From assumptions to evidence, faster
Prediction Markets started with data rather than a defined workflow: four tabs of information and an open question about how any of it could become genuinely useful to users.
Instead of treating the existing tables as the product, AI became a way to rapidly explore the problem space, form plausible hypotheses, and turn them into something concrete enough to challenge with Product and put in front of real users.
Role: Senior UX/UI Designer
Scope: Product exploration · UX strategy · AI-assisted research & prototyping · User validation
Collaboration: Product Manager · Domain experts · Users · Engineering
Status: Ongoing experiment
Scope: Product exploration · UX strategy · AI-assisted research & prototyping · User validation
Collaboration: Product Manager · Domain experts · Users · Engineering
Status: Ongoing experiment
Starting with data
The information was there. The workflow wasn't.
The first step was understanding what someone might actually want to accomplish with prediction-market data. Who could plausibly use it? What questions might they have? What makes a prediction relevant? What could trigger an investigation? And where could that investigation eventually lead?
Claude helped accelerate that exploratory work by generating plausible personas, questions, use cases, and investigation paths around the data. These weren't research findings—they were hypotheses. They provided a richer starting point than simply putting four tables in front of users and asking what they wanted.
Those assumptions were reviewed with the Product Manager, using domain and product knowledge to eliminate implausible scenarios and sharpen the ones worth exploring.
A simple model started to emerge:
Signal → Relevance → Investigation → Action
Working with Claude in code made it possible to turn that model into functional interactions quickly, explore different approaches, and start discussing something tangible rather than an abstract workflow.
A portfolio-first hypothesis
One early assumption became particularly influential: relevance would start with the user's existing exposure.
Connecting prediction events to a portfolio provided a natural starting point. Instead of navigating an enormous universe of events, users could begin with signals potentially connected to what they already owned: What is happening that could affect my portfolio?
The first concept therefore leaned toward risk monitoring—surface a relevant event, understand its potential impact, investigate, and decide whether action was needed.
It was a credible hypothesis and gave the experience a clear direction.
User feedback would reveal that it was only half of the story.
A second path emerged
The All Events view was initially treated as less important than portfolio-specific signals. Putting the concept in front of a user challenged that assumption.
The broader event universe was valuable as a starting point in its own right. Users weren't only interested in “What could affect my portfolio?” They also wanted to know “What's happening in the world?”
That distinction opened another path through the product. An interesting event could trigger investigation even without a connection to an existing position—and potentially reveal an opportunity the user wasn't already looking for.
The experience was beginning to support two different intents:
Managing risk: start with existing exposure, identify relevant signals, investigate their potential impact, and decide whether action is required.
Discovering opportunity: start with what's happening, notice something interesting, investigate its implications, and determine whether there is an opportunity to act.
What had looked like a secondary data view was potentially a primary entry point into the experience.
Making it cheaper to be wrong
The first workflow wasn't completely right. That was part of the point.
AI dramatically reduced the effort required to go from incomplete information to a coherent hypothesis. Plausible users and questions could be explored quickly. Assumptions could become interactions. Alternative workflows could be built instead of debated. Product could react to something concrete while it was still inexpensive to change.
And then actual users could prove some of it wrong.
The value of the synthetic exploration wasn't that AI accurately predicted user behavior. It didn't. Its value was getting to something credible enough to test much sooner.
Product knowledge progressively grounded the initial assumptions. User feedback then started replacing them with evidence—including the discovery that the broader event universe mattered much more than initially expected.
AI didn't remove research from the process. It shortened the distance between assumption and evidence.
Designing the investigation
Finding a relevant signal is only the beginning.
The next iterations are increasingly focused on what happens between noticing something and deciding to act: what catches attention, what makes an event worth investigating, what context users need, how portfolio exposure changes its meaning, and what gives someone enough confidence to continue—or stop.
Those questions are now shaping the workflow and information hierarchy based on actual feedback rather than the assumptions that started the project.
The goal is no longer simply to organize prediction-market data.
It's to help users move from noise to something worth acting on.
Designing in the real product
Working with real design systems, components and application code creates opportunities to bring prototyping closer to the product itself, testing ideas against actual interactions and technical constraints.
Whether the resulting code ultimately ships isn't the experiment. Engineering can review, reuse or replace the implementation. The value lies in reducing the translation between design and development, exposing problems earlier and getting to viable solutions faster.
The goal is to keep tightening the iteration loop—moving from prototype → handoff → implementation → correction toward a continuous process of design → test → refine, directly in context.