AI agents that take routine policy service and claims intake off the queue, while licensed staff keep every coverage, underwriting and claim decision.
Agentforce lets insurers deploy AI agents that read policy and claim data in Salesforce and complete defined tasks, such as updating a mailing address, issuing a certificate of insurance or opening a first notice of loss. Agents can serve policyholders on the web or in messaging, and they can work beside service representatives by summarizing a claim file or drafting a reply. Each agent is limited to the topics and actions you approve, grounded in your own policy records and knowledge, and designed to pass the conversation to a licensed person when a request needs judgment.
Insurance conversations sound simple but often carry legal weight. A policyholder asking whether a loss is covered is asking something only the policy language and an adjuster can answer, and a casual reply from an agent could be read as a coverage position. That makes scope the central design problem. Agentforce handles it with topic instructions that separate informational requests from coverage interpretation, actions that can read coverages but cannot change them, and handoff rules keyed to line of business and state licensing. The Einstein Trust Layer adds masking and logging around each exchange, which matters when any interaction may later surface in a complaint file or market conduct exam. Independent agents are a second audience with different needs, so producer-facing agents are usually built separately.
Commercial policyholders and their brokers request certificates of insurance constantly, often with specific certificate holder wording. An agent can verify the requester, confirm the policy is in force, generate the certificate from an approved template and deliver it. Requests for additional insured endorsements or unusual wording go to an underwriting assistant instead, because those change the carrier's exposure and need human review.
When a policyholder reports a loss, the agent asks the questions the claims team needs: date, location, parties involved, photos and police report details. It confirms the policy number, creates the intake record and hands it to the claims system. It explains next steps in plain language but never estimates a payment or comments on whether the loss is covered.
Independent agents ask about appetite, submission status, commission statements and binding authority. A producer-facing agent can answer from underwriting guidelines and submission records, restricted to that agency's own book. Questions about exceptions or referrals route to the underwriter who owns the account, with a summary attached so the underwriter can respond without rereading the whole thread. Producers see only their own submissions.
Long claims accumulate notes, emails, estimates and recorded statements. An employee-facing agent can summarize the file, list open tasks and draft a status update to the insured for an adjuster to review and send. Adjusters stay in control of the message while spending far less effort reconstructing what happened on a claim they inherited from a colleague. Adjusters can edit the draft freely.
Before building any agent, decide which policy data it may see and in what form. Most carriers keep the policy administration system as the record of truth and expose a read-only summary of coverages, limits and status to Salesforce. Next, define the relationship model of named insureds, households, commercial accounts and producers, so the agent can verify who is asking and what they are entitled to see. Last, tag knowledge articles by line, state and audience, because an answer that is correct for personal auto in one state may be wrong in another, and the agent can only respect that distinction if the content carries it.
Supplies coverages, limits, billing status and document generation, so agents can answer questions and issue certificates without holding a separate copy of policy terms.
Receives structured loss reports from the agent and returns claim status, adjuster assignment and open requirements for policyholder updates during the life of the claim.
Stores declarations pages, endorsements and correspondence the agent retrieves or attaches, with retention applied under the insurer's own records policy and schedule.
Write topic instructions that let the agent explain what a policy document says and where to find it, while refusing to predict whether a claim will be paid. Test with phrasing policyholders actually use, since coverage questions rarely announce themselves, and log every refusal for review.
Some tasks, such as discussing specific coverage changes, may require a licensed person depending on the state. Map which topics are purely administrative and which need licensed handling with your compliance team, then build handoff routing on that map rather than on general complexity.
A self-service claims agent is also an easy channel for someone fishing for information. Require identity verification before revealing policy details, cap what the agent discloses about limits and send suspicious patterns to your special investigations workflow rather than letting the agent judge them.
Client names are anonymized.
It should not. Coverage decisions depend on policy wording, facts developed during investigation and state law, and an automated answer could create expectations or disputes. We design agents to explain the claims process, collect information and report status, then route any coverage question to an adjuster or licensed representative who can document a proper response in the claim file.
Pick a high-volume request with clear rules and low consequences if the agent errs, such as certificates of insurance, billing questions or address changes. Those tasks prove the data connections, verification steps and handoff paths in production. Once they run cleanly and reviewers trust the conversation logs, extend into intake flows like first notice of loss.
Yes, though the use cases shift. An agency might use an agent to gather renewal information from clients, answer questions about office procedures or prepare account summaries before a producer's call. Because agencies rely on carrier systems for policy data, integration is often the harder part, and some carrier data may only arrive as periodic downloads.
Conversation transcripts, the records the agent accessed and the actions it took are logged in Salesforce and can be retained according to your records schedule. Compliance teams can sample transcripts, review refusals and handoffs, and flag answers to correct in knowledge. That review loop belongs in the operating model from the first pilot, not as an add-on.
One onshore team with 150 Salesforce certifications, a Salesforce Consulting Partner since 2017.
A monthly brief for the people who own Salesforce, AI and revenue technology
What changed in Salesforce and AI this month, and what to do about it.
One email a month. Written by the consultants who deliver the work, not by a marketing team, for the leaders who make the technology decisions.
Consultant analysis, not vendor recaps. One click to leave.