Out of HomeMeasured
89%
narrower than your guess. The data narrowed the range by 89% and the model settled cleanly.
guessed $1.22 · concluded $0.86 ($0.24–$2.19)
You came in with six guesses. Here is what your data did to each one.
The pale shape is what you believed before the model saw anything. The solid shape is what it concluded afterwards. If the solid shape is much narrower, your data taught it something. If it sits in the same place at the same width, the number is your own assumption dressed up as a result — and you should treat it that way.
TV — BroadcastMeasured
82%
narrower than your guess. The data narrowed the range by 82% and the model settled cleanly.
guessed $1.22 · concluded $1.67 ($0.51–$3.38)
Video — CTVMeasured
63%
narrower than your guess. The data narrowed the range by 63% and the model settled cleanly.
guessed $1.22 · concluded $1.21 ($0.27–$4.03)
Paid SocialPartly
46%
narrower than your guess. The data narrowed the range by 46%, so a good part of this still rests on the starting assumption.
guessed $1.22 · concluded $1.33 ($0.30–$5.21)
Programmatic DisplayWidened
79% wider
The data pulled this well away from your assumption but could not pin it down: the range came out wider than the one you started with. Trust the direction, treat the level as open.
guessed $1.22 · concluded $1.65 ($0.33–$8.26)
Paid SearchWidened 4×
396% wider
The data pulled this well away from your assumption but could not pin it down: the range came out wider than the one you started with. Trust the direction, treat the level as open.
guessed $1.22 · concluded $9.50 ($1.59–$16.82)
The one thing to take away from this screen
Paid Search is your best-performing and your worst-identified channel at the same time.
Its return looks like the strongest in the portfolio and its evidence is the weakest. Both of those things are true and neither cancels the other. You can spend into it — the direction is what the data is emphatic about — but do not build a plan that depends on $9.50 (somewhere between $1.59 and $16.82) specifically, and do not let anyone quote that figure without the range beside it.
What would fix it
Worth watching rather than fixing. If this channel also shows low contraction on the Evidence tab, treat its individual number as unreliable.
That paragraph is the engine’s own remedy for the finding that binds this channel, shown word for word. It is not general advice and it was not written by this screen.
What happened, week by week
Every period’s outcome split into the baseline the business would have earned anyway and the incremental effect of each channel. The black line is what actually happened. Each channel’s weeks add up to exactly the contribution reported on the overview — this is the same calculation, drawn out over time rather than summed. The shaded band at the right was held out of the fit entirely, so the model was never shown it.
Paid Search12.3%
TV — Broadcast3.8%
Paid Social2.7%
Programmatic Display2.3%
Video — CTV1.9%
Out of Home0.7%
Everything else76.3%
What you actually soldthe line
- On the weeks it learned from
- typically 5.5% outover 195 periods · 91% of the movement explained
- On the weeks we hid from it
- typically 11.1% outover 13 periods · 23% of the movement explained
- No better than assuming nothing and repeating the same period last year, which would have been 10.4% out.