Featured on Wall of Portfolios
‹ Work

Intuit

Sales Orders

Order-to-invoice for millions — Desktop PO parity, manufacturing flows.

Role
Senior Product Designer · QuickBooks Online

Highlights

  • Designing Sales Orders in QuickBooks Online, used by over 36 million companies worldwide, reducing manual accountant effort across order-to-invoice.
  • Introduced Purchase Order capabilities carried over from QuickBooks Desktop, closing parity gaps for migrating customers.
  • Building AI-powered product enhancements to automate repetitive accounting tasks and surface proactive insights for accountants.
  • Led multiple rounds of customer research with accountants and small businesses.

Proving out an AI-first seller vision

Intuit Design for Delight

Proving Out an AI-First Seller Vision

Two-day workshop process: diverge, converge, re-ground, converge, experiment

  1. Deep Customer Empathy

    Service Blueprint + JTBD Seller pain, shared baseline

    Day 1 · 10:30 AM
  2. Go Broad

    Define “autonomous,” build AI-native inspiration, ideate 4 groups, unconstrained futures

    Day 1 · 12:45 PM
  3. Go Narrow

    Prototype, share out, surface top 6-8 jobs + assumptions “What needs to be true?”

    Day 1 · 2:45 PM
  4. Re-Ground in Empathy

    VPs join, Blueprint revisit Sales Leaders add field color

    Day 2 · 8:35 AM
  5. Prioritize & Sequence

    Dot vote on seller impact E2E story readout to VPs

    Day 2 · 10:15 AM
  6. Rapid Experiments

    Quick wins + phased pilots Architect gut-checks

    Day 2 · 12:30 PM
  • Deep Customer Empathy
  • Go Broad
  • Go Narrow
  • Rapid Experiments

Day 1 · Apr 22Day 2 · Apr 23

A case study from QuickBooks Online

Rebuilding sales orders for QuickBooks

Customers were using sales orders. They just couldn’t finish with them, because every next step meant typing the order again.

A packing line in a factory with stacked finished goodsWhiskey barrels aging on wooden racks
RoleLead Product Designer, sales orders
Team5 designers, product managers, 10 engineers
PlatformQuickBooks Online and Intuit Enterprise Suite
TimelineFY26 to FY27
6.7×more customers creating a sales order each month, 1,977 to 13,210
2×more people tried creating with AI autofill, 2.6% to 5.4% in 8 weeks
41%came back to create again within 14 days
70+usability sessions in 2 months

The record that runs the order

A sales order is a non-posting transaction that records a customer’s request to buy specific goods at agreed prices and terms. It reserves stock and kicks off fulfillment. Every step after it reads from it.

EstimateA proposal
Sales orderConfirms and reserves
Purchase orderBuy what’s short
Pick, pack, shipFulfil
InvoiceGet paid
Whiskey barrels aging on wooden racks
Stock is money on a rack.
Workers sorting raw produce into bins
Raw materials, bought first.
Hands adjusting a production machine
Work in progress.
A packing line with stacked finished goods
Finished goods.
A forklift loading pallets into a shipping container
And when it leaves, the books have to know.

I designed for small and mid-market product businesses: makers, distributors and manufacturers who live in this world every day.

The problem

People wanted it. Most didn’t come back.

Before the redesign, only 40% of first-time users created another sales order within 60 days. The product had demand, not fit.

37kcustomers had created a sales order since launch
428ksales orders created
6k+monthly active users
40%came back within 60 days

Most product businesses never reached it

Bars not to scale. Source: QuickBooks Online usage dashboard.

700+ pieces of feedback, two clear themes

8,731invoices already built from several sales orders, by hand
35,332sales orders linked to more than one invoice
“I lack a streamlined way to convert estimates to sales orders, create invoices from sales orders, or generate purchase orders for out-of-stock and drop-ship items.”

The customer problem statement. This one problem drove 28% of negative feedback on sales orders.

