How JAKALA Cuts Campaign Cycles from 7 Days to 24 Hours with Claude Agents

JAKALA is a data and AI company headquartered in Milan, operating across more than 30 countries with 3,600+ staff. The firm serves clients in banking, insurance, luxury hospitality, telecommunications, cruise lines, and education. Its central offering is the AI Factory: a governed delivery model that pairs proprietary data infrastructure, in-house AI products, and production agents to replace labor-intensive consulting with AI-augmented service.

For more than a decade, JAKALA had been building hundreds of AI and ML models in-house. But scaling them into repeatable, regulated delivery exposed two structural problems. Inside the Activation business—which handles media, campaign, and engagement delivery—senior planners on complex multi-market accounts were spending 40 to 60 percent of their weeks pulling data from 8 to 12 systems, hand-consolidating it in business intelligence tools, and building scenario decks before a single strategic recommendation reached a client. A complete campaign optimization loop took 5 to 7 working days per iteration. Separately, generative AI engines were reshaping how customers discover brands, and JAKALA’s clients had no way to measure their visibility inside those engines.

After evaluating leading frontier models on representative workloads—documents and prompts up to 200,000 tokens, multi-step agents with tool calls, and synthesis tasks graded on factual accuracy and auditability—JAKALA selected Claude as the reasoning core of the AI Factory. The deciding factor was consistent, trustworthy behavior inside multi-step agentic chains at the reliability threshold regulated clients in banking and insurance required. Today Claude runs on two tracks: Claude Enterprise serves more than 3,000 employees across data science, consulting, planning, engineering, and corporate functions; and a network of Performance Agents built on the Claude Agent SDK handles end-to-end campaign intelligence per client. Each agent connects via MCP to client ad accounts and JAKALA’s geo-intelligence engine, reasons across the combined data, and proposes specific optimization actions for human review. Claude Opus 4.6 handles the most demanding analytical work, Claude Sonnet 4.6 serves as the default for agentic workflows, and Claude Haiku 4.5 handles high-volume, latency-sensitive steps.

The shift was immediate on accounts running the hybrid agent model. Campaign optimization cycles—previously 5 to 7 working days of data assembly, scenario building, and presentation preparation—compressed to under 24 hours. Roughly 70 percent of the senior time previously absorbed by data collection and reporting moved to strategic and advisory work. Client satisfaction scores rose 35 to 40 percent. The agent deployment process itself matured: JAKALA’s first production agent took a full quarter to go live; by the fifth, the team had a repeatable pattern that now ships new agents into client environments in weeks.

The heaviest Claude users inside JAKALA turned out to be directors and partners, not junior staff—the agents absorbed the assembly work, concentrating value where senior expertise matters most. JAKALA has begun extending the AI Factory into private equity operations, applying Claude to due diligence and portfolio management for Ardian and additional European PE firms. “AI at JAKALA is not a pilot anymore,” said Marco Di Dio Roccazzella, Global Managing Director of Data and AI. “It is the default operating mode of the company. The roadmap is not about adding more AI to JAKALA—it is about making the AI Factory the reference operating model for enterprise AI transformation in Europe.”