SaaStr cracked autonomous sales by deploying AI on lost leads

This article summarizes an episode of SaaStr AI’s video series featuring its executives Amelia Lerutte and Jason Lemkin.

Amelia Lerutte, Chief AI Officer of SaaStr (left) and Jason Lemkin, CEO and Founder of SaaStr (right) / Photo credit: Amelia Lerutte

Company technology projects often fail when leaders plan vast, theoretical changes instead of fixing the daily problems that slow their teams down.

By using software to handle repetitive tasks and establish strict operational rules, companies can unlock unexpected value from the workflows their employees already hate.

Amelia Lerutte, chief AI officer at SaaStr, and Jason Lemkin, founder and CEO of SaaStr, build custom digital workflows to eliminate internal bottlenecks and chase sales leads that human teams typically ignore.

Building from broken daily processes

This approach to fixing broken processes starts small. The most enduring internal tools do not begin from huge corporate mandates, but from a simple wish to solve one annoying administrative task.

The most effective method involves isolating a single manual task and connecting a custom tool directly to company databases like Salesforce. Teams can bypass standardized software menus and letting the tool grow organically through daily use.

“A lot of these did not start as full-blown agents; it was just something where we wanted to replace something,” Lerutte says, adding that a simple dashboard built to replace their Notion setup eventually evolved into a much more capable system.

This organic growth proves that people only keep using tools if they actually want them. The first internal dashboard evolved into an automated marketing system because the tool was useful enough to warrant adding more data and decision-making power.

Bypassing traditional interfaces
Lemkin suggests that the simplest starting point is to spin up a basic AI builder and connect it to Salesforce.

By asking the software to build analytics views that aren’t natively available, he argues that teams can spend just an hour to build something highly useful that Salesforce could never do on its own.

To build these tools successfully and bypass traditional software friction, teams must execute these steps:

• Pick one specific workflow that has obvious problems: Good starting points include tracking projects, routing messages, gathering files, booking meetings, processing standard discounts, and conducting routine customer check-ins.
• Connect the workflow directly to company systems: The SaaStr team linked their custom tools directly to platforms such as Salesforce, Marketo, Bizzabo, Slack, Qualified, Resend, and Clerk to create a functional foundation.
• Design the first version as a read-only view or basic form: The initial launch must focus entirely on saving employee time before granting the software autonomy to make choices.
• Require daily usage: Regular use builds necessary context, uncovers missing data connections, and forces the team to build highly specific features that top-down planning meetings often miss.
• Allow independent actions only after building confidence: The internal tool known as 10K moved from providing basic spreadsheet views to generating campaign ideas only after months of reliable work.
• Let the role of the software reveal itself over time: Another internal project began as a static information website and organically grew into a comprehensive event management system by tracking visitor data and parking details.

Hidden biases in software design

While building these custom tools solves immediate bottlenecks, fixing simple problems often uncovers larger structural flaws in the underlying AI models. Giving the exact same instructions to different systems rarely produces the exact same answers.

The risks of unsupervised speed

Using software in ignored sales channels