How should we measure the long-term effect of ads beyond the click window?
Our answer
Measure what lands after the click window in your own sales data. Keep reading geo holdouts after the ads stop, and follow new customers as cohorts for 12 months. Use a calibrated mix model only for effects lasting months.
- Confidence
- Medium
- Strongest proof
- 1 controlled test
- 3 official rules, 4 datasets, 4 practitioners, 7 opinions or claims
- Sources
- 14 sources
- 19 statements
G
New here? Take the 1-minute tour
Why this is the answer
3 reasons and our judgmentMuch of a test's lift can land after the ads stop. Haus found 24% of the typical YouTube test's lift landed after that point. In TextNow's 26-week geo test, about 87% of the effect came after the first two months.
Customers keep paying off well past a 7-day window. Wicked Reports, which sells this tracking, found customers from October Meta clicks worth $222 in October and $333 by December 31.
Platform estimates of late sales are disputed. About 20% of a brand's sales on Amazon arrive after its window, by Amazon's own figure. Some agencies told Digiday the uplift it credits to streaming TV doesn't show up in weekly sales.
Our judgment
- Measure after-window effects in your own sales data, not by trusting a longer platform window. Jon Loomer advises confirming late buying in data first, and Wicked Reports traced Meta click cohorts in order data through December 31. Some agencies in Digiday's report treat Amazon's long-term figure as a forecast.
- Keep each geo holdout in place for a window after the ads stop, and report the result with and without it. This follows Haus's YouTube tests, Measured's warning that ending early understates lift, Blck Alpaca's cooldown period and Google's cooldown range. On Meta's Conversion Lift, read the test window, since Meta advises against its post-test window without the holdout.
- Follow each channel's new customers as cohorts at 30, 90 and 180 days and 12 months, consistent with Measuring customer lifetime value. This combines RetentionX's checkpoints with ATTN's advice to track customers acquired during a test for 6 to 12 months.
- Use a mix model with carryover only for effects lasting months, and only past the bar in Whether you need a media mix model. Meridian's guidance says effects past the set lag are cut off, and in TextNow's test most of the effect came after two months.
- For email, direct mail and other channels you send yourself, keep a standing group that never receives them. This applies Sellforte's persistent no-mail group for recurring print to other owned sends.
Our judgment, built from Jon Loomer, Wicked Reports, Digiday, Haus, Measured, Blck Alpaca, Google, Meta, RetentionX, ATTN Agency and Sellforte: our reading of these sources, not a finding from any one of them.
What the sources say
19 statements from 14 sourcesMeta advises against using the post-test window to calculate Conversion Lift results, saying it can add variance and shrink the measured effect once holdouts end.
“While we offer the option, we do not recommend using the post-test conversion window for calculating results.”
Recast, a mix modeling company, describes an add-on that uses awareness and consideration surveys to estimate how much baseline sales each dollar of brand spend generates, by channel and over time. No results shown.
“how much baseline sales each dollar of brand spend generates”
Google says geo-based Conversion Lift usually needs more budget than the user-based kind, and to wait for the study's end before reading results. It reports incremental ROAS with a confidence interval.
“Conversion Lift based on geography typically requires a higher budget than the user-based alternative.”
Taylor Holiday says Meta works on a short attribution window because every ad optimizes toward a click-based conversion, so halo effects are minimal.
“the halo effects are actually de minimis”
In Haus YouTube geo tests, a typical 24% of lift arrived after the ads stopped, and counting that window raised typical ROI about 32%. A day after a flight, YouTube still booked about 64% of an in-flight day's lift.
“In the typical test, 24% of the lift landed in the post-treatment window.”
Amazon estimates about a fifth of a brand's Amazon sales value arrives after the usual 7 to 30 day attribution window, from 2 million June campaigns. Some agencies say its projected TV uplift does not show up.
“roughly 20% of the sales value for a brand shows up over time and isn't captured in that immediate attribution window”
Wicked Reports, which sells click-to-order tracking, says customers from October 2025 Meta clicks were worth $222 by the end of October and $333 by December 31, value a 7-day window misses.
“A 7 day cookie sees none of that $112. A 30 day window sees a fraction.”
Sellforte says no test type is best: use Meta Conversion Lift for Meta, geo-lift for Google Search, Shopping and Performance Max, and customer holdouts for email, push and direct mail.
“There is no single best universal test type.”
Sellforte says a well-powered Meta Conversion Lift test is credible for what it measured, but advertisers cannot audit assignment, so check it with occasional geo-lift studies before calibrating a model.
“We have not seen a systematic pattern in which the Meta result is always higher.”
Blck Alpaca, an agency, says Meta Conversion Lift suits comparisons inside Meta, while geo-lift suits budget moves between channels; it relays about $500,000 a year in spend as a geo holdout threshold.
“Use platform Conversion Lift for comparisons inside one system. Use geo-lift for larger channel and budget questions.”
In a 26-week Haus geo test of TextNow's YouTube awareness ads, lift in active users stayed near zero for about five weeks, and about 87% of it came after two months. A shorter test could have missed it.
“We need to measure our brand like we measure our performance, but brand doesn't act like performance.”
Measured, which sells geo testing, says incrementality tests commonly run four to eight weeks and almost never under four, since ending early misses lagged sales. Its illustrative Meta example needs about six weeks.
“almost no test should run under four weeks regardless of what the power math allows”
Google's Meridian lets media affect results for a set number of later periods (max_lag), assuming effects fade toward zero. It suggests 2 to 10 periods for geometric decay, and a longer lag need not raise ROI.
“Increasing max_lag doesn't necessarily mean that ROI estimates will also increase.”
RetentionX, which sells retention analytics, says put time-bound LTV on the weekly scorecard (at least 30, 90 and 180 days), and set a CAC ceiling from a cash payback limit such as day 90 or 180.
“A customer worth $200 over 12 months is not the same as a customer worth $200 over 3 years.”
CTC's CEO says Meta's 7-day click optimization window is too short for long-consideration brands, such as a $2,500 order with 60 days of thought, so he adds 28-day windows from Northbeam.
ATTN Agency suggests running brand campaign tests for 12 to 16 weeks and tracking customers acquired during a test for 6 to 12 months for retention and lifetime value. It gives no results.
“Brand campaigns often show incrementality over months rather than weeks.”
Jon Loomer says to first confirm with data that buyers really convert after the 7 or 28-day window, since a returning buyer can start a new 7-day window; he would keep a conversion goal.
“While the buying cycle might be longer, that doesn't necessarily mean that 7-day and 1-day click are no longer relevant.”
Haus, which sells incrementality tests, says branded search tests need 2 to 3 weeks, Meta conversion tests 3 to 4, and YouTube about 5 to 6 with a post-campaign window, which raised measured iROAS 79% on average.
“If you stop measuring the day your campaign ends, you are throwing away data.”
An Amazon Ads scientist's model of 15 US brands found upper and mid-funnel ads showed only 30 to 50 percent of their effect within a two-week window, against 60 to 90 percent for lower funnel.
“track their upper and middle-funnel ad products for at least a month to capture their longer-term effects.”
In the full answer
For readers who sign in- What to do this week3 steps
- The numbers behind it3 figures from cited sources
- Where each source standsA chart placing each source on the question
- The strongest disagreementThe best case against the answer, from a cited source
- What the research saysWhat academic research adds, kept apart from the answer
- How sure we areWhat its confidence rests on, and what would change the answer