SaaS companies have spent a decade building knowledge bases, help centers, tooltip tours, and chatbots, all optimized around one metric: deflection rate. How many tickets didn’t get filed. How many users didn’t reach a human. The whole model is measured by absence.
Nimbia is an AI onboarding agent for SaaS that does 1-on-1 screen-sharing calls with users, seeing their screen, controlling their browser, and guiding them via real-time voice. We work in a category that treats deflection as victory. We think that’s the wrong measurement.
The Accepted Wisdom: Deflect Tickets Before They Start
The conventional playbook goes like this. You launch. Support requests come in. You write docs. You build a help center. You add a chatbot that suggests articles when someone opens a ticket. You instrument your help center and measure “deflection rate,” which is the percentage of users who visited self-service and didn’t file a ticket afterward.
The metric looks good. Tickets are down. The team celebrates.
The model was never wrong as a hypothesis. Self-service genuinely does resolve some questions. Sixty-seven percent of customers prefer solving issues themselves when the solution is clearly available. When someone needs to know where to find their invoice, a help article is fine.
But the hypothesis became doctrine. Deflection got optimized aggressively, with chatbots trained to suggest articles before letting anyone reach a person, help centers structured around company-internal product taxonomy rather than user mental models, and onboarding flows designed to reduce inbound questions rather than to actually teach the product. The goal shifted from “users succeed” to “tickets don’t arrive.”
Why It’s Wrong: Deflection Is Not Resolution
Deflection measures the absence of a ticket, not the presence of a resolved problem.
According to CMP Research, 30% of customers start support journeys in self-service, but only 25% of cases get resolved there. The gap between those two numbers, support that started in self-service but didn’t finish there, is the failure rate the deflection metric hides. A user who spent 10 minutes in your help center, found nothing useful, closed the tab, and churned: that counted as a deflection. A win, by your metric.
The data on self-service abandonment makes the picture clearer:
- 43% of customers report being unable to find relevant information in self-service portals.
- After 10 minutes of searching, 52% of customers give up and contact the company by another channel.
- 46% of customers abandon self-service due to inaccurate or unhelpful information.
So roughly half of your deflected tickets are not resolved tickets. They are frustrated users who stopped trying. Some escalated to another channel. Some churned. Your deflection rate counted all of them as success.
The Evidence: What Deflection Actually Costs
Poor support doesn’t just create support tickets. It creates churn.
Seventy-three percent of consumers say they will abandon a brand after just one poor customer service experience. One in three customers leaves a brand they love after a single bad experience.
In SaaS specifically, the compounding effect is faster. Sixty-eight percent of users cite poor onboarding as the primary reason they left a product. Up to 75% of SaaS users churn within the first week when the onboarding experience is poor. Users who don’t engage with the product within the first three days have a 90% chance of churning.
These aren’t onboarding statistics. They are support statistics in disguise.
The real cost isn’t the support ticket. It’s the user who left without filing one.
Why “Deflect to Docs” Exists at All
To be fair to the model: it exists for a reason. Human help is expensive and doesn’t scale.
A customer success manager handling 200 accounts can’t do 1-on-1 screen-sharing sessions with every new user who gets stuck. A support team of 10 can’t hop on a call with every trial user who can’t configure their first integration. The math doesn’t work. So the industry built the next best thing: written documentation, tooltip tours, chatbots that route users to articles. Deflect tickets before they start, because the alternative (actually helping everyone) was impossible.
That constraint was real. The constraint no longer exists.
AI can now see a screen, understand what’s on it, control a browser, and hold a natural voice conversation at the same time. The thing that made 1-on-1 guidance impossible, the cost and scarcity of human attention, has been removed. An AI agent can join a user in a live session, see exactly where they’re stuck, and walk them through the fix in real time. Not with a tooltip pointing at a button. With a voice, in their browser, doing the work alongside them.
See also: Customer Success Without the Team
What Replaces It: Help That Resolves
The alternative to “deflect more” is not “hire more humans.” It is measuring resolution instead of deflection.
Resolution means the user reached their goal. They set up the integration, completed the workflow, understood the feature. They did the thing they came to do.
The infrastructure for resolution-first support looks different from the infrastructure for deflection-first support.
Deflection-first builds: a help center organized around product features, a chatbot that routes to articles, tooltips that appear once and never repeat, onboarding flows optimized to check off steps rather than confirm outcomes.
Resolution-first builds: live guidance that sees the user’s screen and adapts in real time, voice explanations that match the user’s context, AI that can perform actions in the user’s browser alongside them, measurement based on whether the user completed their goal rather than whether a ticket was filed.
Nimbia is built on the second model. One line of JavaScript. The agent joins users in a live session, 24 hours a day, in any language, and works through the issue with them on-screen. The metric is whether the user succeeded.
For SaaS companies where trial-to-paid conversion depends on users actually reaching value in the first session, the distance between deflection and resolution is the distance between activation and churn.
See also: Best User Onboarding Tools in 2026
Implications: What to Audit in Your Own Stack
If your support strategy is currently built around deflection rate as the headline metric, here are the questions worth asking.
What happened to your deflected users? If your help center deflects 40% of potential tickets, do you know what those users did after they left? Did they succeed? Did they churn? Deflection rate without a downstream cohort analysis is measuring silence, not resolution.
What does your chatbot do when it can’t find an answer? Most support chatbots are trained to suggest articles and then offer to escalate. The escalation rate tells you how often the chatbot’s article suggestion was actually useful. If 60% of chatbot sessions end in escalation, the chatbot is adding a friction step before support, not deflecting it.
Where do users churn in the first two weeks? Map your churn timing against your support touchpoints. Users who hit their first wall without help tend to churn faster. If your churn cluster is in day three to seven, and your self-service content covers that same feature area, the correlation is worth investigating.
What is your help center’s resolution rate, not deflection rate? This requires post-session surveys or behavioral tracking (did the user complete the task after reading the article?). It is harder to measure. It is the right thing to measure.
The Model Is Built on a Compromise That No Longer Has to Exist
The economic argument for deflection-first support was never that documentation was better than live help. It was that live help was too expensive to give everyone. That argument is obsolete.
SaaS companies that keep optimizing for deflection in 2026 are optimizing for a problem that no longer exists, at the cost of the outcome that matters: users who succeed, convert, and stay.
Book a call to see how Nimbia resolves users’ problems on-screen, in real time.
Part of Nimbia’s ongoing analysis of the SaaS onboarding and support space. March 2026.