Josh Levine runs Manychat’s performance creative program with a two-person team, tests new static ads every week, and puts four out of ten paid impressions behind creators. He does not put AI-generated copy or an AI creator persona into market.

That configuration reflects a specific operating principle. Josh, who built the program from scratch after joining Manychat in January 2025, has structured it around a view that the elements most commonly over-engineered in performance social are the ones that matter least.

“The best way that we found to create high-performing campaigns is working with people that have unique voices that can tell stories to audiences as if they’re one of them,” he says.

Manychat is a conversational marketing platform that helps creators and small businesses convert social interactions into paying customers across Instagram, Facebook Messenger, and WhatsApp. Josh’s team owns every paid visual asset the company puts into market, from concept through production, testing, and performance analysis. The framework governing those decisions is built around a single inversion: the hook comes last.

Hold First, Hook Later

Hook rate, or the proportion of viewers who stop on an ad within the first three seconds, is one of the most discussed metrics in performance social. Josh treats it as the wrong thing to optimize toward.

His team operates on a model he calls “hook, hold, and harvest”: capture attention, maintain it, then drive an action. The metric he actually optimizes against is the 3-second to 15-second retention rate, which measures how many viewers who stopped on an asset continue watching through 15 seconds.

“If you can isolate the retention, you can always go back and re-engineer what the hook is,” Josh says. “But if you create something that is intended to hook someone in and they don’t stay, you have a useless data point because you don’t have a way to bring them through the rest of your funnel.”

His team first identifies what drives conversion, then constructs messaging that retains viewers through to that point, then builds the hook to match. Engineering the hook first, he argues, gives you a metric that can look strong while the underlying message fails. “Thumbstop is something that most marketers over-engineer around,” Josh says. “While important, it shouldn’t be the primary focus.”

Four in Ten Impressions Go Behind Creators

Josh allocates four out of ten impressions in Manychat’s paid campaigns to creators. “They’re trusted voices,” he says, “and they’re the ones that can best represent our products because they’re users of it themselves.”

The highest-leverage move his team made was activating whitelisted ads, also called partnership ads, in which paid placements run directly from a creator’s account. He evaluates operational decisions using a consistent framework: level of effort against level of impact. Whitelisting has a low setup cost as long as a creator pipeline is already in place, and conversion rates are stronger from creator accounts than from brand handles, in his experience.

“Whitelisted ads have been the simplest lever that we’ve been able to pull at scale,” Josh says. “The level of effort to implement whitelisted ads is always going to be low compared to the high impact that it has.”

Every video asset in the program is human-generated. “We are very firm that everything that we put in market is human-led copy,” Josh says. The decision is partly a performance judgment and partly a deliberate brand position. He applies the same standard to creator content: every creator-produced video in market was made by a human.

Ads Have to Win to Stay

Determining whether a piece of creative is effective, rather than coinciding with a strong week of platform delivery, requires a specific benchmark. Josh uses a competitive model he describes with a tennis analogy.

In “King of the Court,” the holder keeps position until a challenger wins a single point outright. “When we have different ads in market with a similar format, a similar angle, a similar message, we are always trying to beat what we currently have in market,” Josh says. “If we find a different way to build it and beat it, that will be the new ‘King of the Court.’”

The same logic governs how he thinks about scale. Not every ad needs to reach a high spend ceiling to be considered a success. “Just because an ad is performing at $5,000 a month or $10,000 a month, if it’s not able to scale past that, it doesn’t make it unsuccessful,” he notes. His model treats the total addressable market as a collection of overlapping segments, each served by different creative, with multiple placements running simultaneously at different spend levels. 

Pushing additional budget behind a creative that is already performing, he argues, is generally less productive than building new work that captures a distinct pocket of the market. Meta’s machine learning, he adds, is reasonably effective at spending money behind something it identifies as successful; the strategic value lies in building the creative that reaches a segment the existing work does not.

One Asset, One Goal

Josh’s recommendation for brand marketers on creative structure is to stop requiring a single asset to accomplish multiple objectives.

“We’re not trying to get someone to laugh at our brand or trust our brand and then also learn about our brand and then sign up for a trial, all at once,” he says. “As long as you’re trying to get your brand to take away one specific thing, you’re going to be doing something correct.”

He attributes the prevalence of multi-goal creative to organizational structure rather than strategic intent. “My personal opinion is because they are trying to hit internal deliverables or goals,” Josh says, “rather than trying to do the thing that they might feel like is the right thing to do.”

The pressure, he argues, is structural. “There is the pre- and post-LLM marketing era,” he says, describing a period defined by heightened expectations for output velocity and efficiency. Under that pressure, Josh adds, teams compress multiple campaign goals into a single asset rather than building a cleaner brief. An ad competes in a feed against video content, other ads, and posts from people viewers actively follow. “If everything gets cluttered and your asset is hyper-focused,” he says. “You’re probably going to break through a lot stronger.”

Photo credit: Taylor Prinsen Photography

Twelve Days from Idea to Launch

The operational change that most accelerated Josh’s program was a redesign of how the team batches static ad production. Previously, large creative volumes were assembled monthly and distributed over the following weeks, with performance data arriving roughly every 30 days. Josh rebuilt the process around a weekly cadence, with smaller batches launching continuously. The team now moves from ideation to in-platform launch in about 12 days, down from 30. “We’re no longer waiting four weeks between insights,” he says. “We’re waiting seven days.”

One question arising from that faster cycle remains open. Josh’s team ran an analysis in April that tracked creative performance from day one through day 180. Within that window, assets behaved predictably: if something was working, the data said so consistently. Past that threshold, results became erratic. “Once it hit that 180-day mark is when we started to see real volatility,” he says, noting that the cause is not established. It could reflect ad fatigue, a data set that thins as an asset ages, or cumulative negative brand sentiment from audiences who have seen the same message for six months.

“We are unsure if that means there is negative brand sentiment or any pushback about having the same message in market for more than six months,” Josh says. His team is working through whether a principled threshold exists for pulling a high-performing asset before it accrues negative exposure, even when conversion metrics remain solid. “It involves not just quantitative data, but also some form of brand involvement,” he says. “It’s a problem that can’t be isolated to just one asset.”

Subscribe to Our Newsletter

Check Out Our Podcast