Agentforce for producers: enabling AI inside the CRM

First, a word that matters here. In American insurance, a producer is the person who wins and places business. They "produce" revenue: finding prospects, understanding their risks, and putting the right cover in place. Think salesperson and broker rolled into one. Their time is expensive, and most of it should go on clients, not on hunting for information.

We worked with a global insurance broker to test whether Salesforce Agentforce could give producers that time back. This is the story of a proof-of-concept and a six-week pilot: what we built, how we kept it safe, and how we set up the decision on what comes next.

Before a renewal conversation or a prospecting call, a producer needs a clear picture of the client. Who they are. What's changed lately. When their cover renews. Who currently holds the business. Which producer internally is the best fit. That information exists, but it's scattered across the CRM, Zywave, an internal knowledge site and the open web, so producers rebuild it by hand, one tab at a time.

Most AI tools people reach for don't touch the CRM. They're clever, but they don't know your customers, so someone still has to feed them context and stitch the answers together. Agentforce itself isn't new. What's new is Agentforce Coworker, which we enabled here for the first time, alongside the Employee Agent. Together they mean enabling AI inside the CRM rather than bolting it on beside it, with the AI sitting on top of the data that's already there.

The opportunity was straightforward. If the briefing writes itself from trusted data, producers walk into calls better prepared and spend their time on the conversation instead of the prep. Looking further out, the same connection lets AI update records directly in Salesforce, which starts to move producers away from manual form-filling.

The centre of the pilot is the meeting-prep summary. Ask Agentforce to prepare for a client, and it pulls together more than 14 fields from four places into one briefing: the CRM, third-party enrichment from Zywave, an internal knowledge site, and information available on the web. The summary surfaces the renewal information, the best-fit producer, the incumbent broker, and the latest signals on the account, including things like changes in senior personnel. It's tailored to the client in front of you, not a generic dump, so producers save time on prep and can push further because the AI already holds the full context from the CRM.

Beyond the summary, we enabled two further capabilities. The first is the Agentforce coworker, which puts an assistant alongside producers as they work. The second is the employee agent, made up of three subagents: a general CRM agent for questions across records, a record summary agent for a fast read on any single account, and a web search agent for information that lives outside Salesforce. Between them they cover the everyday questions producers ask, whether the answer sits inside the CRM or out on the web.

None of this ships without governance. The change went through a robust review led by an Architecture Review Board, which brought compliance, legal, infosec, data protection and AI governance around one table. Insurance is regulated, and the broker's security posture had to hold, so the board's job was to confirm the AI could be deployed safely and to the standards the business is held to. That sign-off is what let the pilot run on real data with real users.

We kept the team lean, around 10 people, and the scope honest. Fifty producers took part as pilot users, running a six-week production pilot on live accounts. At the end of the six weeks we check consumption and the value producers are getting, then decide together whether to continue as-is or build further enhancements on top.