UX Design Mistakes That Don't Show Up in Figma: Six Data Fingerprints

This article was originally written for external publication, yet ended up here instead. Enjoy!

Design review went well. Everyone liked the screens. The prototype tested clean. The team shipped.

Eleven weeks later, activation was down 19%. Support volume had climbed. A retro produced fourteen action items and no root causes.

Some UX design mistakes only introduce themselves through data.

The UX design mistakes that cost the most survive every review because every review is run by people who already understand the product. They know what the button does. They know where the feature lives. They've been inside the logic long enough that none of it feels ambiguous. So the designs pass. The users arrive. Something breaks quietly.

The break doesn't show in Figma. It shows in your activation rate eleven weeks later. Every one of the patterns below has a specific signature in your UX design analytics. The data already caught them. What follows is how to read what it's been trying to tell you.

The product's internal logic is coherent. The user's understanding of what they're about to do is completely different. Nobody on the team caught it because everyone on the team already knows both languages.

The data fingerprint: Not "how do I do this" tickets – "why does this work this way" tickets. Users describing your product to colleagues using your competitor's terminology. High call volume on a feature that's accessible, visible, and working correctly.

One B2B analytics product had 23 support tickets in 90 days about a core feature. The feature worked. The documentation was accurate. The product called the feature one thing; users understood it as something else entirely. A naming decision from sprint one, still running as a tax on the support team eighteen months later.

Mental model debt compounds. A product that's 7% misaligned in month three tends to be significantly more misaligned by month eighteen, because every new feature gets designed to fit the existing product logic rather than the user's existing understanding. By the time it shows up in NPS, it's been in the data for a year.

Map what your product assumes users know when they arrive. Map what new users actually bring in. The gap between those two is the debt, and it's almost always a naming problem wearing a design problem's clothes. A targeted onboarding correction closes most of it. You don't need a redesign. You need to rename three things and explain one.

There are two meaningfully different user types. The product averaged them into a composite persona that doesn't exist. The interface feels coherent from the inside – it was designed for this fictional median user, consistently – but real usage data splits along a fault line.

The data fingerprint: Bimodal NPS that reads as mediocre until you filter by role – some users at nines and tens, another segment at fours and fives, the overall average landing in the middle and looking like a product nobody loves or hates. Support tickets that map cleanly to one user type when you look at who's submitting them. Session recordings where one cohort navigates efficiently while another restarts the same flow three times before leaving.

This is the standard condition in SaaS product design where the buyer and the daily user are different people – which in B2B is almost always. The admin who configured the account and the team member who uses it every day have different contexts, different goals, and a different tolerance for complexity. B2B SaaS onboarding designed around the buyer loses the end user at activation. Every time.

Segment behavioral data by role. If usage patterns split cleanly, the product has two products inside it that nobody built intentionally. The path out is two onboarding tracks – one for the buyer context, one for the daily user – and targeted adjustments that stop the end user from navigating around decisions made for someone else.

Progressive disclosure is a sound principle. Hiding things users rarely need keeps the interface clean. The implementation problem is that "rarely needed" is usually a guess made in week two, by designers who didn't have usage data yet. The principle is sound. The calibration is a bet made in a vacuum.

The data fingerprint: High documentation usage for features that are findable but buried – users motivated enough to go looking, just not able to find things without help. In-app search activity for functions that exist in the product but don't appear in primary navigation. Onboarding completion rates that look solid while adoption of supposedly core functionality never moves.

During the design sprint, features get classified as "edge cases" or "advanced" before there's any evidence to support that classification. A feature used by 73% of active users three times a week is not advanced. Calling it advanced was a design decision made before the product had any users. Nobody went back.

Pull feature usage data and map it against your information architecture. Any feature with high usage rates more than two levels deep is a miscalibration. Move it. The pushback will be that it disrupts the design system hierarchy. The design system hierarchy was built without usage data and can be updated.

The absence of feedback is a basic UX problem. It shows up immediately in testing and is fixed before launch. Latency is different. The feedback exists. The timing is off. Validation that fires too early shuts down exploration. Confirmation that arrives too late creates enough uncertainty that users abandon before they see it.

