How should we judge results on channels with less data than Meta?
Our answer
Set the cost per new customer and decision date before launch. Then judge a smaller channel on total sales and new customers, discounted platform numbers and a post-purchase survey together. Run a holdout only when your volume can read one.
- Confidence
- Medium
- Strongest proof
- 3 official rules
- 3 datasets, 6 practitioners, 5 opinions or claims
- Sources
- 15 sources
- 17 statements
R
MI
New here? Take the 1-minute tour
Why this is the answer
3 reasons and our judgmentSmaller platforms over-credit themselves. In Q1 2026, Snapchat reported a 6.18 times return against 1.85 times after incrementality (Common Thread Collective, ecommerce ad agency). Three new channels each reported a 7 ROAS for one client while revenue stayed flat (Marin Istvanic, creative strategist).
Small volume slows holdouts. An illustrative holdout needs about six weeks at 5,000 orders a week and past 20 weeks at 1,250 (Measured, incrementality testing firm). Sellforte advises related tests or benchmarks for a channel too small to test, with a trigger to test later.
A bar set in advance keeps the read honest. Eightx (fractional CFO firm) advises setting a cost per customer and payback target before launch, then reading a Reddit test against a post-purchase survey and blended efficiency, not the platform's own number.
Our judgment
- Discount reported returns until you have your own test. Multiply a channel's reported return by a benchmark factor for that channel, such as Common Thread Collective's 0.30 times for Snapchat and 0.42 times for Pinterest.
- Judge on total sales and new customers, discounted platform numbers and a post-purchase survey together, at a date set before launch. Each misleads alone: platform reports over-credit and a survey only hears from buyers who answer it.
- Use early events only to check delivery. Early in a test, add to cart or page visit counts show whether the channel delivers. Make the keep or cut call on sales and new customers, because Reddit moves low budgets to those events while Eightx and SecondWin judge on customer cost and new customer count.
- Compare channels on 7-day click alone. Set the platform's window to 7-day click and 1-day view where it allows, but read every channel on 7-day click when comparing, because adding 1-day view did not bring Meta's numbers closer to measured lift in Haus's holdout benchmark (Attribution windows for each campaign).
Our judgment, built from Common Thread Collective, Marin Istvanic, Eightx, Cody Plofker, Reddit, SecondWin, Elevarus and Olivia Kory: our reading of these sources, not a finding from any one of them.
What the sources say
17 statements from 15 sourcesReddit's default attribution counts conversions within 28 days of a click and one day of a view. Windows of 1, 7 or 28 days are available and apply to all your campaigns.
“Your view window can't be longer than your click-through window.”
Pinterest says changing a campaign in Learning mode may reset it, the Learning indicator takes about two weeks on average to clear, and campaigns are best judged weekly or monthly, not daily.
“If you make changes to your campaign while it's in Learning mode, it might reset and take longer.”
Marin Istvanic says one client added Pinterest, TikTok and Snapchat, each reporting 7 ROAS, but total revenue did not change. If MER falls with the new spend while revenue stays flat, the channel is just taking credit.
“if you add a new channel and MER drops proportionally to the new spend while total revenue stays flat, that channel is stealing attribution.”
Haus's holdout tests found Meta's Incremental Attribution tracked measured lift more closely than standard windows, and optimizing to it drove about 1.2x the incremental return. Sample size not given.
SecondWin, which sells ChatGPT ad tools, suggests no paid channel drive more than roughly 60 to 70% of new customers, and testing a second channel with 10 to 15% of the Meta budget for 30 to 60 days. No data.
“Treat the first 30 to 60 days as paid learning.”
Measured's panel of 269 expansion-channel programs put Snapchat at 1.9x the panel's incrementality average, with conversion buys far ahead of awareness buys. It says to size tests with a power analysis.
“Size the test with a power analysis, not a round number.”
New Public Media, an agency, gives each podcast show its own promo code and vanity URL and lists each by name in a buyer survey. In its accounts, survey plus code data recovers 20 to 35 percent more revenue than pixels.
“run a 60 day geo holdout, keeping two comparable metros unexposed, to separate incremental revenue from orders you would have received anyway.”
Sellforte, which sells testing, says no fixed spend makes a channel too small to test; it is too small when no feasible design can detect an effect that would change the budget call. Then use broader evidence.
“A full pause in selected units produces more signal than a shallow cut everywhere”
Stackmatix, a paid media agency, says "where did you hear about us" answers are directional: buyers name the channel they remember, so check them against click and UTM data rather than taking them at face value.
“Respondents simplify: they pick the channel they remember, not the one that last touched them before purchase, so paid social often gets over-credited and dark social under-credited.”
Cody Plofker's first independent check on TV is the post-purchase survey, scaled up by the response rate (if half of buyers answer, double it), then compared with other attribution models.
“never allow anyone to create their own homework”
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”
Web Tonic, which sells marketing to contractors, says below about $10,000 a month in media, call tracking plus offline conversion uploads capture nearly all the value; above that, last touch over-credits branded search.
“Below roughly $10,000 in monthly media, call tracking plus offline imports capture nearly all the value and multi-touch adds cost without changing decisions.”
Reddit tells advertisers to budget at least $100 a day when optimizing for purchases with lowest cost bids, moving to Add to Cart or Page Visits on lower budgets, and to judge results after 7 to 14 days.
“Set budget at least to $100/day for Purchase events with Lowest cost bids. For lower budgets, move to Add to Cart or Page Visits.”
Eightx says to treat a Reddit test as a cheap experiment funded from test budget, set CAC and payback targets first, and discount Reddit's own ROAS, including its 8% Collection Ads lift.
“Fund it from a test budget, not from a winner.”
PipelineOn relays one plumbing owner who confirmed lead source on every call: Angi produced 4% of booked jobs, not the 12% it reported, and cutting it lowered CAC 18% the next quarter.
“Six weeks of clean data showed Angi producing 4% of booked jobs, not the 12% the platform reported.”
Elevarus, an agency selling Reddit ad work, says Reddit product ads suit research-driven buys in active subreddit categories, and to plan for 60 to 90 conversions and a 30-day learning window.
“Plan for 60 to 90 conversions before you judge ROAS against Meta.”
In CTC's Q1 2026 benchmark of 299 DTC brands, Snapchat's platform ROAS of 6.18x fell to 1.85x once adjusted for incrementality, and Pinterest's 6.57x to 2.76x. Each took under 0.5% of total spend.
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
- How sure we areWhat its confidence rests on, and what would change the answer