Orders
Orders by week, with the touchpoints on the path to each one.
For ecommerce
Bayesin models your weekly orders and revenue, separates what the season would have sold anyway, and ranks channels by what the next dollar returns.
Which channel's next dollar clears your margin
Modeled return on the next dollar, with the 80% credible range
| Paid search | 3.4× (2.9× to 3.9×) |
|---|---|
| Microsoft Ads | 3.1× (2.4× to 3.8×) |
| Paid social | 2.1× (1.5× to 2.9×) |
| 1.6× (0.9× to 2.6×) |
The baseline
A holiday week looks like a great week for whichever channel ran in it. The model first estimates what you would have sold with media at zero, then credits channels with the rest.
Your yearly pattern is fitted from your own history once it covers a full year, so a holiday peak lands in the baseline.
Sitewide sales and discount weeks are modeled beside media, so a promotion is not credited to the channel that happened to run.
A baseline that looks too small or too large is flagged before any channel number is shown.
Your conversions
The model explains the number on the books, net of refunds, not a platform's count of attributed sales.
Orders by week, with the touchpoints on the path to each one.
Weekly revenue beside one spend column per channel.
Charges and refunds reconciled, so returns come out of the total.
Recurring charges count as revenue for replenishment and subscription lines.
What the model needs
Before a fit is attempted, your data is checked against these. Anything missing is named, with what to add.
| History | 52 weeks to start. 104 recommended, so the season can be estimated twice. |
|---|---|
| Spend that varies | A channel whose weekly spend barely moved is flagged. The model cannot learn from a flat line. |
| Time in market | A channel active for under 26 weeks is marked, and its range shows it. |
| Cadence | Data refreshes nightly. The model refits every week. |
Questions
No. A weekly file with one revenue column and one column per channel is enough. A connector saves the upload; it does not change the model.
Under 52 weeks, the data check tells you what is missing and when you will have enough. No number is shown until then.
No. Bayesin shows the case and the modeled outcome. Your team decides and makes the change.
Tell us about your channels. We set up your first fit with you.