What an attribution API returns that dashboards hide
Dashboards are edited. The API response is not. Four fields that tend to exist in the payload and never make it onto the screen, and what each one tells you.
By Joris van Huët, Founder & CEOUpdated 4 min read
Run the numbers for your store: the free return rate calculator.
A dashboard is an edited view, and the edit usually removes the fields that would have made you cautious. Reading the raw response is the cheapest way to find out what a tool knows and does not show.
This is not an accusation of bad faith. Interfaces are designed to be legible, and legibility means dropping columns. The problem is that the columns dropped first are almost always the ones that qualify the headline.
Four fields worth looking for
| Field | Typical dashboard treatment | What it tells you |
|---|---|---|
| Interval width | Hidden, or shown as a faint band | Whether the estimate can support a decision |
| Coverage | Absent | What share of real orders the read explains |
| Declined channels | Silently dropped or imputed | Which channels the method could not measure |
When a channel has too little spend for its effect to be separated from noise, a well-built method declines to score it. A dashboard that shows a tidy ranked list of every channel has either measured them all or quietly filled the gaps. Those are very different products and the interface looks identical.
How to check without building anything
You do not need an integration to find this out. Ask the vendor for a sample response payload for one account and one window. A vendor confident in its method will send one. Compare the fields in the payload against the fields on the screen, and ask about anything present in one and absent from the other.
We think this is a reasonable thing to ask of anyone, including us, and set out the wider case in why attribution vendors should explain methodology openly. The buying-side version is the checklist for vetting an attribution vendor.
The interval is the field that changes behaviour
A channel with an estimate of 1.8 and an interval from 1.6 to 2.0 supports a decision. The same 1.8 with an interval from 0.4 to 3.2 does not, and the dashboard renders both as 1.8.
The practical consequence is that teams scale channels on estimates that were never resolved, then conclude attribution does not work when the scaling does not pay. It worked fine. The width was hidden. Attribution tools that report confidence intervals covers who publishes them.
Coverage is the field that changes the framing
Coverage answers a question nobody thinks to ask: of the orders you actually took in this window, what share does this read explain? If it is high, the channel estimates describe your business. If it is low, they describe a slice of it, and the rest is sitting in unassigned traffic doing something the model cannot see. That is a normal condition and it is worth knowing before you reallocate.
What we return
Causality Engine returns, per channel, what it caused next to what last-click gave it, a data-health score from 0 to 100 and a next step, with receipts that check every number against the export. The method is written out on how it works.
The question to carry into any demo
Ask what the model does with a channel it cannot measure. The answer separates tools that are honest about their floor from tools that have no floor at all, and it takes about ten seconds to ask.
Related answers
Frequently asked questions
Why do attribution dashboards hide confidence intervals?
Mostly for legibility rather than deception. A ranked list of single numbers is easier to read than a list of ranges, so the range is the first thing an interface designer drops. The cost is that a resolved estimate and an unresolved one look identical on screen.What does coverage mean in an attribution report?
Coverage is the share of your actual orders in the window that the read explains. Low coverage does not invalidate the estimates, but it does mean they describe part of your business rather than all of it, which changes how much budget you should move on them.What should a tool do with a channel it cannot measure?
Name it as unmeasured at the current spend rather than assigning it a number. A method that returns a confident figure for every channel regardless of how little it spends is not measuring the small ones, it is filling them in.
Go deeper: Incrementality testing, explained.
Sixty-second versions of these ideas: Causality Engine on YouTube Shorts.
Have an audience of ecommerce operators? Earn 20% recurring commission by recommending the €99 read.
Keep reading
Terms in this article
- AttributionAttribution identifies user actions that contribute to a desired outcome and assigns value to each. It reveals which marketing touchpoints drive conversions.
- Attribution ReportAttribution Report shows which touchpoints or channels receive credit for a conversion. It identifies which campaigns drive desired actions.
- CausalityCausality is the relationship where one event directly causes another, essential for identifying specific actions that drive desired outcomes in marketing.
- ClickClick is the action a user takes to interact with a digital advertisement, redirecting them to a website or landing page. Clicks are a fundamental metric for measuring ad engagement and a primary input for click-based attribution models.
- Confidence IntervalConfidence Interval is a statistical range of values that likely contains the true value of a metric. In marketing analytics, it quantifies uncertainty around estimates, indicating the precision of an outcome or causal effect.
- DashboardA dashboard is a visual display of key information required to achieve specific objectives. It consolidates data onto a single screen for quick review.