Intuit · QuickBooks
AI-powered invoicing
A from-scratch rethink of QuickBooks invoicing

~64% of researched invoices never get sent, with $13B+ stuck in drafts. I led the from-scratch rethink of QuickBooks invoicing to fix create-to-send.
Starting from zero, every time
Most invoices and proposals start from something that already exists: a past invoice, a client's proposal template, an old PDF. QuickBooks made people rebuild that from scratch.

Merchants creating invoices and accountants drafting proposals both had the same workaround: keep a reference document on hand and manually recreate it in QuickBooks, field by field. The editor had no way to start from what someone already had.
“Worst screen orientation ever. You have to scroll down just to see any details of the invoice. The entire top section is wasted space.”
Edit · layout“I create it in QuickBooks but I always add some colors to it in Canva before I send it from my own email.”
Design · off-platform
From reference document to draft, instantly
Upload what you already have. AI turns it into a ready-to-use QuickBooks template.
I designed a feature that lets users upload an existing invoice (for merchants) or proposal (for accountants), and AI auto-creates a matching template inside QuickBooks, pulling in layout, structure, and line-item patterns so people start from something close to finished instead of a blank page.
What we optimized for
The calls we made when every option had a real downside.

Create-to-send vs. showing every field
Prioritized line items and totals on the edit canvas; moved advanced and cohort-specific controls into Settings.

One editor vs. cohort-specific flows
Kept AE, MAIP, and merchant in a single invoice experience, with conditional payment sections instead of forking the product.

North-star vision vs. incremental rollout
Defined the full Edit · Settings · Design split upfront; shipped payments settings first while the Design tab targets a later release.
One AI toolchain, from insight to shipped code
The call: design an AI upload-to-template feature, then use AI to build it end to end, solo, in days instead of quarters.
I clustered VOC into design principles with Claude, mapped the upload-to-template flow in Figma, and built a clickable prototype in Cursor, so leadership reacted to real flows, not static frames. That same toolchain carried into production: when polish sat in the eng backlog, I opened PRs in the codebase and traced defects with Claude. Three examples:
Insight → prototype
Claude to cluster, Cursor to prototype
Clustered VOC into design principles with Claude, then built a clickable upload-to-template prototype in Cursor, so leadership reacted to real flows, not static frames.
Pull requests
UI polish merged to production
Navigated the QBO codebase in Cursor, opened PRs for spacing, copy, and component fixes, and shipped them through normal review.
Bug fixes
LLM-assisted debugging
Traced invoicing defects with Claude, patched regressions, and merged the fixes the same week.
From concept to shipping product
The AI upload-to-template feature gave merchants and accountants a faster path from reference document to finished invoice. Throughout, I shipped production PRs and fixes rather than waiting for a handoff.
Want to see the actual designs?
This case study covers confidential work, so the real screens aren't shown here. Email me at dianlee99@gmail.com and I'll walk you through it.