Research

We went to the floor before we opened Figma

I visited factories and warehouses, then synthesised everything with the team at Intuit’s Mountain View office. Amplitude showed where people dropped off; usability sessions showed why.

40follow-me-home sessions
250+survey responses from Online and Desktop customers
40+manufacturing businesses, small to mid-market
3accountants, to validate the money side
A whiteboard used to track orders in a small manufacturer
Smaller manufacturers ran orders on whiteboards.
A whiteboard covered in handwritten inventory counts
Inventory counts, by hand, on the wall.
1A document to us, the start of everything to them

Customers saw the sales order as the beginning of buying, making, shipping and billing, which is why retyping each next step hurt so much.

2Status was invisible

Whiteboards ran the floor because the screen couldn’t show where an order was.

3They fixed it after

People converted a sales order, then immediately corrected the purchase order it produced. So we fill in only what’s certain, and say why.

Goals

Three North Star numbers defined fit

Repeat use

40% before60%+ goal
60%+

Monthly active users

6k+ before30k+ goal
30k+

Desktop parity

80% goal
80%
Approach

The first fix didn’t work. The second did.

We shipped a quick fix for one problem and it didn’t move the KPIs. So we stepped back, changed how we worked, and rebuilt the order in phases.

Attempt one

  1. Found it. Retyping to convert drove 28% of negative feedback.
  2. Flagged it to product and leadership.
  3. Shipped a quick fix for one part of the problem.
  4. The KPIs didn’t move toward 60% repeat use or 30k monthly users.

Attempt two

  1. Test first. Define the scope and the process before designing.
  2. Five scopes: link, customize, find, money and trust, and AI. Linking first.
  3. Rebuilt in phases: linking, then templates and statuses, AI autofill, deposits and many-to-one invoicing.
  4. It moved: 1,977 to 13,210 monthly users.

Desktop got the hard parts right

Sales orders came from QuickBooks Desktop, decades-old software. Our brief: keep its power, lose its weight.

QuickBooks Desktop Create Sales Orders window with Create Invoice and Create Purchase Order buttons
Desktop: convert straight from the order.
QuickBooks Desktop warning that there isn’t enough quantity to sell
Desktop: a stock check before you promise.

What we kept

Convert from the order

Stock check before you promise

Customer context on the form

Templates for pick lists and packing slips

What we changed

A toolbar of buttons became one Action menu

Blocking pop-ups became status you can see

One desktop became anywhere, online

Deposits off to the side became deposits on the order

QuickBooks Online sales orders list with the Action menu open
Online: every next step in one Action menu, with order and payment status side by side.
QuickBooks Online sales order showing linked transactions
Online: every linked record in one place.
Decisions

Every big call changed because of evidence

AreaWe thoughtWe learnedWe changed
Scope

Fix everything customers asked for.

Linking alone drove 28% of negative feedback.

Linking first, inside five scopes.

Invoicing

One sales order, one invoice.

8,731 invoices were already built from several orders by hand.

Many sales orders onto one invoice.

Purchasing

Auto-split one purchase order per vendor.

It added risk to the core conversion.

Kept auto-split as a later stretch goal.

Deposits

A deposit is just a payment on the order.

It’s a liability, and passing it along wrongly counts it twice.

Reused the proven deposits-on-estimates accounting pattern.

Shipping

Ship-to addresses can wait.

13.9M Desktop customers use several ship-to addresses.

Made it the first FY27 item.

Deep dive one

AI autofill: every customer writes orders differently

Owners receive purchase orders from their customers in every format and retype each line into a sales order. Customers in different regions write orders in different ways, so the AI has to read each one, and a wrong line is worse than an empty one.

A pill or a drawer?

AI entry points at Intuit have a dedicated pattern, and that started an internal debate about the right component. We worked it through with the wider design team against the QuickBooks design system and its values, then tested both.

DrawerTested
Rich, but out of the way until opened.
PillShipped
Right where the work starts: 15% better intent discoverability.

