Designing Trust into AI-Powered Analytics

How I solved the black-box problem — and why showing uncertainty builds more confidence than hiding it.

The problem no one was talking about

Business analysts are drowning. Not in a lack of data — in too much of it, from too many places, with no way to trust any of it.

When I started research for Klarify, an AI-powered analytics platform, I expected to find people frustrated with slow dashboards or messy UIs. What I actually found was much more fundamental: people didn’t trust AI at all.

After 12 contextual interviews with analysts across manufacturing, logistics, and retail, a brutal pattern emerged:

• 70% of analysts’ time is spent manually gathering data
• 25+ hours per week pulling from multiple disconnected systems
• Zero existing tools offered proactive monitoring

“I can’t walk into my CEO’s office and say ‘the AI said so.’ I need to understand why before I can act on any recommendation.” — Manas Kumar, Operations Analyst, Logistics

Every AI tool on the market — Mixpanel, Amplitude, Tableau — shared the same flaw: they were reactive. You had to remember to check them. And when they surfaced something, there was no explanation, no confidence level, no data trail. Just an answer floating in a void.

What I was asked to design — and the harder problem underneath

The brief was to design Klarify: a platform that alerts business users to anomalies and recommends actions.

But the research revealed something deeper. The core design problem wasn’t about dashboards. It was about trust.

An AI that surfaces the right insight at the wrong confidence level is worse than useless — it erodes credibility over time. My task became twofold: build a system proactive enough to catch issues before they escalate, and transparent enough that analysts would stake their professional reputation on its outputs.

I also had to design for three completely different mental models in the same product — the deep-diving analyst, the operations manager triaging fires, and the VP with 90 seconds who just wants to know if anything is broken.

Three design decisions that changed everything

1. Make uncertainty visible — and celebrate it

The conventional instinct is to hide uncertainty. If the AI isn’t sure, don’t show it — it’ll confuse people. I tested the opposite.

• V1 — Gray confidence text in corner → users missed it entirely ❌
• V2 — Colored badge in header → better, but users still skipped it ⚠️
• V3 — Gradient pill with sparkle icon, anchored to the insight title → “The confidence badges completely changed how I feel about AI recommendations” ✅

The breakthrough: transparency beats perfection. Analysts don’t need AI to always be right. They need AI to tell them when to trust it. A “Medium — 73% confidence” badge turned out to be more reassuring than a confident-looking answer with no provenance.

2. Match the interface to how people actually think

Every analytics tool I audited gave users data and left them to construct the narrative themselves. I designed Klarify’s Insight Detail view around the three questions every analyst asks in sequence:

What happened? → Why did it happen? → What should I do?

This became the three-column layout. In testing, participants described it as “exactly how I think about problems.” Not because it was clever, but because it was obvious once you understood their mental model.

3. Stop building one product for everyone

The most contentious decision was proposing role-based interface adaptation — effectively three UIs instead of one. But the research was unambiguous: an executive and an analyst have opposite needs.

The executive interface: 2 navigation items, 3 metrics, one-sentence summaries, high-confidence alerts only (85%+). The analyst interface: full exploration depth, data source attribution, calculation transparency. Same underlying data, radically different presentations.

What the testing showed

After two rounds of moderated testing with 8 business analysts:

• 92% trust increase when confidence badges were visible
• 3 minutes average time from alert to action (vs 2–4 hours manually)
• 100% task completion rate across all 6 test scenarios
• 9.0/10 average screen quality rating

Participants who had been the most skeptical of AI in initial interviews became the most enthusiastic advocates for the confidence badge system.

The lesson I’ll carry forward

The most impactful work I did on Klarify wasn’t designing a beautiful interface. It was the decision to make the AI’s uncertainty visible and prominent. That single choice accounted for most of the trust increase.

Good design sometimes means designing against your instincts. Hiding uncertainty feels safer. Showing it builds trust.

Want to see the full visual design, screens, and prototype? View the complete Klarify case study on Behance →

Jaya Gershon T — Product Designer *jayagershon@gmail.com*