Most onboarding statistics pages are copies of copies. The same numbers circulate for years, the original study is nowhere to be found, and half the figures mutate a little with each repost. If you've ever tried to trace "73% of customers churn within 90 days of poor onboarding" back to an actual study, you know the feeling.
This page is the version we wanted to find. Every statistic below links to a primary source, with the methodology and the year noted where the source publishes it. Where a famous number has no traceable origin, we left it out.
The most concrete numbers on what bad onboarding costs come from Wyzowl's customer onboarding survey, which polled 216 consumers. It's a small sample and it dates to 2020, so treat these as directional. They've held up as the most-cited primary data in the category because almost nobody else has published survey data at all.
- 8 in 10 users say they've deleted an app because they didn't know how to use it.
- 55% of people say they've returned a product because they didn't fully understand how to use it.
- 86% say they'd be more likely to stay loyal to a business that invests in onboarding content that welcomes and educates them after purchase.
- 65% say video is their favorite way to learn how to use a product or service.
The pattern behind all four: users don't file a complaint when they're confused. They leave.
Intercom has published the SaaS version of the same behavior for years: 40 to 60% of people who sign up for a free trial use the product once and never come back. Their framing of why is worth keeping: people abandon trials for reasons within the company's control, because they never saw value, and rarely because the product genuinely couldn't help them.
The single most quoted number in product adoption comes from Pendo's 2019 Feature Adoption Report, and unlike most quoted numbers, this one has a real methodology behind it: an analysis of feature usage across 615 companies with more than a year of usage data.
- 80% of features in the average software product are rarely or never used.
- Pendo estimated that public cloud software companies had invested $29.5 billion in developing those rarely-used features.
The report is from 2019, and we'd love to see someone rerun it in the AI era. But the direction of the finding is hard to argue with, and it explains why feature adoption became its own discipline: shipping the feature turns out to be the easy half. Getting anyone to find it is the work.
For benchmarks, the most recent large-sample source is Userpilot's Product Metrics Benchmark Report, which analyzed 547 SaaS companies in 2024.
- Product-led companies average a 34.6% activation rate.
- Sales-led companies average 41.6%, which surprises people until you remember a human is walking the customer through setup.
- Among companies with activation rates above 50%, 80% use multimedia elements like video or animation in their onboarding.
If your activation rate is somewhere in the thirties, you're normal. That's also the uncomfortable part: the normal outcome is that roughly two thirds of signups never reach the moment your product becomes useful to them.
Onboarding and support are converging on the same problem: users want to figure things out inside the product, and mostly can't. The support-side numbers make the gap measurable.
- 81% of customers try to resolve their issue themselves before contacting a live representative, per research published in Harvard Business Review by Matthew Dixon and co-authors in 2017.
- Yet a Gartner study found that while 73% of customers use self-service at some point in their journey, only about 1 in 7 issues (14%) are fully resolved through self-service.
- A June 2025 Gartner survey found 60% of customer service agents fail to promote self-service to customers, which tells you how much confidence support teams have in their own self-service tooling.
Read together: nearly everyone tries to self-serve and almost nobody succeeds. The people running support channels know it. That gap is the entire reason in-app assistance became a category.
The forward-looking number that matters most comes from Gartner's March 2025 prediction: by 2029, agentic AI will autonomously resolve 80% of common customer service issues without human intervention, reducing operational costs by around 30%.
The word doing the work in that sentence is "agentic." Gartner's distinction is between AI that generates text about your problem and AI that takes the actions to fix it, like navigating the interface on the user's behalf. Given the 14% self-service resolution rate above, the prediction says more about how low the bar is than about how good the AI needs to be. We wrote up what that action-taking architecture looks like in practice in our guide to workflow-aware assistants.
Production data on in-product AI is still scarce, but some exists. Frigade's State of In-Product AI report, based on a year of real AI assistant usage across production apps, is the most direct published measurement so far. Frigade is our product, so apply the same skepticism we apply to every vendor number on this page. What it found:
- About 35% of in-app assistant responses were guided walkthroughs rather than text answers.
- Roughly 70% of users who received a walkthrough followed it, advancing at least one step in the same session.
- 62% of returning users asked three or more questions, meaning users who come back treat the assistant as a way to get work done, not a novelty.
Set that 70% follow-through next to the 14% self-service resolution rate above and you can see why Gartner is betting the category moves from answering to acting.
Pendo's 2019 Feature Adoption Report, based on usage data from 615 companies, found that 80% of features in the average software product are rarely or never used. It remains the only large-scale published study of feature-level usage.
Userpilot's benchmark of 547 SaaS companies puts average activation at 34.6% for product-led companies and 41.6% for sales-led companies. Products above 50% are outliers, and 80% of them use multimedia in their onboarding.
The best available survey data (Wyzowl, 216 consumers) found 8 in 10 users have deleted an app because they didn't know how to use it, and Intercom's long-running estimate is that 40 to 60% of free-trial signups use a product once and never return.
Mostly not yet. Gartner found 73% of customers use self-service at some point, but only about 14% of issues get fully resolved there. That resolution gap is why Gartner predicts agentic AI, which takes actions rather than just answering, will handle 80% of common support issues by 2029.
The production data published so far says yes. Frigade's State of In-Product AI report, drawn from a year of AI assistant usage in production apps, found about 70% of users who received a guided walkthrough followed at least one step in-session, and 62% of returning users asked three or more questions. It's vendor data, but it's the only public dataset measuring this behavior at scale.
