SaaStr found $2.7 million in CRM leads sales had left behind

SaaStr had nearly a thousand people in its CRM who had explicitly asked for information and then received no reply. Salesforce says AI-powered follow-up to that overlooked segment produced an additional $2.7 million in closed revenue and $3.5 million in pipeline.

The interesting part is not that an AI wrote emails. The value already existed in the customer history. What was missing was the capacity to restart those conversations and hand them to sales at the right moment.

Source: Salesforce — Agentic outreach turns SaaStr’s warm leads into $2.7 million in sales

The revenue did not come from a new channel

SaaStr runs a large community, media business, and events platform for SaaS companies. According to Salesforce, three full-time employees run the operation. Requests about tickets, sponsorships, and partnerships arrive throughout the year, but salespeople prioritized the largest deals. Activity logs showed that lower-value contacts were left untouched.

These were neither cold prospects nor active sales conversations. They had already raised their hands, shared their details, and entered the CRM. The case is therefore better described as reactivating dormant demand than as ordinary prospecting.

SaaStr's own early account describes roughly 1,000 overlooked leads and about 3,000 emails per month. The first useful question was not “can AI find new customers?” It was “which customer signals are we already paying to collect and then allowing to decay?”

Source: SaaStr — SaaStr and Agentforce: It’s Early, But the Results Are Really, Really Strong

What the AI sales agent actually did

An AI sales agent is a system that can read customer data, write and send follow-ups, and carry work across several steps. SaaStr named its agent Hexi.

Hexi did four things that are difficult to reproduce with a generic email blast:

• used up to nine years of CRM history, previous interactions, and event attendance as context
• adapted each email to the company and earlier relationship
• queued follow-up by time zone and appropriate sending time
• handed the contact to a person when buying intent became clear or the question became more complex

Salesforce says the first version was built in three weeks. SaaStr manually uploaded 1,250 leads to test the approach. In May 2026, Data 360 replaced that manual import so event registrations could flow directly into follow-up.

That sequence matters. The early pilot tested a bounded use case before SaaStr removed the manual step. The result did not come from switching on full automation across the entire CRM database on day one.

Source: Salesforce — SaaStr customer story

The handoff created the business value

Salesforce highlights one $250,000 deal. A company that had previously bought sponsorships had replaced its entire marketing team. Hexi identified a relevant contact, restarted the conversation, and then passed the full context to a person who closed the deal.

This shows the economically interesting division of labor. The agent restored contact and surfaced the signal. The salesperson took over where relationships, negotiation, and commercial judgment mattered more.

SaaStr also limited the agent's authority. It could not book meetings, discount below a set floor, or promise speaking slots. Salesforce Flows handled deterministic events, while agent instructions covered more open-ended judgment. When a contact needed human attention, a Slack notification included the history and reason for the handoff.

Those limits are worth noting, but they are not the main point. The handoff did not begin with an empty message saying “someone replied.” It arrived with usable commercial context.

Source: Salesforce — SaaStr customer story

A 72% open rate is not the business case

Salesforce reports more than 3,000 emails to over 1,000 contacts, a 72% open rate, and a response rate above 10%. Those are strong attention metrics, but they do not show profitability on their own.

Open rate is also an imperfect measure. Privacy protections and automatic image loading can affect the count. An opened email can still be irrelevant.

The measurement chain therefore needs to continue:

• qualified replies, not every reply
• human handoffs that were genuinely worth sales time
• pipeline created under a clear definition and time period
• closed deals attributable to the reactivation
• contribution margin minus software, implementation, and ongoing work

The $2.7 million in closed deals is the most consequential number in the story. But without cost, margin, and a comparison period, it does not produce a verified return on investment.

What the customer story does not prove

Both Salesforce and SaaStr benefit from presenting a successful outcome. The figures are vendor- and customer-reported; this is not an independent impact study.

The sources provide no control group, complete conversion baseline, or public method for attributing the $2.7 million. They also omit implementation cost, the specific license cost for this deployment, gross margin, and the number of sales hours required after handoff.

SaaStr's early account says enterprise tools in this category often start around $50,000–$100,000 per year once licenses, training, and onboarding are included. That is SaaStr's broad market estimate, not a disclosed price for Hexi. SaaStr also received help from Salesforce's forward-deployed engineers and described continuous training as necessary.

None of this makes the result uninteresting. It changes the question from “does it work?” to “which part can we realistically reproduce with our data quality, deal size, and operating effort?”

Sources: Salesforce customer story and SaaStr's own early account

Look for neglected demand before buying new technology

This case does not begin with Agentforce. It begins with a list of people who had already shown intent but whom the organization lacked the time to help.

A useful economic model is simple in form: contactable dormant leads × qualified response rate × close rate × contribution margin per deal, minus implementation and operating cost. Every factor needs your own data. SaaStr's outcome is not a sensible default forecast for another company.

The initial decision brief should therefore show how many contacts are genuinely reachable, why they fell out of the process, what a restarted conversation could be worth, and who takes over when it becomes commercially important. Only then can you judge whether an agent, simpler automation, or a different division of work is the right answer.

If that opportunity exists in your CRM, Tool Forge can help you measure and build a bounded reactivation workflow without turning one vendor case into a revenue promise.

SaaStr did not automate sales as a whole. It automated the absence of follow-up. That is a narrower idea—and probably a more valuable place to begin.