Renewal management is central to any recurring revenue business. The signals that indicate an account is at risk are usually already sitting in the CRM: contact titles, deal stages, call logs, account descriptions, and more. What's typically missing is something that reads all of these together, on demand, and tells the Customer Success Manager (CSM) what to do next.

The signal problem

A CSM managing forty accounts can't manually cross-reference contact titles, deal stages, call logs, and account descriptions for every account before every renewal conversation. Synthesising these signals consistently, at scale, is genuinely difficult. Deterministic tools alone don't help here either. You can't build a blueprint transition that reads an account description and judges whether it contains churn language, or a kiosk rule that assesses the seniority of every associated contact and flags a gap in executive coverage. These are reasoning tasks, not rule execution tasks. But handing that reasoning to an AI agent with no guardrails introduces a different risk: a confidently wrong classification based on incomplete retrieval, or a recommendation that contradicts something the CSM already knows from a conversation last week.

The approach: bounded AI reasoning inside a deterministic review flow

This session builds a renewal risk classifier that runs entirely inside Zoho CRM using two Zia Agents invoked natively inside a Kiosk Studio flow. The first, a Risk Analysis Agent, reads the account and its related records, contacts, deals, tasks, events, and calls, then classifies the account as Green, Yellow, or Red, returns a confidence score, ranks the top signals driving that classification, recommends a retention play, and explains its reasoning in plain language. The second, an Assessment Logger, takes the CSM's final decision and writes a structured, timestamped note back to the account record, creating an audit trail of every renewal assessment made.

How it works for the CSM

1. Trigger the assessment. The CSM opens an account record and clicks Start Analysis, launching the kiosk, which passes the account ID to the Risk Analysis Agent.
2. The agent reads the CRM and classifies. The agent retrieves the account record, inspecting fields such as the description, health indicators, and subscription status, then pulls related contacts, deals, and recent activity via COQL, and evaluates the signals against a structured classification rubric. It returns a JSON payload containing the classification, confidence score, the top three ranked signals, a recommended play, the rationale for that play, and alternative plays.
3. The CSM reviews and decides. The kiosk shows this output on an Assessment Review screen, the human-in-the-loop checkpoint, where the CSM sees the classification, confidence level, driving signals, and suggested play, and can accept it or override it from a dropdown if they have context the CRM doesn't capture. Nothing is written to the database until the CSM clicks Log Assessment, at which point the Assessment Logger writes the structured note and the kiosk ends cleanly.

How it's configured

The kiosk's current record is set to the Accounts module. A screen with a single button lets the CSM trigger the whole flow from the account record. A Zia Agent component performs the synthesis, passing the account ID in the user query, tested and mapped before use. A review screen shows the agent's output alongside a dropdown of override options, with nothing committed until the CSM clicks Log Assessment. On that click, a second Zia Agent component is called, passed the chosen play, the decision on whether to notify a VP, and the first agent's output, and this agent both adds the note to the account record and, if selected, sends a Cliq notification to the VP. A success screen confirms the outcome and shows the chosen play, and a separate screen handles the agent's failure path, advising the user to contact the admin. Once configured, the kiosk is published to the Accounts module's record detail page.

The audit trail

Every assessment is captured in two places. The structured note on the account record shows the classification, the recommended play, the CSM's final decision, and whether they accepted or overrode the agent's recommendation, visible to anyone opening that record. Separately, every agent execution is captured in Zia Agent Studio's Observability tab, where you can drill into any session and inspect the agent's step-by-step retrieval, the fields it evaluated, the classification logic applied, and the exact JSON it returned, useful if a CSM flags an unusual recommendation or you want to understand why a specific account came back Yellow instead of Green.

Where to take it further

This first version was kept deliberately narrow in scope. A natural next addition is a product usage signal, since usage decline close to a renewal window is often the earliest reliable churn indicator, provided a product usage module is already linked to the Accounts and Contacts modules. Traversing a linked Subscriptions record would also add renewal window dates and contract value context, sharpening confidence scoring further. On the interface side, a custom SlyteUI component could replace the plain Assessment Review screen with a colour-coded classification banner, a clearer signal display, and an inline override dropdown, making the assessment easier for CSMs to act on at a glance.