The account finished onboarding on schedule. Kickoff on day three, integration live in week two, training delivered, go-live signed off a day early. Onboarding completion rate: 100 percent. Every task green, no escalations, no red flags in a single weekly review.
Six months later they didn't renew. The person who told us so hadn't been on a call since March.
Nothing in that story reads as failure until the last sentence, which is exactly what makes it hard to catch. A stalled onboarding announces itself. A finished one gets celebrated.
Call it checkbox churn: the customer completes your plan and leaves anyway.
Why this hurts worse than a visible failure
A visible stall has an advantage nobody talks about. It creates attention. Someone escalates, someone reschedules, someone gets on a call and asks what's actually going on. The account stays in front of a human because it's making noise.
A completed onboarding does the opposite. It gets a congratulations in Slack, a closed-out project, and a lifecycle stage that flips from onboarding to active. That flip is the moment the account stops having an owner whose job is to watch it. It graduates into the pool of accounts that are presumed fine until they tell you otherwise, and they tell you otherwise at renewal.
So completion is not neutral. It actively removes the account from your attention, on the strength of a metric that never asked the customer anything.
Completion measures you, not them
Look at any onboarding plan and read who performs each task. Schedule kickoff. Configure the integration. Run the training session. Send the credentials. Confirm the data migration. Almost every line is work your side does, or work you ask the customer to do so that your side can proceed.
Which means completion rate is an honest measure of one thing: your team ran the project it said it would run. That is worth knowing. It is not the same question as whether the customer built a habit, told a colleague, or planned next year's budget around you.
A project plan tracks delivery. A relationship tracks intent. They share a timeline and answer different questions, and only one of them shows up at renewal.
A customer who's leaving will still check your boxes
Here's the mechanic that makes this so quiet. Telling a vendor you're done is socially awkward. Sitting through the training is not. For a customer who has already privately decided, riding the checklist to the end is the path of least friction: no confrontation, no negotiation, no conversation where somebody has to say the disappointing thing out loud.
You've seen the shape of it. The champion who joins the go-live call, says the right things, signs off, and is never on a call again. Nothing broke. They just stopped being there.
It is the politest way to leave, and on a task board it is indistinguishable from success.
Four tells that actually predict renewal
None of these require a product. All four are readable in what you already have, if someone reads.
Attendance
Who actually showed up to the last two calls, compared to who was in the room at kickoff. Champion attendance decay is the earliest cheap signal you get, and it usually precedes silence by weeks.
In the record it looks like a shrinking accept list, a delegate sent in the champion's place, or a reschedule that never gets rebooked. None of those trip an alert.
This week: for every account in its first year, write down who attended the last two calls next to who attended kickoff. Any account where the person who signed the deal has quietly dropped off the invite goes on a short list you review by name.
Question volume
Engaged customers are annoying. They push back on your defaults, they complain about the export format, they ask detailed questions with an edge in them. That friction is a customer investing in getting this right.
Silence is not satisfaction. In the record it looks like a thread that used to run four replies a week and now holds your last two unanswered messages, or replies that have compressed down to "sounds good" and a thumbs up.
This week: sort your accounts by days since the customer last wrote you something that wasn't a one word confirmation. The bottom of that list is your risk list, and it will not match your escalation list.
Promise to delivery
Not whether the tasks completed. Whether the specific thing that closed the deal actually got delivered. The migration timeline, the integration with the tool they already use, the report format their board expects. That promise is the reason they bought, and it frequently lives in a deal note or an email thread that nobody opens after the handoff.
This is the seam between sales and onboarding, and it's where a plan can run perfectly against the wrong requirements. We wrote about that failure at length in churn doesn't start at renewal.
This week: for each account in its first 90 days, find the one commitment made during the sales cycle and confirm out loud that it shipped. If you can't find the commitment, that is the finding.
Breadth of logins
One user is a pilot, not adoption. A single active person means a single point of failure, and that person will eventually change roles, go on leave, or leave the company, taking your renewal with them.
In the record it looks like seats provisioned against seats used. If you don't have usage data, count the distinct human beings who have ever emailed you from that account. One name is a warning regardless of how enthusiastic that name is.
This week: list every account where you've only ever heard from one person, and make an introduction to a second one the goal of your next call.
Why small teams can't just watch for this
Notice what all four have in common: none of them live in a task list. They live in calendars, inboxes, deal notes, and the gaps between messages. Watching them means rereading the record across every account, every week, and forming a judgment about each one.
One person carrying 60 accounts can't do that, so they triage by volume, which is the correct move given the constraint and also the reason a quiet account never makes the list. The bottleneck isn't diligence, it's reading time. That's the structural problem, and it's the one we've written about before, so we won't re-argue it here.
What we're building instead
Herofy watches the relationship alongside the plan. Connect HubSpot and Gmail, and Sidekick reads each account's threads and notes for the things a task board can't hold: who's still engaged, what the tone is doing, which commitments are still open, who's gone dark.
That becomes a Relationship Health Score, built from relationship signals rather than a usage number a customer success platform infers from logins. Every signal shows its work: the actual sentence from the actual thread that made it think so, not a score you're asked to trust. Then it drafts the nudge that gets things moving, and you edit it and send it, or you skip it.
Two hard limits worth stating plainly. Sidekick never contacts your customers. A human sends every message, even in the most autonomous mode. And the rules it follows aren't buried in a model somewhere, they live in a plain English handbook you can read and edit.
The line worth keeping
The boxes measure you. The renewal measures them.
A 100 percent onboarding plan tells you your team did what it promised, which is genuinely worth knowing and is not evidence that anyone is staying. The evidence for that is in whether they still show up, still argue with you, got the thing they were promised, and brought someone else along.
Common questions
What is a good onboarding completion rate?
Most teams that measure it land somewhere north of 90 percent, and hitting that is a reasonable bar for operational discipline. It's also an easy number to hit, because you control nearly every task in it. Treat it as a measure of your delivery, not as a predictor of renewal, because it does not correlate with one.
Why do customers churn after a successful onboarding?
Because completion measures your team's execution, not the customer's commitment. A customer who has already decided to leave will still attend the sessions and sign the go-live, since compliance is easier than confrontation. And a finished onboarding moves the account out of active watching, so nobody is looking when the real signals show up.
What should you measure in customer onboarding instead?
Four things that track intent rather than delivery: whether the champion still attends calls, whether the customer still asks questions and pushes back, whether the specific promise made during the sales cycle was actually delivered, and how many distinct people at the account are using you. All four are visible in your calendar and inbox today.
Running onboarding at a small B2B SaaS team on HubSpot, where every account matters? We're pre-launch, building with a small number of design partners. Come build it with us.