The Agentforce readiness checklist: 10 things to fix before you launch agents
Agentforce agents are only as good as the data, permissions and processes underneath them. Use this checklist to find the gaps before your customers do.
Agentforce makes it surprisingly quick to set up an AI agent in Salesforce. Making that agent reliable enough to talk to your customers is another story. In our experience, most problems come down to the foundations, meaning data, knowledge, permissions and process, rather than the agent itself.
Data and knowledge
- Clean up your core records. Duplicates, outdated contacts and missing fields turn into confident wrong answers. Start with the objects your agent will read.
- Decide how much Data Cloud you need. Figure out which outside data (orders, billing, product usage) your agents need, and whether to bring it in through Data Cloud or call it through actions.
- Curate your knowledge base. Retire outdated articles, fill the gaps for your top contact reasons, and structure content so retrieval returns precise answers.
Security and trust
- Give agent users the least privilege possible. Agents should have access only to the objects and fields they need. Review sharing rules for anything exposed by accident.
- Configure the Einstein Trust Layer. Confirm that data masking, the audit trail and zero-retention settings meet your compliance requirements.
- Define escalation paths. Every topic needs a clear hand-off to a person, with the full context passed along.
Agent design
- Start with narrow topics. Pick two or three high-volume, well-understood requests instead of building a do-everything agent.
- Build actions around real processes. Flows and Apex actions should reflect how the work actually gets done, including policy limits like refund thresholds.
Testing and rollout
- Test against real conversations. Build test sets from past cases and run them in Testing Center before every change.
- Plan for the people side. Train your service team to work with agents, decide who owns agent performance, and agree on the metrics you’ll review each week, such as deflection, resolution and CSAT.
Where to start
A two-week readiness assessment is usually enough to score each area, prioritize the fixes and choose your first topics. From there, most companies can launch a first agent in a controlled channel within four to eight weeks.