Lisha Lokwani
Back to work

Case Study · AI · Data

Replacing chat with a workbook for serious analysis.

At Petavue, the chat-first model worked for exploration but collapsed at decision time. Each query was isolated; insights had to be stitched together by hand. Workbook reframed the core experience around how analysts actually think — in flows, not questions.

Role

Founding Designer

Scope

Core product redesign

Timeline

2025 — 2026

Team

1 designer · 4 engineers · 1 PM

Workbook canvas showing a Q1 Pipeline Review plan with five analysis steps on the left and a Segment Activity Effectiveness data table on the right.
Workbook · multi-sheet analysis flow with persistent context

The problem

Chat is great for asking, terrible for deciding.

Analysts came to Petavue to make calls — pricing changes, retention interventions, churn forecasts. The chat surface answered each question well in isolation, but every decision required a chain of twelve. Context evaporated between prompts. Edits meant re-prompting. The transcript became a flat record of attempts, not a record of thinking.

Hand-drawn sketch comparing the current Q→A chat experience, which stops at answers, with the analyst's real mental model: data scope, overview, breakdown, drivers, details, and actions — a flow that builds understanding.
Field-note sketch · chat gives answers; the analyst's mental model builds understanding.

Key insight

Data analysts don't think in questions. They think in flows — a goal, broken into steps, built incrementally, refined until the insight is clear.
From eight weeks of analyst interviews · 2025

The solution

A surface built around workflows, not turns.

Workbook replaces the chat thread with a multi-sheet canvas. Each sheet is a discrete step in the analyst's reasoning, fully editable and re-runnable. Three product moves carry most of the weight:

Multi-sheet workflow

Sheets become explicit thinking steps — break a goal into stages, organize them spatially, and keep each step inspectable.

Persistent context

Every prompt, output and edit lives inside the workbook. Pick up where you left off, branch a step, or hand work to a teammate.

Iterative building

Modify any earlier step and downstream blocks recompute. Iteration replaces the chat-style restart-from-scratch loop.

Workbook memo view rendering a Sales Activity Effectiveness analysis with key findings, ranked patterns, and a contextual action menu for turning the analysis into a chart, quick analysis, or saved definition.
From analysis to artifact · the same workbook turns into a shareable memo, chart, or saved definition.

Design decisions

Three calls that shaped the whole product.

Structure

Structure as a feature, not a constraint.

Chat looks lightweight but pushes every cognitive burden onto the user. The workbook trades a heavier first impression for far lower load during the actual work.

Iteration

Design for the second pass, not the first prompt.

The most valuable analysis happens on iteration three, not query one. Every interaction was tuned for editing, branching, and re-running — not just authoring.

Adoption

Behavior change is the real shipping risk.

Users had built muscle memory around chat. We staged the rollout so existing prompts opened inside a workbook, letting people migrate without learning a new mental model upfront.

What I took from it

Designing for iteration beats designing for output.

Workbook reframed how the team measured success: not the quality of a single answer, but the speed at which an analyst could refine toward a defensible decision. Behavior change carried more risk than UI change — the rollout had to teach a new mental model without asking anyone to abandon a familiar one.

· Workflow, not query· Editable, not append-only· Persistent, not ephemeral

Up next

More case studies in progress.

See all work