Marketing’s AI Shift Is From Measuring Campaigns to Running Them

Omri Shtayer, vice president of data products and DaaS at Similarweb, told PYMNTS that the company launched AI Studio to address what he sees as a structural limitation in many AI deployments: a lack of context.

“One of the things that most AI systems are lacking is context,” Shtayer said.

For Similarweb, that realization led to a broader effort to rethink how its proprietary data could power agents capable of delivering not just analysis, but action.

“I always say data is the fuel, AI is the engine, and the agents and actions are the aircraft that deliver those actions to real life,” Shtayer said.

Rather than layering a general-purpose chatbot onto existing dashboards, the company built what it calls a “super agent” trained specifically on Similarweb’s datasets and defined business use cases. The objective, Shtayer said, was to make the company’s data usable in what he described as “the new age of AI,” where conversational interfaces are expected to produce actionable insights. That positioning reflects a broader industry shift in which proprietary datasets are becoming as strategically important as the models themselves.

From Dashboards to Recommendations

Shtayer said he led the early effort to determine how Similarweb’s data could be delivered directly to agents in ways that empower them to generate insights and recommendations.

“I led the effort to determine how to bring our data to agents so they could use it to gain insights, actions and recommendations,” he said.

The result, he added, goes beyond a simple AI feature.

“It’s much more than that. We built a whole system,” Shtayer said.

Instead of requiring users to sift through hundreds of data points, the AI translates performance metrics into prioritized next steps. A marketing team might receive guidance on whether to increase paid search investment, focus on organic traffic or adjust social strategies. An investor might structure due diligence around traffic trends and competitive positioning. In that sense, the platform is designed to collapse the distance between analytics and execution, turning dashboards into workflow engines.

In one beta example, Shtayer described using the system during a client discussion about how to win a major publisher account. When faced with a tactical question about approach and timing, he turned to the platform.

“I’m not an expert. Let’s ask the AI,” he recalled.

What began as a single query expanded into a structured set of recommendations covering timing, contacts and optimization priorities. The interaction became, in his words, “a battle plan,” translating analytics into a tactical roadmap.

Building Trust With Guardrails

Enterprise adoption, Shtayer said, depends on trust. During beta testing, early users questioned whether they could rely on AI-generated outputs to guide revenue and marketing decisions.

“The first question in beta was how do I trust the AI? Nobody trusts AI 100%,” he said.

To address that skepticism, Similarweb embedded guardrails designed to constrain what the system can access and generate. The AI is limited to approved first-party and competitive datasets and restricted from relying on broad, generalized model knowledge.

“We put guardrails in place so the AI will not do anything except what we taught it and instructed it to do,” Shtayer said.

By narrowing the system’s scope and grounding it in curated data, the company aims to make outputs auditable and defensible for executives and department leaders increasingly accountable for AI-driven decisions.

Embedding AI Into Daily Workflows

Shtayer said AI Studio is designed to integrate directly into enterprise processes, particularly across marketing and sales teams.

“It fits into day-to-day strategy, marketing, content creation and Salesforce sales,” he said.

In practice, users are asking highly tactical questions, from how to craft outreach emails to which prospects to prioritize. In payments, Shtayer said the system can identify fast-growing websites not yet integrated with a provider, analyze merchant adoption patterns and flag potential fraud risks based on site characteristics.

Beyond individual use cases, Shtayer framed the initiative as part of a broader transformation in how enterprise data is consumed.

“We’re changing the paradigm of how data is being used,” he said.

As companies move from AI experimentation to embedded deployment, Shtayer suggested the differentiator will not simply be access to powerful models, but the ability to pair them with trusted, domain-specific data and clearly defined guardrails. In that environment, AI becomes less of a novelty interface and more of an operational intelligence layer guiding everyday business decisions.

For all PYMNTS AI coverage, subscribe to the daily AI Newsletter.