Declining usage means different things for different customers
Renewal risk scores treat any drop in usage as fully negative. For a customer who has finished their rollout and settled into a steady state, that drop is expected. It is a serious problem when the customer is only halfway through implementation.
The score cannot tell those customers apart, because the number is the same. The path they have taken is what distinguishes them. That means comparing the customer to their own history rather than to a benchmark.
The recommendation matters more than the risk level. You can choose among four responses: a check-in, a training session, a commercial conversation, or no action.
Doing nothing must always remain an option. If a renewal program contacts every account that has had a dip, your attention looks like a machine is handling it.
How the agent is built in Agent Studio
The comparison standard is specific to each customer, so the system prompt must say so. If it does not, the model uses the average across all customers, because that is the more typical way to put it.
Prompt skills list your intervention options and, more importantly, what each one costs. Commercial concessions and training sessions are not interchangeable, and the agent must know which of them you are unwilling to offer.
A Start Scheduled flow opens a window just before the renewal date. HubSpot or Salesforce supplies the account. An HTTP Request node retrieves the usage figures, and the Agent Chat node requests an assessment.
Recommendations go to Notion as a short document with the evidence attached. Accounts the agent regards as high-risk go to Slack immediately, because a window means you act inside it.
The customer is never contacted automatically, because commercial concessions require human approval. Put that boundary in the flow as a Human in the Loop node, not as an instruction the model is expected to follow.
The only way to tell whether the agent is assessing risk from trends, or only matching decline patterns, is to compare those assessments with actual renewals each quarter.
What this agent is built from
- System prompt: the instruction to compare each customer against their own baseline, not a benchmark.
- Prompt skills: your intervention options and what each one costs you.
- HTTP Request (Util): pulls usage data. Salesforce and HubSpot supply the account context.
- Agent Chat (Util): a scheduled flow opens a window before each renewal date.
- Notion (Integration): receives the recommendation with evidence attached.
- Slack (Human in the Loop): carries high-risk accounts immediately and gates every concession.
- When an ActionFlow is enough: If you need renewal date reminders, those are a workflow, and they are the higher priority. Build those first.
Frequently asked questions
Start building AI workflows
Create a free account, open a template or a blank canvas, and run your first ActionFlow.
Newsletter
Get product updates
New nodes, agents, and product notes. We send mail only when we have something worth opening.