Scaling customer success
The problem
A growing software company had built a strong self-serve motion, but the customer success team behind it couldn't keep up with the volume. The math didn't help: average account sizes were small enough that hiring more CSMs would have wrecked the unit economics, so the team stayed lean. The result was predictable. Routine questions ate the day. The high-leverage moments, the ones that quietly drive retention and expansion, were going unseen and unactioned. Net retention had started to slip, and the cost structure couldn't carry a bigger team.
The engagement
Build a customer success layer that could scale with the business without scaling headcount. The brief landed in three parts: respond to customers in seconds, hold retention through the next phase of growth, and surface the expansion conversations the team didn't have bandwidth to catch.
What we did
We started by mapping the customer journey end to end with the team: where questions came in, how product usage trended, what billing signals meant in practice. The discovery surfaced where customers were getting stuck and where retention and expansion moments were going unseen.
The platform layer is Fin (formerly Intercom) at the surface, with a Claude-powered middleware between Fin and the company's Snowflake master data model. Snowflake was already the source of truth for product, billing, and account data, so anchoring the middleware there meant every agent response could draw on real account context, not a partial copy. Routine questions get answered in seconds. Routine actions (plan changes, billing updates, password resets) execute end to end, with human checkpoints on anything sensitive.
We extended the same surface upstream into onboarding so new customers felt the responsiveness on day one. Usage-signal monitoring sits on top of the Snowflake layer: the agent engages on a defined threshold, or hands off to a human when the deal warrants the touch. Dashboards track resolution, satisfaction, escalation rate, and revenue impact in real time, so the team tunes the system weekly rather than guessing.
The impact
AI ended up handling 85% of customer interactions. Response time dropped from 5+ hours to under 4 minutes. Retention rose from under 80% to over 90%. And in the first 90 days, the system surfaced $120K in incremental expansion revenue, from opportunities the team didn't have bandwidth to catch before.