Make every matchday feel closer to every fan.

The blocker is rarely data. Ticketing history, favourite players, purchase behaviour and engagement patterns already sit in the CRM. The blocker is that acting on them is manual: someone has to decide the segment, write the variant, check it against brand rules and schedule it — per campaign, per language, per match.

So the club sends one newsletter to the whole base, because that is what one person can produce before kickoff. The segmentation exists on a slide and never reaches the inbox. An agent changes the unit of work: it composes per fan from the same source, and a human reviews the rules rather than the variants.

All of it drawn from the same grounded source, so the stat in the newsletter matches the stat in the push and the one on the site.

A general model can understand a fixture and use tools. It does not automatically have access to a club's consented fan history, channel preferences, or operating rules. Those customer-approved inputs are what allow a configured workflow to shape a more relevant experience.

That also sets the integration bar: the agent has to read the CRM, respect consent and channel preferences, and write back what it sent so the next decision is better informed. A system that only generates, without closing that loop, ends up producing personalized content nobody can measure.

Four constraints, and none of them are about model quality.