Onboarding benchmarks from real data across 464 SaaS products: median tour completion is 29%

The tooltip stat matches what we saw — we shipped hover tooltips on a couple of features, checked the analytics a month later, and basically nobody had ever opened one. Ripped them out. On the NPS number: is the 4.3% response rate per survey impression or per unique user? We re-show the survey to non-responders, which inflates impressions fast, so the denominator matters a lot for sample size planning.

The announcement decay data may be the most immediately actionable finding. If 43% of lifetime impressions arrive in 48 hours, teams should treat announcements as timed campaigns: segment the audience, coordinate in-app and email, and schedule a second exposure for eligible users who missed the first. I would love to see the same curve split by release type, because a major workflow change should age differently from a minor update.

29% is lower than I would have guessed, but it also makes me wonder whether completing the tour is even the right success metric. Did you see any correlation between shorter tours and higher completion?

The methodology line deserves more attention than the headline stats: per-company medians instead of pooled averages is exactly what most public benchmarks get wrong. One whale and an "average" means nothing.

Two selection effects worth printing next to the numbers, though. Your sample is companies that installed an adoption tool, teams that already care about onboarding more than the median SaaS does, so the true population number likely sits below 29%, and your "you're median, not failing" reframe is even stronger than stated. And the 23% vs 69% auto-start gap probably isn't the trigger's doing: user-initiated tours get taken by people already engaged enough to open one. Ship "switch to user-initiated" expecting 69% and the disappointment will be data-driven. The causal read needs the same tour randomised across triggers, which, with 464 companies on the platform, you might genuinely be able to run.

The completion numbers become decision-useful only when paired with activation. I would cohort by start mode and tour length, then compare 7-day activation and 30-day retention among users who were eligible, who started, who completed, and a matched no-tour group. Otherwise the 69% user-initiated figure may mostly measure pre-existing intent. The product decision changes only if a tour lifts downstream behavior, not merely if users finish it.

The 23% vs 69% gap is a great reminder that completion is not activation. I would tie the tour to the first meaningful outcome. In Speechara.Ai, that would be the first successful transcript or translation in a real meeting, not simply finishing an onboarding flow. Did you have enough downstream data to compare tour completion with first value or week 4 retention?

The 464 figure in your post caught my attention — real numbers like that are rare in build-in-public threads. We've seen similar dynamics with AI in our own workflows, and the gap between demo and production use is usually where the surprises hide. Out of curiosity, how did you decide what to measure first?Reading this made me think about how AI decisions tend to look obvious in hindsight but are genuinely messy at the time. One thing we've learned shipping AI-adjacent tools: the bottleneck is almost never the model, it's the surrounding workflow. What part of your stack ended up being the real constraint?

The user-initiated gap is useful, but I'd tie completion to the first real result. For DictaFlow, that means the first successful text insertion into another app, not finishing a tour inside DictaFlow. Someone can finish a short tour and still never try DictaFlow in their actual workflow. Tracking how long it takes to get that first successful insertion in another app would probably tell you more than tour completion alone.

this is a great public service, the "youre median, not failing" reframe alone will save a lot of founders from torching a perfectly fine onboarding. one thing id add on top of your length finding: tour completion is a proxy, and a slightly dangerous one, because completing a tour isnt the same as activating. its very possible to have a "completed" 8-step tour that walked people through features they didnt need and left them no closer to their first real win. so the number id chase isnt higher completion, its a shorter path to first value, which sometimes means DELETING tour steps until the only thing left is the click that produces the aha. a 2-step tour at 60% that lands someone on their first real outcome beats a 6-step tour at 40% that ends on a settings page. did your data let you correlate tour completion with actual downstream retention or activation, or just completion in isolation? because if you can show "completion alone doesnt predict retention but reaching step X does", thatd be the most valuable slide in the whole report.

The methodology choice matters as much as the numbers here: company-level medians keep one large account from pretending to be a universal benchmark. I’d extend that by splitting the funnel into exposure, intentional start, completion, and the first product outcome the tour is meant to create. The 23% versus 69% auto-start gap suggests intent is doing a lot of work, but the next question is whether user-initiated tours are simply selected by already-motivated users. A useful follow-up would be a matched comparison by account age, traffic source, and prior activity, then week-4 retention or repeat usage after each tour type. That would turn “shorter and user-initiated wins” from a strong correlation into a more reliable onboarding decision. Are those downstream retention events available in the same dataset?

The auto-start vs user-initiated gap (23% vs 69%) matches what we saw at SocialPost.ai: intent beats interruption, so we moved the tour behind a "show me around" button and activation improved even though fewer people ever saw the tour. One question on the data: did you cut tour completion against week-4 retention? Completion is the easy number to inflate, and I would love to know whether a 73% completion on a 2-step tour actually predicts anyone staying.

The auto-start vs user-initiated completion gap is really interesting. It shows that getting users into the flow isn't the same as getting them to actually engage with it. I'd be curious to see how this changes when onboarding is segmented by traffic source or user intent.

This is genuinely useful—especially the choice to report medians and quartiles by company rather than pooled averages. That makes the numbers far more actionable for the typical small SaaS, where a few huge accounts could otherwise distort the benchmark.

The auto-start vs user-initiated finding is the one that should change how most products ship tours by default. A 46-point completion gap isn't a UX detail, it's telling you that autonomy matters more than exposure. People don't abandon because the tour is bad. They abandon because they didn't choose to be in it.

The 9+ steps at 8% completion is brutal but expected. The more interesting question is whether there's a cliff between 4-step and 5-step tours, or if it's a gradual decay. If there's a step-change somewhere in that range, that's where to draw the line.

The 43% of lifetime impressions in 48 hours for announcements changes the sequencing logic entirely. If you're sending emails tied to announcements, the email needs to go out in hour 1, not day 3.

The more interesting benchmark is the gap between tour completion and actual activation. A 73% completion rate for a two-step tour sounds great, but it doesn't necessarily mean the user reached the outcome the tour was meant to create.