SaaStr Built an AI Renewal Agent That Makes Custom Decks for Every Account. Their Smallest Customers Responded Best.

Every SaaS company with a sponsorship or enterprise tier runs the same triage: the biggest accounts get custom renewal decks built by a human who spent time with the CRM, combed through event data, and wrote messaging tailored to each contact. Everyone else gets a templated email and a prayer. The cutoff isn’t cruelty — it’s arithmetic. A week to build 25 custom decks is a week that doesn’t exist.

SaaStr has replaced that arithmetic with an AI renewal agent. The result inverts the expected outcome.

What the Agent Does

The renewal agent assembles custom decks and renewal emails for every account in SaaStr’s sponsor database — not just the top tier. The scope means accounts that previously received template treatment now get decks that reference their specific outcomes, their team’s activity across SaaStr events, and their presence in SaaStr’s content archive.

The agent writes the narrative first. Before generating a deck, the agent produces the intended pitch narrative for human review. This step allows corrections before any materials are built — a critical override point, because the agent will occasionally fabricate numbers despite explicit instructions not to. The narrative review step catches these before they reach the prospect.

Data aggregation spans six sources: Salesforce CRM (contract details, lifetime value, call history, email open rates), a WordPress API (published articles mentioning the customer), social media accounts, a podcast archive (customer name mentions), Bizzabo (lead counts, attendee seniority data by sponsor tier), and Gmail via a Replit connector for email context. The combination produces a contact-level picture that a human researcher would need hours to assemble for a single account.

Deck generation uses Gamma’s API to build the final presentation, preserving actual customer logos and brand templates rather than generating approximations. The agent addresses individual contacts within an account with messaging customized per contact type — champion, economic buyer, renewal contact.

The operational model is human-gated: people handle initial outreach. The agent fires the custom deck when a prospect replies. Human attention is spent on warm accounts; the agent handles the collateral for every temperature.

What Happened When They Tested It

The expected result would be that personalization scales the performance of the top tier: diamond sponsors (largest accounts) see better reply rates, smaller accounts see modest improvement. That’s not what SaaStr observed.

Silver sponsors — the smallest tier, the accounts that previously received template treatment — replied at higher rates than diamond sponsors. The most plausible interpretation: diamond accounts already received enough attention from the human-managed process that incremental personalization added less. Silver accounts had been getting nothing custom; any actual personalization moved the needle more dramatically.

The Catch

The fabrication problem is real and not fully solved. Despite explicit constraints, the agent periodically invents numbers — performance statistics, attendance figures, outcomes that weren’t in the source data. The narrative approval step is a mitigation, not a fix. Any team deploying a similar agent should build verification into the workflow, not assume the agent produces accurate claims.

The human-only initial outreach model also means the agent’s impact is invisible until a prospect engages. Cold outreach still depends on people.

Source: SaaStr

Content created with AI assistance and reviewed for accuracy.