The short answer
Yes. Web-only incrementality tests can inform app sales calibration in MMM, but the app effect remains an assumption until it is measured. Verify which purchases the test captures, check whether ad routing or tracking has changed, and use a justified transfer assumption with wider uncertainty. Prefer experimental evidence covering both web and app.
This article is part of Asked by Marketers, a series answering real questions from marketing leaders.
Sellforte's team holds more than 1,450 meetings each year with marketing leaders in Ecommerce and Retail about Marketing Mix Modeling and incrementality testing. Each week, we anonymize at least one question from those conversations and answer it in depth, based on what marketers are actually struggling with, not what keyword tools suggest. About the series →
Why marketers ask this
A business sells through its website and app, but its testing setup only follows purchases completed on the web. The MMM includes both. The analytics team now has a credible web result and an unresolved app estimate, while the marketing team needs a budget recommendation for the whole business. An old test becomes harder to interpret when ads that once opened a website now send customers into the app.
Which purchases did the test actually measure?
The test supports conclusions about the purchases included in its outcome data. Before using it to calibrate Marketing Mix Modeling, confirm whether that outcome is web revenue, app revenue, or both. A campaign's destination and a reporting label such as "web purchases" are not enough to establish coverage.
Check the conversion-event configuration against the order data. Can an app purchase enter the reported event through a server integration? Are web and app purchases correctly classified and counted once? Does "web" include both mobile browsers and desktop browsers? A smartphone purchase is not necessarily an app purchase.
| What the test captures | What the result supports |
|---|---|
| Verified web purchases only | Evidence for web impact. App impact needs separate evidence or an explicit assumption. |
| Verified web and app purchases in one total | Evidence for combined impact. The split between web and app is still unmeasured unless analyzed separately. |
| Purchases with unclear coverage or classification | Reconcile the event data before assigning the result to either sales channel. |
Coverage is separate from test validity. A time-split analysis that compares advertising-on and advertising-off periods still needs to account for other changes over time. Adding app events improves outcome coverage; it does not remove confounding. The same distinction applies to incrementality testing more broadly: recording a purchase tells you that it happened, while a credible comparison estimates whether advertising caused it.
When can web evidence inform app calibration?
Transferring web evidence is more defensible when the tested advertising can plausibly drive purchases in both places and the relevant customer journeys are comparable. Specify exactly what the assumption transfers. Equal incremental ROAS, equal percentage lift, and an equal ratio of incremental to attributed revenue are different assumptions.
You can apply the web test's incrementality factor to app-attributed revenue for the same campaign group, provided you justify that choice. Here, the incrementality factor means incremental revenue divided by attributed revenue. This assumes that app-attributed revenue has the same relationship to incremental revenue as web-attributed revenue. It is not established merely because both appear in the same attribution report.
Before using that assumption, assess differences in customer mix and purchase behavior. An app audience dominated by existing customers may respond differently from people making their first website purchase. Campaign objective matters too: a test of ads intended to generate purchases provides limited support for an app-install campaign whose value arrives later.
Use the tested campaigns and period when assessing the web/app split. The app share of the company's total sales is not a substitute for the app share associated with those campaigns. Even a campaign-specific attribution split can be distorted by missing events or different attribution settings.
If the model accepts a prior for web alone, attach the test evidence there. If it requires a combined web-and-app prior, document how you estimated the missing app contribution. Entering the web-only iROAS as the combined return silently assumes zero app impact.
In the model's calibration record, distinguish the measured web result from the assumed app contribution. A prior is the expectation supplied to the model before it learns from the sales data. A transferred app prior should reflect uncertainty about the transfer as well as uncertainty in the original test.
What if ads now send customers into the app?
A change in ad routing can change where the advertising generates sales. If clicking an ad now opens an installed app, a web-only outcome may miss purchases that used to remain visible on the website. Record when that behavior changed and which campaigns, devices, or operating systems it affected.
Keep a separate timeline for tracking changes. App-attributed revenue can rise because more app events started reaching the reporting system, even if customer behavior stayed the same. Compare the event feed with the underlying web and app order totals before treating a reporting jump as evidence of a larger app effect.
