Just do it - Measuring Complex Customer Journeys
Measuring complex user journeys is nearly an impossible task without the proper tools. Learn how MMM & Geo Tests can help advertisers measure the impact of their cross-marketing efforts.
What is an attribution window?
An attribution window is how long after an ad interaction a conversion still counts as attributable to it. It is a setting rather than a measurement, and changing it changes every reported number without anything about the advertising having changed.
In plain English
An attribution window defines how long after seeing or clicking an ad a conversion is still credited to it. Seven days after a click, one day after a view, thirty days after either — these are configuration choices, not properties of customer behaviour, and every reported performance figure depends on them. Widen the window and every channel's numbers improve without a single additional sale.
Windows come in two kinds and the distinction matters enormously. A click-through window starts when someone clicked, which is a deliberate action and a reasonable basis for crediting what followed. A view-through window starts when an ad was merely displayed, possibly unseen and unnoticed. View-through windows are much more generous in what they capture and much weaker as evidence, and a platform quoting impressive numbers on a long view-through window is measuring coincidence as much as influence.
The defaults differ between platforms, which makes cross-platform comparison quietly meaningless. One channel reporting on a 7-day click and 1-day view basis and another on 28-day click and 7-day view are not measuring comparable things, and the second will always look better. Before comparing any two channels' reported performance, check that the windows match — they frequently do not, and the gap is often the entire difference between them.
Double counting follows directly. A customer who clicked a display ad on Monday and a search ad on Thursday is inside both windows on Friday, so both platforms claim the conversion. Summing platform-reported conversions therefore exceeds actual conversions, sometimes substantially, and a media report that adds up to more sales than the business made is usually this rather than a data error.
Setting one honestly means looking at your actual conversion lag: the distribution of time between first exposure and purchase. A window covering most of that distribution is defensible; one much longer captures coincidence, and one much shorter undercounts slow deciders. Fix it before the campaign and leave it alone — a window adjusted after seeing which value flatters the result is choosing the answer, and nothing in the output records that it happened.
The formula
The rule, the two window types, and the arithmetic of why platform numbers sum to more than reality.
- The rule
credit if t_conversion − t_interaction ≤ windowA configuration threshold. Nothing about it is estimated from data.
- The two types
click-through (started by a click) vs view-through (by an impression)View-through captures far more and evidences far less.
- Why totals exceed reality
Σ platform-attributed conversions > actual conversionsOverlapping windows mean several platforms claim the same conversion.
- Setting it
cover most of the observed conversion-lag distributionMeasure the lag from your own data — see the proportion confidence interval calculator.
Worked example
A retailer audits its reporting and finds three platforms using different windows. The same quarter is then recomputed on a single consistent basis to see how much of the reported performance was a settings artefact.
Platforms claimed 41,200 conversions against 26,800 actual — a 54% overcount — and one platform's apparent performance was mostly window generosity.
The overcount is the clearest symptom and the easiest to check: if your platforms collectively claim more conversions than the business recorded, overlapping windows are the reason and the individual figures cannot be compared. Platform C's 30-day view-through window is the outlier — it credits itself for any conversion within a month of an impression the customer may never have noticed, which on a high-reach display buy is a large share of all conversions. Recomputing everything on a 7-day click basis with no view-through cuts its share by 58% and puts the three platforms on comparable footing. Two things follow. Cross-platform comparisons require matched windows before any conclusion is drawn, and the reconciliation should be run periodically rather than assumed. And even matched windows only fix comparability — attribution on any window still counts conversions that would have happened anyway, which is what [incrementality](/glossary/incrementality) measurement addresses.
Common misconceptions
- “A longer attribution window gives a more complete picture.”
- It captures more conversions and increasingly ones the advertising had nothing to do with. Beyond the genuine conversion lag, a longer window is adding coincidence rather than insight — someone who saw an ad three weeks ago and bought today probably did so for other reasons.
- “Platform-reported conversions can be summed to a total.”
- Overlapping windows mean multiple platforms claim the same conversion, so the sum routinely exceeds actual sales. If your reporting adds up to more conversions than the business recorded, that is the mechanism rather than a tracking fault, and no individual figure should be read as a standalone contribution.
- “View-through and click-through attribution are broadly equivalent.”
- A click is a deliberate action; a view may not have been noticed at all. View-through windows credit conversions to impressions that were served, possibly below the fold, possibly never seen. They are far more generous and far weaker as evidence, and a channel's apparent performance can rest almost entirely on them.
Frequently asked questions
How should I choose an attribution window?
Measure your own conversion lag — the distribution of time from first exposure to purchase — and pick a window covering most of it. For impulse categories that is days; for considered purchases it can be weeks. Then fix it before the campaign and leave it alone. Adjusting it after seeing which value flatters the result is choosing the answer, and nothing in the output records that it happened.
Can I compare channels reporting on different windows?
Not directly. A platform on 28-day click and 7-day view will always outperform one on 7-day click and 1-day view, whatever the underlying reality. Recompute both on a common basis before comparing, and check the defaults rather than assuming they match — they usually do not, and the difference is often larger than the performance gap being discussed.
Should I count view-through conversions at all?
With considerable caution and a short window if at all. A view-through credit assumes an impression the customer may never have seen influenced a purchase days later, which is a strong claim on weak evidence. Where display genuinely contributes, an incrementality experiment establishes it properly — a generous view-through window is a way of asserting the contribution rather than demonstrating it.
Knowing the term is the easy part
Applying it to a live measurement problem is the part that goes wrong. If you are designing an experiment, reading a result you do not trust, or trying to work out what your marketing actually caused, that is the work we do.