Jamloop today opened a household-level holdout methodology inside its self-serve platform and announced a partnership with Recast, giving connected TV advertisers two separate ways to test whether streaming ads drove business outcomes that would not have happened otherwise. The performance CTV platform, based in Walnut Creek, California, tied the launch to an early result from a streaming service campaign: 3,224 incremental subscriptions measured at 99.98% confidence, an estimated 365% return on ad spend.

The distinction the company is pressing sits at the center of a measurement argument that has followed connected TV advertising for years. Impressions, reach, and completion rates describe delivery. Attribution assigns credit for conversions that may have occurred regardless of the ad. Incrementality asks a narrower and harder question: did the campaign cause outcomes that would not have existed without it? According to Jamloop, its new capability is built to answer exactly that.

How the holdout works

The mechanism is a randomized holdout applied before a campaign launches rather than a reporting label attached after it ends. According to Jamloop, the platform randomly assigns eligible households into treatment and holdout groups ahead of delivery. During the campaign, the company's proprietary bidder serves ads to the treatment group while actively suppressing exposure to the holdout group. Jamloop then validates the holdout households against real bidstream activity to confirm they were reachable and impression-eligible, comparing outcomes between exposed and unexposed households.

That validation step is where the company draws its technical line. Households are removed from the control group if they were never impression-eligible or were incidentally exposed during delivery. The claim is that owning the bidding workflow lets the holdout function as an enforced condition of delivery rather than a set of records assembled once a campaign concludes.

The design targets a known weakness in post-bid randomization, a method in which test and control assignment happens after a bid is won. Under that approach the advertiser pays for the impression regardless of whether a real ad or a placeholder is served, so a 10% holdout can consume roughly 10% of budget on non-campaign delivery. Selecting and suppressing the holdout before the auction is the structural alternative Jamloop is positioning against modeled audiences, synthetic controls, and exposed-versus-unexposed comparisons built retroactively.

The reported result

Jamloop attached a single campaign to the launch. According to the company, it measured 3,224 incremental subscriptions at 99.98% confidence for a major CTV streaming service focused on subscriber growth, representing an estimated 365% return on ad spend. The figure is an interim read. After the campaign's 30-day measurement window closes, Jamloop said it will provide a final read on cost per incremental subscriber, incremental revenue, and return on ad spend.

Two qualifications sit inside those numbers as stated. The ROAS figure is described as an estimate, and the streaming service is unnamed, so the result cannot be independently verified against a specific advertiser or campaign structure. The confidence interval refers to the statistical reliability of the measured lift, not to the size of the underlying return, which remains subject to the final read once the window closes.

The 99.98% confidence figure is worth reading carefully, because incrementality experiments have a documented history of imprecision. Academic work on advertising lift tests has shown that individual customer responses carry high variance while advertising effects tend to be small relative to that noise, producing sales-lift estimates that stay imprecise even in very large experiments. Compounding the problem, advertisers frequently run small control groups, sometimes 5 to 10% of test size, to limit the opportunity cost of withholding ads. A high confidence level indicates the measured difference between exposed and unexposed groups is unlikely to be an artifact of chance. It does not, on its own, describe how large the return was or how the holdout was sized, both of which shape how far the result can be generalized to other campaigns.

"CTV is trying to earn its seat at the performance table," said Jeff Fagel, CMO at Jamloop. "But to do that, it cannot rely only on impressions, reach, completion rates, or attribution that gives media credit for conversions that may have happened anyway. Performance marketers want to know: did CTV drive more visits, more sales, more subscriptions, or more revenue? That is the standard search and social helped create, and it is where CTV needs to go next."

Two methodologies, not one

Alongside the household-level holdout, Jamloop is partnering with Recast, a marketing planning and analysis platform, to add geographic lift testing to CTV campaigns. According to the announcement, Recast's GeoLift methodology compares performance across defined markets such as ZIP codes, DMAs, or other geographic regions, estimating whether CTV drove incremental business impact at the market level rather than the household level.

The two approaches measure different things. Household-level holdouts isolate the effect of exposure at the audience level. Geo-lift captures broader market effects that user-level measurement can miss. According to Jamloop, the intent is to give advertisers both lenses and let campaign structure, scale, objectives, and available business data determine which applies.

"We built GeoLift by Recast because we believe advertisers should be running more experiments on their media, and not just accepting modeled answers," said Michael Kaminsky, Co-founder and CEO of Recast. "CTV is a natural fit for geographic-based testing as you can measure whether markets that saw a given campaign drove stronger conversion than comparable markets that didn't. We love that Jamloop is building incrementality into how campaigns are actually delivered, and that commitment to rigor is exactly what we want to support."

Daniel Gulick, Head of Product at Jamloop, framed the pairing as a hedge against relying on any single measurement. "Household-level holdouts and geo-lift each provide a different lens on incrementality," he said. "Household-level holdouts are powerful for measuring the impact of ad exposure at the audience level, while geo-lift can capture broader market effects that aren't always visible through user-level measurement alone. The strongest advertisers will use multiple methodologies, compare results, and avoid relying on any single source of truth."

