Prediction Credibility 
Designing trust into AI-powered company intelligence at Crunchbase
UX Case Study will be linked soon!

Project Summary
Crunchbase launched Predictions & Insights, a suite of AI-generated signals designed to help investors, salespeople, and founders evaluate companies more confidently. The feature surfaced growth scores, funding predictions, and acquisition signals derived from proprietary data and machine learning models. The product vision was clear: move beyond raw company data and give users an opinionated view of where a company was headed. But after launch, a gap emerged. Users found the predictions interesting. Very few acted on them.

Research across two rounds of moderated interviews pointed to a consistent pattern. Users were not dismissing predictions outright. They were pausing because they lacked the context to evaluate them. Without visible evidence or reasoning, most users defaulted to the funding and investor data they already trusted, treating the new AI signals as interesting but unverifiable. The design challenge was not to make predictions more prominent. It was to make them believable.

This case study covers four interconnected areas of work that addressed that challenge:
1. Prediction Credibility design -- restructuring the prediction cards to surface contributing factors, timing distributions, confidence levels, and freshness signals so users had enough context to act.
2. Win Predictions -- surfacing historical model accuracy and confirmed outcomes to give users a track record they could evaluate.
3. AI-Powered Company Header and Contextual Summaries -- introducing AI-generated intelligence at the top of the profile, with its own set of trust and transparency decisions.
4. Predictions & Insights Feedback Modal -- give users a direct path to flag inaccurate data and closed the loop between user input and model quality over time.

Each of these projects addressed a different layer of the same underlying problem: when users encounter AI-generated content in a high-stakes research workflow, trust is not assumed. It is earned through evidence, transparency, and the ability to verify. Across all four areas, the design work focused less on showcasing what the model could predict and more on giving users the tools to decide when to believe it.

Taken together, the work produced a reusable framework for AI credibility design, one that has since informed how the team approaches new AI surfaces across the Crunchbase platform.