calls for is only a candidate until a full fit has tested it on your own spend, and every entry says what it would cost you.
A full fit will test these on your own data before choosing
None of these is a change yet. Before fitting, a full fit hides returns we chose in copies of your own spend and tries setups built from these changes on them, side by side with your model exactly as you specified it. It uses a change only if its setup does better, and holds back from the budget plan any channel whose return it cannot find. A quick preview tests nothing and uses none of them.
A fit that expects the occasional extreme week
To be testedraised by a few extreme weeks in your salesSome weeks in your sales sit far from their usual level. A standard fit would have to explain them, and the only way it can is by changing how much it credits your media. We fit in a way that expects the occasional extreme week, so rare spikes and dips count as noise rather than as evidence. How often such weeks happen is learned from your data, not set by us.
What it would cost you, if a full fit uses it
Genuine one-off events are now down-weighted along with the errors. If a spike was a real promotion you want credited, add it as a control column instead — this setting protects the fit, it does not explain the spike.
One combined return for two channels bought together
To be testedraised by two channels bought togetherPaid Social + Paid SearchThese two channels were bought so closely together that the data cannot tell their effects apart, so any split between them would be a guess. We estimate one return for the pair from their combined spend, starting from an assumption that blends the two channels' own, weighted by how much each spent.
What it would cost you, if a full fit uses it
You get one number for the pair and no split within it. If you need each channel's own return, only an experiment that moves one without the other can give it to you.
A firmer starting assumption for a channel the data barely sees
To be testedraised by spend that runs in short burstsfor each ofTV — BroadcastOut of HomeThis channel's spend barely moves, runs in short bursts or covers only part of the window, so your data holds little evidence about what it returned. We narrow its starting assumption, so a few weeks of noise cannot hand it credit that belongs to the baseline or to another channel.
What it would cost you, if a full fit uses it
This channel's number leans on the starting assumption more than on your data. Read it as an informed default, not a measurement.
A return that can change from year to year
To be testedraised by a window long enough for returns to change from year to yearYour data covers at least two years, and each channel ran through most of every year it was on — long enough for its return to have changed over the window. We let each channel's return drift from one year to the next, starting from the same assumption a single average would.
What it would cost you, if a full fit uses it
Each year's number is less certain than one average across the whole window would be, because each year has only its own weeks to go on.
A shared starting assumption for two channels bought together
To be testedraised by two channels bought togetherPaid Social + Paid SearchThese two channels were bought so closely together that the data struggles to tell their effects apart. Each keeps its own return, but both start from the assumption that they return about the same, so the split between them leans on that rather than on noise.
What it would cost you, if a full fit uses it
The two returns are pulled toward each other. If one channel genuinely returns far more than the other, this setting narrows the gap between them.
An assumed diminishing-returns curve for a channel the data barely sees
To be testedraised by spend that runs in short burstsfor each ofTV — BroadcastOut of HomeThis channel's spend barely moves, runs in short bursts or covers only part of the window, so your data cannot trace how its returns fall off as spend rises. We fix that curve at the starting assumption — half its full effect at about its typical weekly spend — and estimate only how much the channel returned.
What it would cost you, if a full fit uses it
How fast this channel runs into diminishing returns is assumed, not measured, so its number, and any budget advice built on its curve, leans on the starting assumption.
How much each channel's spend moves about
Ups and downs are what let the model see what a channel does. Spend the same amount every week and there is nothing to learn from — the return comes back as whatever we assumed going in, however large the budget.
| Channel | Spend share | Variation | Overlap cost | Weeks with nothing spent |
|---|---|---|---|---|
| TV — Broadcast | 27.8% | 1.66 | 1.2 | 59% of weeks had no spend |
| Video — CTV | 15.5% | 0.93 | 3.0 | spent every week |
| Paid Social | 19.6% | 0.74 | 2.5 | spent every week |
| Paid Search | 17.3% | 0.77 | 4.8 | spent every week |
| Programmatic Display | 11.5% | 0.71 | 1.9 | spent every week |
| Out of Home | 8.3% | 1.64 | 1.2 | 62% of weeks had no spend |
How closely your channels move together
Read a cell as: when this channel went up, did that one go up too? 1.00 means always, 0.00 means never. Anything above 0.80 and your data cannot tell the two apart. Measured after allowing for advertising that keeps working once it stops running and for the tenth million doing less than the first — checking raw spend instead is the usual reason a tool declares your data fine and then fails to separate anything.
| TV — Broadcast | 1.00 | 0.02 | 0.22 | 0.25 | 0.14 | -0.17 |
|---|---|---|---|---|---|---|
| Video — CTV | 0.02 | 1.00 | 0.14 | 0.13 | 0.01 | -0.01 |
| Paid Social | 0.22 | 0.14 | 1.00 | 0.74 | 0.50 | 0.05 |
| Paid Search | 0.25 | 0.13 | 0.74 | 1.00 | 0.56 | 0.10 |
| Programmatic Display | 0.14 | 0.01 | 0.50 | 0.56 | 1.00 | -0.05 |
| Out of Home | -0.17 | -0.01 | 0.05 | 0.10 | -0.05 | 1.00 |
| TV — Broadcast | Video — CTV | Paid Social | Paid Search | Programmatic Display | Out of Home |