TextNow Ran YouTube for 26 Weeks. 87% of the Lift Came After Month Two.

If you review a brand channel on the same 4-week cycle you use for performance, you’ll cancel it before most of its effect exists. TextNow now has the number to prove that: in a 26-week YouTube holdout test, 87% of the measured lift accumulated after the first two months, and the first five weeks showed almost no cumulative lift at all.

The test ran December 2025 to May 2026, documented by the measurement firm Haus. TextNow geo-fenced 20% of the US, split it 50/50 into exposed and holdout cells, and ran YouTube reach and video-completion campaigns against the exposed half. The metric was Total Week 2 Primary Users: people who install the app and are still active in week two, a retention measure rather than a download count.

The result was a 1.99% Total W2PU lift at the end of treatment, at an indexed cost per install-and-activate of 98 against a planning target of 100. Roughly 2% better than the plan assumed.

Nick Hughes, Growth Marketing Lead at TextNow, described what they were actually buying with the test:

“We wanted to understand the payback period for upper-funnel media: what we saw in brand lift, what we saw in search lift, what we saw in KPI lift, and when each effect showed up.”

The when is the finding. Everything else is ordinary.

Why the timing matters more than the lift

A 1.99% lift is a modest number. The shape of how it arrived isn’t.

Five weeks in, a marketer reviewing this campaign on a standard performance cadence would have seen essentially nothing and had a defensible case to kill it. The money would have been reallocated to a channel that reports inside the review window, and the 87% still to come would never have existed to be measured. That decision would have looked disciplined in the meeting, and been wrong.

Bonnie Herche, VP of Marketing at TextNow, put the underlying tension plainly:

“We need to measure our brand like we measure our performance, but brand doesn’t act like performance.”

That sentence is the whole problem with how most teams govern this spend. The measurement rigor is right to demand. The cadence borrowed with it is what does the damage, because a review rhythm built for channels that convert in days will systematically defund every channel that converts in months.

Hughes described the internal use, which is the part worth stealing:

“When teams asked whether a first-week increase in installs came from our new brand campaign, we could point to this test. We launched in December and saw the bulk of impact in March. That benchmark has been helpful for the internal team.”

The measurement window is a budget decision

One detail deserves more attention than it got. Extending the read by a six-week post-treatment window changed the answer.

Including that window produced a 1.84% lift and improved measured efficiency by roughly 8% against the treatment-only read. The campaign didn’t perform differently. The analyst chose a different end date and the channel’s apparent efficiency moved 8%.

That’s worth sitting with, because nobody negotiates the measurement window. It gets inherited from whatever the analytics tool defaults to, and it silently decides which channels look fundable. Gabriella Mark of TextNow’s growth marketing team noted the comparison that follows from it:

“The incremental cost per registration was stronger than some of our direct-response campaigns.”

An awareness channel beat direct-response on incremental cost, but only once someone looked over a long enough horizon to see it.

This is not just a YouTube problem

TextNow is a consumer app running national video, which isn’t most readers’ situation. The transferable part is the governance question, and it applies to any channel with delayed payback: SEO, content, community, partnerships, category education, PR.

Every one of those is reviewed on a cadence borrowed from paid search. Every one of them would show the same near-flat first five weeks. Almost none of them has a holdout test behind it, which means the argument to keep funding them is currently instinct against a spreadsheet, and the spreadsheet wins. It’s the same gap that leaves teams unable to tie AI spend to a result they never measured.

The reason vendor numbers rarely settle this is that they are measured on the vendor’s terms. When The Trade Desk published a 32% CPA improvement this week, its own footnote disclosed a sample size and a significance test and still never named a time period. The buyers Digiday spoke to described steady, incremental gains rather than a step change, which is what a number with no clock usually turns out to mean. A geo holdout with a stated window is a different class of evidence, and it is the class you can run yourself.

The counter-case: Haus sells the thing this proves you need

Haus sells incrementality measurement: geo-lift testing, causal attribution, marketing mix modeling. A case study showing that only a rigorous long-window holdout reveals the true value of upper-funnel media is, conveniently, an argument for buying rigorous long-window holdouts.

The design also flatters the finding. A 26-week test with a six-week tail is the kind of study only a company already committed to the channel will fund, so the published cases are selected for campaigns that worked. Nobody publishes the 26-week holdout that found nothing.

What survives that discount is the structure rather than the number. A 50/50 geo holdout is a real experiment, not an attribution model, and the honest thing about it is that it can return zero. Three named TextNow operators went on record with the timeline, which is more than most case studies offer.

Three moves follow.

Write down your review window before the campaign starts, not after. If you can’t say in advance how many weeks you’ll let a brand channel run before judging it, you haven’t funded a channel. You’ve funded an option you’ll close at the first quiet review.

Run one geo holdout on the channel you most suspect is wasted. Half your markets on, half off, one metric that matters, and a window long enough to be embarrassing. It’s the only test that can come back and tell you the answer is nothing.

Stop letting the analytics default pick the end date. An 8% efficiency swing from a six-week tail beats most of the optimizations teams spend a quarter chasing, and it costs one conversation about when to stop counting.