Picsart's Gevorg Kazaryan: inside Vera, the AI that grades your ad before it goes live
Picsart's Gevorg Kazaryan explains how Vera scores ads, turns feedback into edits, and keeps final campaign decisions with marketers.
Upload a finished video to Picsart, an AI-powered creative platform used by marketers and creators to design, edit and generate content. Tell the system which audience it is for and where it will run. The new Vera agent can return a score, flag strengths and weaknesses, recommend a change, and pass that change to another creative agent for a new version.
The loop can move quickly. The decision to ship still belongs to a person.
That boundary matters to Gevorg Kazaryan, Product Director of AI Agents at Picsart. His role sits inside the part of the company building agent-based creative workflows, where marketers can hand off pieces of production while keeping approval over the result.
On September 22, Picsart and consumer-insights company Zappi launched Vera, an AI ad strategist inside Picsart's Agent Marketplace. Picsart says the underlying Zappi model identified the stronger creative 86% to 93% of the time across key metrics when validated against hundreds of digital ads evaluated by consumers.
ContentGrip asked Gevorg what happens after that score appears on screen, where human judgment still enters the process, and how the workflow changes when the same creative moves across markets.
Table of contents
- What Vera does before media spend
- Turn a weak score into an edit
- Keep the final call with the marketer
- Make market context part of the brief
What Vera does before media spend
Gevorg described the starting point as a finished video plus context: the brand, the target audience and the placement. Vera then returns an estimated resonance score together with strengths, weaknesses and recommendations.
That gives a marketer another checkpoint before buying media. The creative team can still use its own experience, consumer feedback or a small-budget campaign. Vera adds a faster assessment earlier in the process.
Gevorg said teams should treat Vera as "another data point that helps teams make better decisions." Feedback that might take days or weeks through a traditional research process can arrive earlier, while the marketing team keeps responsibility for what happens next.
The score is therefore useful only if it leads to a decision. A marketer can accept the recommendation, ask for more detail, ignore it, or send the feedback into another creative pass.
Patrecia Meliana
Turn a weak score into an edit
The handoff between research and production is the practical part of the workflow.
Gevorg gave a simple example. Vera might recommend showing a product from more angles. The marketer can approve that direction and ask Vera to pass the feedback to a Picsart creative agent. The video agent converts the recommendation into detailed generation or editing instructions, produces the revised asset, and returns it for another assessment.
A recommendation that begins as research can therefore end as a visible change in the video. The marketer can compare the new version with the previous one and decide whether the edit improved the work enough to move forward.
This extends a broader shift already underway inside Picsart's agent marketplace: separate agents can take on different parts of the same creative job. The important detail in Gevorg's description is the review loop. A result can come back for another pass instead of moving automatically into publication.
Keep the final call with the marketer
The harder case comes when the creative team disagrees with Vera.
Gevorg said the team can ask the agent to explain its reasoning, challenge the recommendation and request more detail. If that discussion does not resolve the disagreement, the responsible marketer still chooses which version goes live.
"People still own the judgment call," Gevorg said.
He also suggested a practical way to settle a disagreement: run an A/B test. One version can follow Vera's recommendation while another keeps the team's preferred direction. Performance then gives the team evidence for the next decision.
That approach also sets a useful limit on predictive scores. A model can narrow uncertainty before launch. It cannot remove the need to compare the prediction with what happens after real people see the campaign.
Make market context part of the brief
A creative score can change when the audience changes, so Gevorg said the market and target audience have to be part of the input from the beginning.
He also expects the agent to state where its knowledge is thin. "We want it to acknowledge those gaps rather than give an answer simply to please the user," Gevorg said.
That becomes especially important when a campaign moves across countries. Gevorg used Brazil as an example. A team adapting an English-language video could translate the dialogue into Brazilian Portuguese, adjust expressions and slang, and change visual material to fit the local context. Another creative agent can make those changes before the localized version returns to Vera for another assessment.
The result is a workflow where research and production sit closer together. A marketer can test a finished asset, turn feedback into an edit, test again, and decide when the work is ready for media spend. The speed comes from the handoffs between agents. The final approval stays with the marketer.
