Less searching
02 · Evolution of the mental model
Three iterations. Each one moved the product closer to the way analysts actually think.
V1 · The chat model
V1 assumed the analyst knew exactly what to ask. The interface was simple. The analytical process wasn't. A conversational AI could answer questions, but the analyst still had to build the context, frame the question, and interpret the answer themselves.
The mental model gap: the product treated each query as independent. The analyst's reasoning is sequential and contextual.
Design decision
An answer engine maps only to the final stage of the analyst's process. Every prior stage (finding the signal, gathering evidence, testing assumptions) remained outside the product.
V2 · The context model
V2 moved the product closer to the analyst's starting point. Instead of waiting for a question, the system now surfaced what mattered: material market changes, supporting context, and agent confidence levels.
V2 made the market easier to read. It did not yet make the analytical journey continuous. Investigation, scenario modelling and decision-making were still separate from the detection layer.
Design decision
Surfacing the right signal is necessary but not sufficient. The limitation of V2 was not information volume. It was that the workflow broke after investigation ended.
What research changed
The CEO and CTO had conducted early conversations with analysts before I joined the exploration. I subsequently led deeper qualitative design validation with the CTO as a primary domain source. Three observations shaped the V3 direction.
01
Signal
Analysts consistently used AI outputs as directional input, but always applied their own judgement before acting on any conclusion. Existing models were useful but not trustworthy enough to bypass investigation.
Design implication
The AI should support the reasoning process rather than make the decision opaque. Removing judgement from the workflow was the wrong goal.
02
Signal
Understanding why a price moved was more actionable than knowing that it moved. Context and drivers were as important as the signal itself.
Design implication
The product needed to surface evidence alongside conclusions, not in a separate tool the analyst had to switch to.
03
Signal
The analytical workflow naturally spanned detection, investigation, scenario modelling and decision, but each phase used different tooling that didn't connect. The cognitive cost of switching between tools was significant.
Design implication
The interaction model needed to match the analyst's natural reasoning flow, not flatten it into a single query or fragment it across separate tools.
| Analyst's reasoning | V1 | V2 | V3 |
|---|---|---|---|
| Event A market change occurs | Chat | Detect | Detect |
| Evidence What does it mean and why? | Chat | Context | Investigate |
| Drivers What is causing the change? | Chat | Analysis | Analyse |
| Assumptions Does the price model still hold? | – | Analysis | Challenge |
| Scenario What is the price impact? | – | Analysis | Recalculate |
| Decision What should I do? | Answer | Answer | Decide |
| Alignment | Low | Partial | High |