The data fingerprint: Abandonment concentrated in the middle of multi-step flows, not at the beginning or the end. Users reaching the final step and not submitting. Repeat visits to the same screen before forward movement – checking something, leaving, coming back, checking again.

In form-heavy onboarding, this manifests as inline validation firing on keystroke. The user is still typing and the error state is already active. The first input attempt becomes a failure state before they've finished a thought. In setup flows, it appears as progress indicators that move without telling users what was just validated – the bar advances, but they can't tell whether they've already failed something two steps back.

Teams diagnose this as a copy problem and rewrite the error messages. The abandonment rate doesn't move. Add time-in-step to your funnel analysis. A step that should take under 90 seconds where users are averaging four minutes isn't a complex step. It's an anxious one. Fix the timing.

The UX correctly identifies the primary user goal. Then the visual design buries it under the primary business goal. Both were approved in the same review. Nobody noticed they were in conflict. Everyone signed off on both.

The data fingerprint: Diffuse click maps on pages where you expect convergence. Time-on-page increasing without any improvement in completion rate. A/B tests where copy and headline changes move engagement numbers but not conversion – because the problem is visual weight, not words.

This accumulates at the growth stage without anyone intending it. The core team builds the primary flows. The growth team adds upgrade prompts, trial banners, feature announcements. Each addition is justified individually. Collectively, they've buried the primary user action under secondary product priorities. The user engagement silence that follows looks like disinterest. It's usually confusion about what the product actually wants users to do.

Show the page to someone who has never seen the product. Ask them what the primary action is. If they hesitate, or point at the upgrade banner, the hierarchy is inverted. Fixing it requires the growth team to agree to deprioritize their elements. That conversation takes longer than fixing the hierarchy itself. Have it anyway.

The product was designed for new users during a period when almost all users were new. The interface did its job. The users graduated. The interface didn't.

The data fingerprint: Churn concentrated at the 4 – 6 month mark among the most active accounts – the opposite of where churn normally appears. Support tickets from high-usage accounts that aren't about confusion; they're about friction. Too many clicks. No saved workflows. No keyboard shortcuts. These users aren't lost. They've outgrown the interface.

The reason this is one of the harder UX design mistakes to catch internally: the product team is usually among the highest-usage people on the product, and the interface feels obvious to them. The experience gap between someone who's been in the tool daily for a year and a user arriving at activation has compressed to nothing in their perception.

Track behavioral patterns by cohort age. The user at month seven should be navigating meaningfully differently than the user at week six – fewer exploratory clicks, more direct paths, faster session completion. If the patterns are similar, the product hasn't graduated anyone. The fastest confirming signal is support tickets from your longest-tenured, highest-usage accounts about things that work correctly. That's not a bug report. That's a user who's run out of product. Start with keyboard shortcuts and saved-state features. That's usually where the frustration concentrates first. Improving empty state design for returning users is part of this – the empty state a new user sees and the one a seven-month user sees after completing a task are different problems with different solutions.

The common thread isn't that these are hard to fix. Most of them aren't. The common thread is that they don't surface through the processes built to find them. Design review doesn't find them. Usability tests don't find them because participants already know what they're testing. Retros identify symptoms, not sources.

The signal structure is consistent across all six: things appear in the data before they appear in complaints. Mental model debt is in support ticket patterns months before it reaches NPS. Hierarchy inversion is in click map diffusion before it drives churn. The power user problem is in session cohort data long before anyone leaves and names the interface as the reason.

A monthly review against three signals gets most of this: where users stop, where they slow down, where they go for help. Stops are structural breaks. Slowdowns are latency or cognitive overload. Help-seeking is a calibration failure in disclosure or onboarding. Map those three and you'll find at least two of these six inside the first quarter.

These six UX design mistakes have been in the data the whole time. The retro eventually produces fourteen action items. None of them are these.

The method that catches these before they ship: the Fake Persona Test.

If you want a team that catches these before they ship, here is how I approach UX design — and how that informs my product design work.