AI people trust

Analysts ignored the AI because it was a black box. We made it show its work — and they started using it.

The problem

Routine prep ate a huge share of every analyst's day, and the AI could do it. But it handed over answers with no reasoning, so people re-checked everything. Verifying took as long as the work itself.

What I did

  1. 1

    Mapped trust

    Transparency, predictability, competence, control.

  2. 2

    Found the grind

    Tracked which tasks were routine vs. judgment.

  3. 3

    Added reasoning

    Confidence scores + plain-language "why".

  4. 4

    Let users teach it

    Correct a result in place; it learns.

Confident — accept as-is Fairly sure — spot-check Unsure — review first

Every result carried a confidence score. People learned the scale fast and stopped re-checking things that didn't need it.

The results

0
Time saved on
routine data prep
0
Analysts who
opted in
Takeaway

Explainability is a design problem, not just an ML one. People don't need the math — just enough to know how far to trust a given answer.