Two different roles in the sales process

Sales Agent was presented as an agent that could research and prioritise leads, arrange meetings and contact prospects. Sales Chat was designed to help sellers understand an account, identify deals at risk and prepare for customer meetings by combining CRM, presentation, meeting, email and web information. Microsoft said both could be accessed through Microsoft 365 Copilot and Copilot Chat.

The agents could connect to Dynamics 365 and Salesforce, but they were not released for routine use on 5 March. Public preview was planned for May 2025. The announcement therefore described product direction and an invitation to test, rather than evidence of finished functionality or achieved sales results.

Independent perspectives separated promise from adoption

TechTarget placed the launch within competition for autonomous sales agents and highlighted access to both Microsoft and Salesforce CRM data. Computerworld noted that Sales Chat could create documents from several sources, while Sales Agent could follow up leads. The coverage showed vendors competing both for the seller’s workspace and for automation of customer engagement itself.

In December 2024, Anthropic recommended simple, composable patterns and distinguished predefined workflows from more autonomous agents. Forrester had also warned about unclear terminology and implementation risk in the agent market. Those perspectives supported starting with a bounded task and increasing autonomy only when evidence justified it.

Internal assistance and external action carried different risk

Preparing a meeting brief was easier to control than sending a message to a customer. In the first case, a seller could read, correct and select what to use. In the second, an incorrect price, tone or recipient could directly affect the relationship. The organisation therefore needed different authority and approval points for the two agents.

The data foundation also required practical testing. Duplicate contacts, outdated prices, missing consent or weak links between email and CRM could produce personalised responses that were still wrong. A pilot needed a known set of leads, documented permitted sources and a comparison with a human-controlled group.

Useful measures went beyond messages sent or time saved. Sales management needed to track qualification accuracy, meetings that actually occurred, customer reactions, corrections and how often a seller intervened. Only those measures could show whether the preview improved selling or merely increased activity.

Customers also needed a clear route to a person and a way to correct information. An agent working continuously could scale good and poor practice equally fast. A clear sender identity, controlled frequency and respect for opt-outs were therefore part of the sales design.

Sources

Microsoft: New sales agents in Microsoft 365 Copilot, 5 March 2025

TechTarget: Microsoft rolls out sales agents and Salesforce connector, 5 March 2025

Computerworld: Microsoft launches sales agents, 5 March 2025

Anthropic: Building effective agents, 19 December 2024

Forrester: The state of AI agents, 4 October 2024

Further reading

Read more: What is the difference between a Copilot, an agent and a language model?

Join the discussion

Where in your sales process do you spend the most time on manual follow-up? Share your experience.