HelloFresh
How an AI agent built on MCP redesigned the paid social acquisition of a company that invests over 100 million euros a year in advertising.
-
Industry
Food subscription
-
Services
SCALE Paid, GROUND Data
-
Year
2026
-
View project
hellofresh.com
In organisations operating at HelloFresh’s scale, paid social does not stop working as it grows — it stops being governable with the tools and processes it started with.
HelloFresh invests over 100 million euros a year in Meta campaigns distributed across 18 markets and dozens of brands, with a paid social team managing thousands of active ads simultaneously. At the scale reached, traditional tools had stopped handling the required workload: reporting too slow to inform weekly decisions, launches still managed manually, configurations with a significant error rate, and no feedback loop between creative performance and the team producing the creatives.
What was needed was a system that did not replace people but gave them back the time and visibility to make better decisions.
An AI agent built on Model Context Protocol architecture
We developed an AI agent accessible via chat on Slack, built on a Model Context Protocol architecture that allows the system to autonomously query campaign data, interpret it in the client’s business context, and act directly on advertising platforms.
The agent serves three levels of use from the same access point:
– The operational level, which launches ads starting from a creative folder and configures campaign parameters in minutes.
– The analytical level, which queries granular performance numbers for individual ads and adsets in natural language.
– The strategic level, which receives recommendations on budget allocation, scaling timing and ad stops, generated by predictive models trained on historical data.
The entire system runs on a new data infrastructure built to replace the Google Sheets previously in use: a validated database with codified naming conventions and automatic checks on incoming data quality.
This was not a tool problem — it was an orchestration problem
The challenge was not to automate a single step, but to restore coherence to a process that, at the scale reached, had fragmented. Ad launches, reporting, budget decisions and creative feedback lived in separate tools and logics, with distributed ownership and very little shared visibility.
The agent was designed as an orchestration layer: a single access point capable of connecting operations, analytics and strategy, adapting to different seniority levels within the team and returning to each person the information in the format most useful for their role.
The value is measured in time freed, errors avoided and costs reduced
The system went live in September 2025 and the results are visible both in operational efficiency metrics and in advertising performance metrics.
– CAC: 25% reduction at equivalent acquisition objectives, with 15% more budget allocated
– Time spent on reporting and analysis: 90% reduction
– Average time to launch a new ad: from one or two days to approximately 5 minutes
– Campaign configuration error rate: from a 10-20% range to below 1%
TOV note: these numbers should remain as they are, without superlatives around them. The data speaks for itself. Avoid “extraordinary”, “incredible”, “transformative” in proximity.
A system that with scales the organisation
-
Operational efficiency
The same team now manages a significantly higher volume of work without increasing headcount, and the time freed is reinvested in high-value activities such as strategic analysis and new creative testing.
-
Data quality and process quality
The migration from Google Sheets to a validated database drastically reduced configuration errors and made reporting reliable for all levels of the organisation, from the CMO to the operational team.
-
More informed budget decisions
The strategic level of the agent, supported by predictive models, assists budget decision-makers in scaling and reallocation choices, with recommendations based on data rather than intuition.
When paid social grows, the problem shifts from budget to the governance of the system that manages it
If your team spends more time configuring campaigns than deciding how to optimise them, the bottleneck is not the investment — it is the system of tools and processes surrounding it.