Connecteam built an internal AI sales agent on HubSpot data that tells reps which deal to work next
In Geektime's AI Insider column (promoted content), Dotan Fried, AI AgentOps at Connecteam, describes an internal web app he built for the sales team. It connects to the sources the team already uses, HubSpot, correspondence, call and meeting summaries, tasks and product usage data, organizes the information on each lead into one business context, and uses Gemini and Claude models to spot risks and opportunities and recommend the next action: which deal to work and what to write or say. Recommendations appear in a dashboard and in Slack, and the agent also prepares reps for meetings and suggests how to revive a lost deal. The tool was built gradually over six months, developed in Cursor with the model chosen by task complexity. He says sales actions rose 26 percent and phone calls 23 percent, and the time to understand a deal dropped from 10 to 15 minutes to about 4 minutes. His tip: start by asking where the company is losing money, not which agent you could build.
Otomatzia editorial · Published
Source published: · Geektime
