How Operation Smile Canada used Prescient to find its best-performing channel hiding behind near-zero platform ROAS

Context

Operation Smile Canada Foundation (OS Canada) is a registered non-profit that provides free cleft surgery and long-term comprehensive cleft care for children in some of the hardest-to-reach places in the world. Every donation dollar has a job: a surgery, speech therapy, oral health care, nutritional supplements, training and education, and research, so the team has no appetite for spending on a channel that is not earning its place. To grow giving, OS Canada runs paid media across five channels at once: Google Ads, Meta, Connected TV (through StackAdapt), Pinterest, and Microsoft.

The Challenge

Each platform reported on its own performance, and read at face value the move was obvious: put more into search, pull back on Connected TV. On last-touch reporting, Google Ads looked like a top channel and Connected TV looked like a near-total loss.

But each platform could only measure its own contribution. Connected TV reaches people who see an ad, feel moved, and give days (or even weeks) later through a different channel, a path a last-click view cannot follow. Search tends to meet people who have already decided to give and typed the organization's name into Google. For a mission-driven organization, matching budget to the wrong read means fewer children helped. Operation Smile Canada needed one cross-channel view before moving another dollar.

The Solution

Operation Smile Canada brought all of its channels into Prescient's marketing mix model for one cross-channel view. Rather than reading each platform's report on itself, the model looks at total giving across the full window and works backward to measure how much each channel contributed, including the giving a channel influenced but did not directly close.

That is where the value sits. A platform can credit a donation only when it sees the last click or last touch before it happens. It cannot connect the donor who watched a Connected TV spot on Tuesday to the search click on Friday. Prescient's model assigns credit to both roles: the channel that created the demand and the channel that captured it. That reframed the question: rather than asking which channel to scale, they started asking which channel creates demand and which one meets it.

The Results

The modeled view reorders the channels

Ranked by return, the modeled view reorders the channels against where last-touch reporting placed them. The two channels furthest from their last-touch rank are the two that matter most to the decision.

Google Ads ranked near the top on last-touch but fell to last once modeled: heavy spend mostly capturing demand that would have arrived anyway. Connected TV climbed from #4 to #2, roughly 6x the contribution last-touch credited it, because it is view-based media with almost no clicks for last-touch to attribute. Pinterest jumped from last to first on a small budget. Inside Google Ads, the branded search campaigns are the weakest slice: they reported about 9x ROAS on last-touch, the kind of number that pulls budget toward them, yet Prescient ranked them lowest for true contribution because they mostly capture donors already searching for Operation Smile by name.

Where demand is created, and where it is captured

The clearest way to see the difference is to line up what each channel reported on last-touch against what Prescient measured. Google's branded search reported about 9x ROAS on last-touch, yet Prescient ranked it near the bottom: most of the donors it reaches had already decided to give before they searched for the organization by name. Connected TV read the other way, close to zero on last-touch because it is view-based, and a top-tier contributor once Prescient measured the giving it set in motion across other channels. The model is what tells you which channel is creating demand and which one is meeting it.

Scaling Connected TV

Measuring Connected TV was one thing. Scaling it was the test. Operation Smile Canada moved more budget into Connected TV through the first half of 2026 and watched the modeled performance at each step. From Q1 to Q2, Connected TV spend scaled 3x. If the return had come from a small, untested budget, this is the move that would have exposed it: modeled efficiency should fall as a channel runs out of new people to reach. Modeled efficiency held close to its Q1 level at 3x the volume, with no sign of saturation.

Key takeaways: What this means for you

1. A view-based channel can look like your weakest performer and still be your strongest. Connected TV ranked #4 on last-touch and #2 once modeled, because the giving it drives closes somewhere a last-click view cannot see. Pulling budget on dashboard signal alone is the fastest way to cut your best channel.
2. Match budget to the role each channel plays. Demand-capture channels like branded search deserve enough budget to meet donors already deciding to give. Demand-creation channels like Connected TV deserve room to grow, because they expand who knows about the mission.
3. Channels that look efficient at low spend can have real headroom. Connected TV scaled 3x and modeled efficiency held. Scaling is the test of whether a measured return is durable, and the team trusted the data through it.