# How We Built an Outbound Voice AI Agent to Recover Churning SaaS Customers *Nikhai Jaysen · August 28, 2026* > How SpaceBlanket.AI built a trigger-based outbound voice AI agent that calls stalling trial users at the right moment — and what changed. Source: https://www.spaceblanket.ai/blog/outbound-voice-ai-agent-saas-churn-recovery A B2B SaaS company was watching trial accounts drift to expiry without any human touching them. We built a trigger-based outbound voice AI agent that calls every stalling trial at the right moment — and handles the entire first-touch conversation automatically. ## The Churn Signal Nobody Was Acting On A B2B SaaS company came to us with a problem that looked like a product problem but was actually an operations problem. Their data showed that users who didn't reach a specific activation milestone within the first 10 days of a trial almost never converted to paid. They knew this. They had the data. They were watching accounts drift toward the end of the trial window — and doing nothing about it until the trial expired. The reason wasn't indifference. It was capacity. Their team was two people handling everything from support to upsells to new user onboarding. Proactive outreach on stalling trials wasn't happening because there was no time for it. By the time someone noticed an account had gone quiet, the trial window had closed. They wanted to fix this without hiring. What they needed was a system that would notice the signal and act on it automatically — calling the user, understanding why they'd stalled, and either solving the problem on the call or routing them to a human who could. ## The Approach: Trigger First, Qualify Second The core design decision was to treat this as a trigger-based outbound operation, not a scheduled calling campaign. Rather than calling everyone at day 7, the **voice AI agent** would call the moment a user crossed a specific inactivity threshold — 72 hours without meaningful product engagement, combined with an activation step still incomplete. This mattered because the window for a recovery call is short. A user who hasn't logged in for three days is still recoverable. One who hasn't logged in for nine days and hasn't replied to two emails usually isn't. The agent needed to reach people while there was still something to save. The qualification logic on the call itself was simple: understand why they'd stopped engaging. The answers fell into predictable categories — too busy to set up, a specific configuration question blocking progress, a feature expectation that wasn't met, or genuine disinterest. Each of those had a different next step, and the agent was designed to route accordingly. ## What We Actually Built The system had three integrated layers: - **Trigger logic in the CRM.** A workflow rule monitored trial user engagement data pulled from the product's analytics. When a user hit the inactivity threshold and had a valid mobile number on file, the CRM pushed that user to the outbound call queue via webhook. - **The voice AI agent itself.** Built on Retell AI, the agent opened the call by acknowledging the signup and the recent trial — not with a generic opener, but referencing the product category and how long the trial had been running. It then asked one question: what had gotten in the way of getting set up? The conversation branched from there based on the answer. Users who had a configuration question got a direct answer if it was within the agent's knowledge base, or an immediate warm transfer to a human support rep with the context already attached. Users who said they hadn't had time were offered a 15-minute setup call booked directly into the calendar from within the conversation. - **CRM integration and handoff.** Every call outcome — completed, no answer, transferred, booked — was logged back into the CRM as a structured record. A booked call changed the trial user's status to "high intent, returning," which triggered a different email sequence and flagged the account for account executive follow-up. For users who didn't answer, the agent called back once four hours later. After two missed calls, it sent a short SMS offering the same 15-minute setup session. No further outreach ran automatically — the CRM flagged those accounts for a human to review at the end of the week. ## What Changed The most immediate change was that the team stopped making manual outreach decisions about stalling trials. The agent handled the first-touch on every triggered account. The team's involvement started at the moment a user was transferred or booked — which meant every human conversation had context and a reason to happen. Trial-to-paid conversion on accounts the agent contacted improved over accounts in a comparable period where the team had relied on email follow-up alone. The biggest shift wasn't the conversion rate itself — it was the distribution. Users who had a technical question and got it answered on the call converted at a high rate. Users who said they hadn't had time and booked a setup session converted at an even higher rate. The agent was surfacing intent signals the company hadn't previously been able to act on at scale. The system runs continuously without oversight. When a trial account hits the threshold at 2am or on a Saturday, the call goes out. The team reviews logged outcomes weekly and occasionally adjusts the qualification logic when new answer patterns emerge. We build outbound **voice AI agents** like this for SaaS companies that have usage data to trigger on but not the headcount to act on it manually. We've built exactly this for SaaS teams at different stages — early-stage with a small ops team, and growth-stage with a full customer success bench. [Get in touch](/contact) and we'll show you what a trigger-based outbound build looks like for your specific trial or churn-risk flow. ## Related - [All articles](https://www.spaceblanket.ai/blog) - [Contact](https://www.spaceblanket.ai/contact)