The Users Who Never File a Ticket
Support volume is a lagging, self-selected signal. How to find the confusion that never reaches your inbox, and why it costs more than the tickets that do.
Filing a support ticket takes a particular kind of person. You have to believe the product is worth the effort, believe someone will read it, be willing to describe your own confusion in writing to a stranger, and have the time to wait for a reply. Every one of those is a filter. By the time a message lands in your inbox it has passed through all four, which means your support queue is not a sample of your users' problems. It is a sample of your most patient users' problems.
Everyone else closed the tab.
The queue is the wrong instrument
Support volume gets treated as the health metric for a help centre. Tickets go down, docs are working. It is a reasonable-sounding measure and it is close to useless on its own, because it moves for two completely opposite reasons.
Tickets fall when documentation improves. Tickets also fall when people stop bothering. A queue that halves because your onboarding article finally explains the API key step and a queue that halves because half your trials now give up on day two look identical on the graph. If that is the only number on the dashboard, the second scenario reads as a win right up until the revenue catches up with it, a quarter later.
The users who leave without contacting you are the expensive ones, and they are systematically invisible. They cost a trial, a slot in your funnel, whatever you paid to acquire them, and - because they never told you why - they cost you the information that would stop it happening again.
What confusion looks like when nobody reports it
It does leave traces. They are just not in the inbox.
Search dead-ends. A query typed into your help centre that returns nothing, or returns results nobody clicks. That is a person who knew exactly what they wanted, went looking for it, and found nothing. It is the highest-intent signal you have and most teams never look at it.
Repeat visits to the same article. Someone reading the same page three times in a session is not learning. They are failing to find the bit they need, and the article has a structural problem rather than a coverage problem.
Questions the answer bot could not answer. If your help centre has AI answering questions, the unanswered ones are a ranked list of missing documentation, written in your users' own words. We surface those specifically because they are the cheapest research anyone will ever run.
The step where sessions stop. Not the page they left from - the last thing they successfully did. People rarely abandon on the screen that confused them; they abandon one screen later, after deciding it is not worth it.
None of those require asking anyone anything. They are all sitting in data you already have, and unlike a survey they are not filtered through willingness to respond.
The one number worth building
If I could give a team a single instrument for this, it would not be ticket volume. It would be a ratio: questions asked of your documentation that it could not answer, over total questions asked.
That number does not fall when people give up. It rises. A user who stops trying stops asking, which removes them from both halves of the fraction. So unlike ticket volume, it cannot be improved by making your product so confusing that people quit - which is the specific failure mode of every deflection metric I have seen.
It also degrades gracefully. You do not need an AI answer layer to approximate it; a search log with a zero-results flag gets you most of the way there.
You cannot prove any of this
The honest problem with everything above is attribution. You cannot prove the person who left would have stayed if article seven had existed. Nobody can. Every claim in this area is inference, and any vendor telling you they have measured prevented churn to two decimal places is selling you a model, not a measurement.
There is also a real risk of over-reading the traces. A search with no results might be someone typing a competitor's feature name to check whether you have it. A repeat article visit might be someone using your docs as a reference the way they are supposed to. Treated as a ranked list of things to investigate, these signals are excellent. Treated as a scoreboard, they will mislead you.
And some silence is fine. Not every trial that ends was winnable, and chasing every abandoned session is how teams end up building for people who were never going to pay.
What to look at first
Pull your help centre's search log for the last month and sort by queries that returned nothing. Read the top twenty out loud.
Those are twenty people who told you precisely what they needed and got nothing back. None of them filed a ticket about it.
About the Author
Simon Moxon has been building SaaS companies since 2009. He's scaled teams to 20+ people, hit £2m ARR, and learned most of the lessons the hard way. Now he's building SupportPages.io - help docs that write themselves when you ship code. He writes about what's changed in SaaS and what hasn't.