SaaStr put an AI agent on 1,000 ignored leads; it reports $2.7M closed

SaaStr, which runs conferences and a media business for software founders, pointed an AI sales agent at the inbound leads its reps had never answered, and says $2.7 million in closed revenue and $3.5 million in pipeline followed. The figures are in a [customer story Salesforce published on 31 July 2026](https://www.salesforce.com/customer-stories/saastr-ai/agentic-outreach/); Salesforce sells Agentforce, the product the agent is built on.

The leads were people who had asked about sponsorships, tickets or partnerships and "never heard back", the story says: the reps were working six-figure sponsorship deals, and activity logs showed the smaller requests going unworked. "Our sales team decided to ghost them to focus on higher intent or bigger budget leads," says Jason Lemkin, SaaStr's chief executive. The agent, named Hexi, sent more than 3,000 emails to over 1,000 contacts, about three each. Salesforce reports a 72% open rate and a response rate above 10%.

How it is set up

• Hexi writes each email from nine years of records in SaaStr's CRM, the sales team's customer database: what the person asked for, which events they attended and what they have looked at on the website since.
• It may not book meetings, discount below a set floor or promise a speaking slot. When a conversation heads there, or a lead looks valuable, it sends the lead and its history to SaaStr's Chief AI Officer, Amelia Lerutte, in Slack. One such hand-off closed at $250,000, about 9% of the $2.7 million.
• The first run used 1,250 leads uploaded by hand as a spreadsheet file. Since May 2026 every event sign-up goes to the agent automatically, with "no human in between".

A second agent, built on Qualified (an inbound-sales tool Salesforce has since acquired), answers visitors on the event website. Salesforce says it [booked 614 qualified sales meetings](https://www.salesforce.com/customer-stories/saastr-ai/agentic-attendee-event-support/) for the SaaStr Annual conference onto reps' calendars, out of 402,000 conversations: one meeting for roughly every 655 conversations (402,000 / 614). It gives an "average ticket size of roughly $85,000" and does not say how many of the 614 meetings became sales.

The people side

The story says SaaStr has three full-time employees and more than 20 AI agents, and also refers to "sales reps". It does not say whether those reps are among the three, are contractors or work part-time, so the size of the human sales team is unknown. No job changed hands that the story reports: the agent took work nobody was doing. Lerutte reviews escalations and rewrites the agent's instructions; Salesforce's own engineers built the first agent with her in three weeks.

The catch

Salesforce's pages disagree with each other. The [overview](https://www.salesforce.com/customer-stories/saastr-ai/) says the $2.7 million was "deals sales reps never touched"; the detail page says the agent "hands them off to a sales rep when they're ready to buy" and calls the $250,000 deal "human-closed". The number of leads is "nearly 1,000", 1,250, "over 1,000 contacts" or, on [a third page](https://www.salesforce.com/customer-stories/saastr-ai/sales-ai-agent-data/), "more than 3,000 previously unresponsive leads".

The story gives no period for the $2.7 million, no count of deals, and no account of how a sale was credited to the agent rather than to the rep who closed it or to a buyer who would have come back anyway. Cost is absent: no licence price, no usage charges, no value for the vendor engineers' time. One failure is on the record: in testing, an event-help agent "would invent scheduling availability" until Lerutte added an instruction telling it to admit uncertainty. The pages carry no later figures as of 15 November 2026.

What to do if you run sales or revenue operations

• Have your CRM administrator pull every inbound lead from the past 12 months with no logged reply. Count them and total the value of what they asked about. If the list is short, the fix is a routing rule, not an agent.
• If you pilot an agent on that list, hold back a random tenth that gets no outreach, and compare closed revenue between the two groups after 90 days. SaaStr's story has no such comparison.
• Write down what the agent may not do before launch (prices, discounts, meeting times, commitments) and name the person who receives its escalations.