What agentic support actually saved

The problem

Support volume scaled with users, and users were in the millions. Resolution time was the number customers actually felt, and headcount was the only lever anyone had ever pulled on it. The obvious move, deflection, is also the one that quietly destroys satisfaction when you get it wrong.

What we did

• Built agentic workflows into the support stack rather than putting a chatbot in front of it, so the automation resolved things instead of deferring them.
• Set CSAT as a hard guardrail before anything shipped. Speed bought with satisfaction isn't a saving, it's a transfer.
• Layered ML-driven personalisation across the lifecycle in parallel, so what a user received matched where they actually were.
• Put AI content and translation tooling in the marketing team's hands, which let a single content manager produce and localise hundreds of SEO articles a month.

The call I had to make

The guardrail was the decision. We set a CSAT floor before we automated anything and we were prepared to give speed back to hold it. We never had to. Being willing to is what made the number trustworthy.

Impact

• Resolution time down 75%
• $300K a year out of the cost base
• CSAT sustained at 80%
• +15% new user activation and +10% NPS from the lifecycle personalisation