Availability and roadmap

According to Jamloop, the incrementality capability is available now for eligible U.S. advertisers with sufficient campaign scale, conversion volume, and duration. Experiment reporting runs in-platform. Self-serve experiment configuration, which would let advertisers set up their own tests without company involvement, is planned as part of Jamloop's 2026 product roadmap rather than shipping at launch.

Why this matters for CTV budgets

The launch lands inside a documented shift in how much money streaming commands and how sceptically buyers now treat its metrics. Nielsen data showed streaming reaching 44.8% of total TV viewership in May 2025, surpassing the combined share of broadcast at 20.1% and cable at 24.1% for the first time. Connected TV advertising spending approached $33.35 billion in 2025, with CTV's share of media budgets doubling from 14% in 2023 to 28% in 2025. As spend has grown, so has pressure to prove it works.

The gap Jamloop is addressing has been mapped repeatedly on PPC Land. Coverage of CTV's conversion problemdocumented why streaming still struggles to close the loop that search and social closed years ago, with performance teams often forced to read CTV impact indirectly through downstream search and CPA movement. That difficulty is why standardization efforts have accelerated: IAB and IAB Europe published an incrementality framework for commerce media on September 9, 2025, distinguishing genuine causal lift from correlation and attribution. A separate strand of infrastructure work has focused on conversion signals: the IAB pushed standardized Conversion APIs for CTV on October 30, 2025, reporting that two-thirds of advertisers improved ROAS after implementation, and streaming platforms have since raced to build server-side conversion infrastructure of their own.

Those two tracks, causal measurement and conversion signal plumbing, address the same underlying complaint from different angles. Conversion APIs give a platform the raw outcome data. Incrementality methods determine whether that outcome was caused by the ad. Jamloop's holdout sits in the second category, and its pitch is that it produces the cleaner causal read by controlling the auction rather than reconciling logs afterward.

Jamloop's before-launch holdout is one design point in a crowded field of competing methods. When Google reduced its incrementality testing budget minimum to $5,000 in May 2025 using Bayesian statistics, it lowered the cost of causal measurement but kept the model built on experimental design and audience exclusion. Other vendors have pushed toward continuous measurement instead of discrete tests: Smartly signed a letter of intent to acquire INCRMNTAL in March 2026 to fold always-on incrementality, which uses causal AI without user-level data, directly into its platform. Nexxenbuilt ghost bidding incrementality into its DSP in March 2026, submitting control-group bids without serving ads. Kochava folded on/off pulse testing into its core product this month, dropping the need for an outside agency or data-science team.

Against that backdrop, Jamloop's argument is procedural rather than statistical. The company is not claiming a novel model. It is claiming that owning the auction lets it enforce the holdout as a condition of delivery, then clean the control group of households that were never reachable or were incidentally exposed. Whether that procedural control produces materially more defensible reads than post-hoc methods is the question the unnamed streaming result gestures at but cannot settle on its own. The final cost-per-subscriber read, due once the 30-day window closes, will offer a firmer test of the approach than an interim estimate can.

For performance-oriented budget holders weighing streaming against channels where causality is easier to demonstrate, the direction of travel is consistent across the vendors moving this year. CTV is being rebuilt to answer the incrementality question directly, because the budgets increasingly depend on the answer.

Timeline

Summary

Who: Jamloop, a performance connected TV platform based in Walnut Creek, California, in partnership with Recast, a marketing planning and analysis platform. Executives quoted include Jeff Fagel, CMO at Jamloop; Michael Kaminsky, Co-founder and CEO of Recast; and Daniel Gulick, Head of Product at Jamloop.

What: The launch of a household-level holdout methodology built into campaign activation, paired with Recast's GeoLift market-level testing across ZIP codes and DMAs. Jamloop reported an interim result of 3,224 incremental subscriptions at 99.98% confidence, an estimated 365% return on ad spend, for an unnamed streaming service campaign, with a final read on cost per incremental subscriber, incremental revenue, and ROAS due after the 30-day measurement window closes.

When: Announced on July 21, 2026. The capability is available now for eligible U.S. advertisers, with self-serve experiment configuration planned on Jamloop's 2026 product roadmap.

Where: Available to eligible U.S. advertisers with sufficient campaign scale, conversion volume, and duration, with experiment reporting delivered in-platform.

Why: Connected TV spending has grown sharply, with streaming reaching 44.8% of TV viewership in May 2025 and CTV budget share doubling from 14% in 2023 to 28% in 2025, while buyers increasingly demand proof that streaming ads cause outcomes rather than take credit for conversions that would have happened anyway. Jamloop's before-launch, bidder-enforced holdout is positioned against post-hoc and modeled measurement methods as a more defensible way to isolate that causal effect.