Autonomous onboarding nudge agent (L5)
An agent watches onboarding milestone data and acts without a CSM in the loop for the routine cases: stalled setup, unused seats, missed training. Humans handle exceptions and anything touching commercial terms. Every action is logged and reversible.
The steps
- 01
Pick the three autonomous cases
Tool: Manual
Only automate cases where the correct action is unambiguous and cheap to undo: setup stalled seven days, seats provisioned but never logged in, training booked but not attended. Everything else stays human. Owner: CS leader plus RevOps. DoD: three cases written with exact trigger conditions.
- 02
Give the agent bounded tools
Tool: Agentforce
The agent may send a templated nudge, book time on a named calendar, and open a task. It may not issue credits, change contract terms, or email an executive sponsor. Pitfall: open ended tool access. Enumerate tools explicitly. DoD: tool allowlist is documented and enforced in config.
- 03
Log every action to the record
Tool: Salesforce
Each autonomous action writes a timestamped entry to the account with the trigger, the message sent, and the outcome. If the CSM cannot see what the agent did, the agent gets turned off. Owner: RevOps. DoD: an account timeline shows agent actions inline with human actions.
- 04
Audit weekly against a holdout
Tool: Slack
Hold out ten percent of eligible accounts with no agent contact. Compare milestone completion and complaint rate. Kill any case that does not beat the holdout after two cycles. Owner: CS ops. DoD: weekly audit post with holdout numbers in the CS channel.
Tools in this playbook
- Manual
- Agentforce
- Salesforce
- Slack
Next playbooks
Unfamiliar terms are defined in the AI and Revenue Dictionary. Related frameworks live in the framework library.