CIMx — Daniel J. Baek

Overview

The Problem

Customers interacting with Intuit's tax and financial products regularly needed to submit, organize, and manage complex documents. The experience was fragmented across products and manual in nature — customers named files inconsistently, uploaded incomplete sets, and internal experts spent significant time chasing missing information.

CIMx introduced AI-powered document intelligence to this experience. Uploaded documents are analyzed automatically, organized into structured folders, checked for completeness and accuracy, and renamed for consistency. What previously required manual back-and-forth was reduced to a guided, intelligent flow.

[replace-me] Upload and organize flow

[replace-me] Document detection states

Role & Process

How I Worked

This work spanned three Intuit products — TurboTax, QuickBooks, and BizTax — each with different user bases and document contexts. I mapped the existing journey for both customers and internal experts to identify where time was lost and where AI could meaningfully intervene.

From there, I designed the AI-powered interactions: the upload experience, folder auto-organization logic, detection states for missing or incorrect documents, and auto-rename conventions. I worked in close partnership with AI and engineering to ensure the model outputs were surfaced in ways that felt helpful rather than alarming.

01

Journey Mapping

Mapped the full document submission journey across customer and expert touchpoints to surface where time and effort were being lost.

02

AI Interaction Design

Designed how AI outputs — folder structure, detection flags, renamed files — were communicated to users in a way that built trust rather than confusion.

03

Cross-product Rollout

Adapted the core patterns across TurboTax, QuickBooks, and BizTax, balancing consistency with each product's unique customer context.

[replace-me] Upload UI

[replace-me] Folder structure

Impact metrics

Key Decisions

Design Choices

AI transparency

[replace-me] How did you handle surfacing AI actions to users?

[replace-me] What was the decision around showing vs. hiding AI behavior, and why?

Error states

[replace-me] How did you design detection flags for missing or incorrect docs?

[replace-me] How did you balance helpfulness against causing unnecessary alarm?

Cross-product consistency

[replace-me] How did you adapt patterns across three products?

[replace-me] What stayed consistent and what varied, and why?

Outcomes

Impact & Results

23%

Handling time reduced

$3.54M

Realized cost savings

46 min

Saved per engagement

[replace-me] Add qualitative outcomes here: improved accuracy, expert satisfaction, reduced friction. What did you learn from this project?