An older test still describes its original conditions. Applying today's app share to a test from before deep linking began assumes a purchase journey that may not have existed then. Review tests on each side of the change separately and document how each contributes to the current calibration.
Opening the app after an ad click confirms that the route is possible. It does not establish that the ad caused an additional purchase. A customer might have opened the app and bought anyway, which is why routing checks cannot replace an incrementality estimate.
What evidence would resolve the app assumption?
To measure the app effect directly, use a credible experiment that observes web and app revenue under the same treatment and control comparison. First check whether an existing experiment can be reanalyzed using order data that identifies the purchase location. If the original assignment and suitable outcome data are available, you may be able to reuse the test.
For a geo-lift test, aggregate web and app orders into the same geographic units and measurement window, then estimate each outcome and their combined effect. This can capture purchases without matching every order to an individual ad exposure, provided the geographic design and sales data are suitable. For a user-level holdout, both outcomes need to be measured consistently across treatment and control groups.
Look at combined revenue as well as the split. Advertising may move a purchase from the website into the app without increasing total sales. Web-only results can therefore miss positive app impact or overlook substitution between the two. An unmeasured app effect should not automatically be set to zero, and zero is not a guaranteed lower bound.
The combined estimate may also be more precise than the separate estimates. Keep uncertainty visible if there are too few app purchases to estimate app lift reliably. Until stronger evidence is available, compare budget recommendations under plausible app assumptions. If the recommended allocation changes substantially, prioritize a test that resolves that uncertainty before committing to a large shift.
How this looks in practice
Consider a hypothetical campaign test with $10,000 of relevant spend and $30,000 of estimated incremental web revenue. For the same campaigns and measurement window, attribution reports $60,000 of web revenue and $20,000 of app revenue. Assume consistent revenue definitions, verified event coverage, and a test comparison against no spend. All figures below are illustrative.
The measured web iROAS is $30,000 ÷ $10,000 = 3.0. The web incrementality factor is $30,000 ÷ $60,000 = 0.5. Applying that factor to the app is a separate modeling choice:
| App assumption | Estimated incremental app revenue | Estimated total iROAS |
|---|---|---|
| No incremental app effect | $0 | ($30,000 + $0) ÷ $10,000 = 3.0 |
| Same 0.5 factor as web | $20,000 × 0.5 = $10,000 | ($30,000 + $10,000) ÷ $10,000 = 4.0 |
Neither app assumption was measured by the web-only test. These are sensitivity scenarios, not confidence limits or the full range of possible outcomes.
In this example, web accounts for 75% of campaign-attributed revenue. Dividing web iROAS by that share gives the same extrapolated total: 3.0 ÷ 0.75 = 4.0. This shortcut relies on the equal-factor assumption and the matched attribution data. Dividing by the web share of all company revenue would use a different denominator and would not support the same calculation.
The spend remains $10,000. A campaign can affect both sales channels. Calculating those effects separately does not create a second advertising cost. Compare the model's recommendations under both scenarios, retain the original web estimate and its uncertainty, and record the extra assumption behind the combined figure.
Related questions
Can we transfer a web result to an app in another country?
That combines two extrapolations: between sales channels and between markets. Assess each separately, because comparable web and app journeys do not establish comparable market conditions. Our guide to using incrementality evidence across countries explains the market-transfer decision.
How should we weight tests from before and after a routing change?
Assess how well each test represents the current purchase journey, alongside its design quality and uncertainty. A recent test with unclear app coverage can still be a poor calibration source. Our guide to weighting multiple incrementality tests covers how to combine evidence without treating every result as equally relevant.
How Sellforte helps
Sellforte combines MMM with incrementality experiments and attribution data to estimate marketing's contribution to sales. Web and app outcomes can be modeled separately where the sales data supports that split, with calibration reflecting the available evidence. Book a demo to discuss your measurement setup.
Authors
Lauri Potka is the Chief Operating Officer at Sellforte and has over 15 years of experience in Marketing Mix Modeling, marketing measurement, and media spend optimization. Before joining Sellforte, he worked as a management consultant at the Boston Consulting Group, advising some of the world's largest advertisers on data-driven marketing optimization. Follow Lauri on LinkedIn, where he is one of the leading voices in MMM and marketing measurement.