The Onboarding Emails Were Opened. Nobody Finished Setup.

Nikhai Jaysen · September 29, 2026

A SaaS platform's onboarding emails had great open rates and mediocre completion rates. We built a voice AI agent that calls stalled signups directly, finds out what they're stuck on, and finishes the setup live on the phone.

Voice AI Onboarding Calls: The Short Version

A voice AI onboarding agent calls only the accounts that have stalled: users who haven't finished key activation steps within a set window after signup. It checks the account's real state, asks what is blocking the outstanding step, and walks the user through it live. In the first month it reached 340 stalled accounts and finished setup on the call for 189 of them.

The Problem: Users Were Reading the Emails and Still Not Activating

A SaaS platform we worked with had done the obvious things right. A five-step onboarding email sequence, a progress bar in the product, a Slack alert to the customer success team whenever a new account went quiet. The emails performed well by every normal metric: high open rates, decent click-through on the setup links. And a meaningful share of new signups still weren't finishing setup. They'd complete step one, sometimes step two, and then stall somewhere in the middle, not cancelling, not asking for help, just not coming back.

The team's working theory going in was that the emails weren't clear enough. So they rewrote them. Completion barely moved. The real problem wasn't clarity. It was that a written instruction can't tell whether the reader actually understood it, and a user who hits a confusing step in a SaaS product rarely emails support to say so. They just quietly stop.

The Approach: Call the Stalled Accounts, Not Every Account

We didn't propose calling every new signup. That's noise, and it would have trained users to ignore the outreach within a week. Instead, the trigger is specific: an account that hasn't completed a defined set of activation steps within a set window after signup gets flagged, and a voice AI agent calls that account directly. Users who activate on their own during that window never hear from the agent at all.

This matters for the same reason we've written about elsewhere: speed and relevance are what make outbound calls land instead of getting hung up on. A call that references exactly where a user is stuck reads as help. A generic "just checking in, how's it going" reads as a sales call in disguise, and gets treated that way.

The Build: A Call That Can Actually Finish the Setup

The agent opens by checking the account against the product's actual state (which steps are done, which aren't) before it says a word about what's missing. It doesn't ask "have you finished setup?" as an open question; it says specifically what's outstanding, based on what the system already shows, and asks what's holding that step up. That distinction is the difference between an agent that sounds like it read a script and one that sounds like it's actually looked at the account.

From there, the call branches on what the user says. A misunderstood setting or a step the user simply skipped gets walked through live, the same way a support rep would talk someone through it over the phone: the agent describes exactly what to click and confirms the change registers on the account before moving on. A technical blocker (an integration that needs an API key from another system, a permissions issue on the user's end) gets handed to a human immediately, with the specific blocker and everything already confirmed on the call passed along, so the user doesn't repeat themselves. This is the same escalation discipline behind how we design a voice AI agent's handoff to a human generally: the agent finishes what it can finish cleanly, and stops the moment a question needs judgment it doesn't have.

Before any of this went live, the agent went through the same grounding process we use on every build, pulling the actual product's setup flow, current UI copy, and the specific wording support already uses to explain each step, the same approach described in how we train a voice AI agent on product knowledge. A generic walkthrough script would have broken the moment the product's UI didn't match what the agent described.

What Changed

In the first month live, the agent reached 340 of the accounts flagged as stalled and completed setup live, on the call, for 189 of them. Those 189 accounts activated the same day the call happened, instead of the two-to-three-week drift that stalled accounts typically went through before either finishing setup on their own or quietly churning without ever using the product for real.

The team's customer success lead flagged one thing they hadn't expected going in: the calls surfaced a UI step that was confusing more users than anyone realized, because it showed up as the same stuck point across dozens of calls in the first two weeks. That's not something an email open-rate metric was ever going to reveal: a live conversation about exactly where someone got stuck did, in volume, almost immediately.

The mechanism here isn't unique to onboarding. It's the same logic behind the outbound voice AI agent we built to recover churn-risk accounts and the demo callback agent that calls new signups within 90 seconds: find the specific moment a customer is stuck or slipping, and put a call in front of it before the gap widens into a lost account.

If new signups are stalling out somewhere between "signed up" and "actually using it," get in touch. We'll show you what a voice AI agent that can see exactly where they're stuck, and finish the setup with them, looks like for your product.

Frequently Asked Questions

Why did this SaaS platform build a voice AI onboarding call instead of relying on an email sequence?

Their onboarding emails had strong open rates but weak completion rates: users read the instructions and still didn't finish the setup steps that predicted long-term retention. A phone call catches the specific step a user is stuck on in real time and walks them through it, which a static email sequence can't do no matter how well it's written.

What does the voice AI onboarding agent actually ask a new user?

It confirms which setup steps are already complete by checking the account in real time, then asks the user directly which part they're stuck on or planning to do next. If the user names a blocker the agent can resolve (a misunderstood setting, a missing integration step) it walks them through it live; anything more technical gets handed to a human with the full context attached.

Does every new signup get a call?

No, the agent calls users who haven't completed key activation steps within a set window after signup, not every new account. Users who activate on their own without help never get a call at all, which keeps the outreach relevant instead of becoming another onboarding touchpoint people learn to ignore.

What happened to activation rates after the agent went live?

In the first month, the agent reached 340 of the accounts flagged as stalled and completed setup live on the call for 189 of them. Those 189 accounts activated the same day instead of drifting for the two-to-three weeks stalled accounts typically took to either finish setup on their own or churn before ever using the product.

Related Service

Voice AI Agents

Handle inbound calls, qualify prospects, and follow up with leads, using natural-sounding AI voice agents that never miss a call.