AI customer success platform for a B2B SaaS company. churn down 38%, expansion revenue up $1.1M

The company sold a vertical B2B SaaS platform to roughly 600 customers across two adjacent industries. Annual recurring revenue was healthy and growing but two trends were eating the business from underneath. Customer churn was steadily climbing as the team scaled past the point where the founders personally knew every customer, and expansion revenue was stuck despite a clear pattern of customers using the platform in ways that signalled they were ready for an upsell. Both problems were really the same problem. the customer success team was reacting to events instead of seeing them coming.

The data to fix it existed. Product usage events were captured. Support history was captured. Billing data lived in the billing system. CRM activity lived in the CRM. The four sources had never been consolidated into a single account-level view because every attempt to do so had been deprioritised by the next release deadline. The CSM team was learning about churn risk from the cancellation request and learning about expansion opportunities from whichever account manager happened to be paying attention that week.

We built a unified customer intelligence platform with four interlocking components on top of a single customer data layer. The data layer normalises product events, support tickets, billing events, and CRM activity into one consistent account-level view that updates continuously. The health-scoring engine sits on top of that layer and produces a live health score for every customer, broken down into the underlying signals that drive it. adoption depth, breadth of feature use, support sentiment, billing health, and engagement cadence with the customer success team. Every score is explainable and every score recommends a specific next action.

The in-product AI assistant is trained on the company's documentation, changelog, and resolved support tickets. It answers user questions inside the product itself with citation-linked answers and escalates cleanly to a human when it is uncertain or when the question requires action it cannot take. The churn risk model scores every account weekly and routes the at-risk accounts to a CSM with a specific intervention playbook based on which signals are moving. The expansion intelligence layer surfaces accounts whose usage patterns indicate readiness for an upsell, with the specific feature-level evidence the account manager needs for the conversation.

An in-product assistant trained on the company's own resolved tickets behaves nothing like a generic chatbot. It answers in the company's voice, with the company's documentation conventions, and it knows which workarounds were officially blessed by the support team and which were one-off fixes that should not be recommended. Every answer carries citations into the documentation and a single click escalates the conversation to a human with the full context attached.

Forty-seven percent of inbound support volume was deflected inside the first quarter, with measured customer satisfaction on deflected questions actually higher than on questions that went to a human first. The Tier-1 support team moved their focus to the complex, high-touch tickets that always belonged to humans and the company avoided two planned support hires.

Net revenue retention climbed from 102 percent to 121 percent. Customer churn dropped by 38 percent against the prior twelve-month baseline. Expansion revenue grew by $1.1 million in the first year, almost entirely on the back of the expansion intelligence layer surfacing the right accounts to the right account managers at the right time. The payback on the build landed inside the second quarter and the company has since extended the platform with an executive-facing customer health dashboard for the board and an automated quarterly business review generator for the top tier of accounts.