AI Storyteller Interface

The first version of Storyteller was deliberately simple: a title, an AI-generated summary, and four insight categories — patterns, momentum, trend, and price — for the Technical Insight product.

Instead of an open chat, it used closed prompts: five pre-generated questions the user could choose from. No free-form input. This was a compliance-driven choice — in finance, an open text field is an open door for AI hallucinations that go unchallenged.

V1 · summary card

V1 · closed prompts

Users were receiving the insights. They just weren't acting on them.

The outputs were too ambiguous, felt unverifiable, or simply didn't surface at the right moment in the workflow. An insight nobody acts on is decoration.

Goals:

1. Deliver actionable insights at the right moment.
2. Ensure compliance and trustworthiness — no unchallenged hallucinations.
3. Streamline content delivery.
4. Increase user engagement.

I prioritised trust and clarity above everything else. In a high-stakes financial context, a user who misreads an AI output doesn't just disengage — they make costly decisions. Every design choice was filtered through one question:

Rejecting the chat paradigm

A conversational interface felt intuitive at first, but it puts the burden on the user to formulate the right question — and we were designing for both finance newcomers and professional analysts. Free-form input in finance creates inconsistency and lets hallucinations slip through unchallenged. I pushed for a guided, pre-structured interface instead.

Defending the transparency layers

Confidence indicators, source attribution, and reasoning trails were challenged as "adding noise." I argued they were the foundation of trust: without them, the AI is a black box — and users opt out of black boxes.

After · the insight

After · guided prompts

After · follow-ups

Simplicity vs. scope

The original vision was a minimal, single-action interface. As compliance requirements and data complexity grew, we layered in more structure. The result is slightly more complex than ideal — but it serves the real use case.

Speed vs. transparency

Loading transparency metadata (sources, confidence) adds latency. We negotiated with engineering on what loads eagerly vs. on demand; some transparency features live behind an expand interaction rather than being always visible.

Personalisation vs. consistency

I wanted prompts tailored to each user's portfolio and behaviour. We deprioritised it in favour of a consistent baseline that could ship and be measured. Personalisation stays on the roadmap.

Storyteller outgrew its first home. It has since been extended to five more products, each tailored to its context — not all keep the closed prompts, but the visual language stayed.

When AI earns trust through transparency, engagement follows — you don't have to ask for it.