How autofill works

QuickBooks Online sales order with the Autofill panel open: Files and Text tabs, an Autofill from file button and a drop zone for PDF, JPG or PNG
1 Autofill opens beside the order: drop in the customer’s file (PDF, JPG or PNG) or paste the text.
Autofill reading a customer's PDF order, with a Does this look right? card asking to match a line to an existing product or create a new one, suggestion 1 of 4
2 Each line is checked before it lands: match it to an existing product or create a new one, one suggestion at a time.
2×more people tried creating, 2.6% to 5.4% in 8 weeks
41%came back to create within 14 days
+12ptsatisfaction on hard prompts, 61 to 73, after the model change
+15%intent discoverability, pill against drawer
Deep dive two

Manufacturers build on deposits, not promises

A maker pays for materials long before the customer pays them. Deposits had to live on the sales order, without ever letting the books lie.

55kDesktop Enterprise users, 20.5% of the base, take prepayments every month
~$40Ma month already flowed through deposits on estimates in QuickBooks Online

A deposit isn’t income. It’s money you owe back.

Deposit received→Customer deposit, a liability→Invoice created→Applied to accounts receivable

Passed from estimate to sales order to invoice the wrong way, it creates double journal entries. What was broken in Online:

Only one deposit per order, where Desktop allows many
No way to take 100% up front
The deposit didn’t travel from estimate to sales order to invoice
One deposit couldn’t be split across several invoices
QuickBooks Online settings with deposits turned on for sales orders
Turn deposits on once, in settings.
QuickBooks Online sales order with the deposit drawer open
Take the deposit right on the order. The drawer asks only what a deposit needs.
QuickBooks Online sales order list showing a deposit paid status
Payment status shows on the list the moment money comes in.
QuickBooks Online journal entry moving the deposit from liability to receivables
Underneath, the proven estimates pattern keeps the books exact.
Design system and values

Built on the system, argued in the open

We worked inside the QuickBooks design system, and when a pattern didn’t fit, we debated it before adding anything new. Every screen followed the QuickBooks content principles.

It’s about them, not us

We designed around a business owner’s Friday, not our data model.

Speak their language

“Receive deposit” and “Convert to purchase order”: words owners already use.

Focus on the payoff

Every confirmation says what changed and where.

Keep it simple

The deposit drawer asks only what a deposit needs.

Genuine

We never guess a vendor or a balance, and we say why a field is blank.

One component, one job

The pill or drawer debate ended with data, not opinion.

QuickBooks Online sales order with a confirmation toast after adding a deposit
The confirmation names exactly what happened.
Team

Carry it, then hand it off

For the first three quarters I carried sales, purchase and manufacturing orders myself. Once it stabilized, two designers took over purchase and manufacturing orders with a free hand, and I moved to guiding them.

JJam sessions

Personal flows and mapping first, then team jam sessions to make decisions together.

EEngineering early

Early buy-in from engineering, so constraints shaped the design instead of breaking it late.

TTools I built

Ship It, for faster internal bug fixes, and Snap comment, to log Jira bugs straight from a comment.

DDisagreements, resolved by definition

Early arguments came from unclear definitions. We tested first, defined the work and the process, then designed.

Results

6.7× more customers, every month

Goal: 30,000 Jan 2025Aug 2026 1,97713,210

Customers creating a sales order each month, QuickBooks Online usage dashboard. 44% of the way to the 30k goal.

85,906customers created at least one sales order, Jan 2025 to Sep 2026
Feb 2026sales orders launched to mid-market in Intuit Enterprise Suite
55%+of feedback themes addressed for that launch
41%of AI autofill users came back within 14 days

Adoption grew, but growth alone doesn’t prove design caused it. The before and after monthly figures may count different customer bases, so I read them as direction, not attribution.

What’s next

The order keeps growing

Next on the roadmap: several shipping addresses, custom tax rates, linking to manufacturing orders, drop-ship parity, partial conversions and printing shipping labels.

If I did it again, I’d use AI to change how we work from day one: a new workflow for talking to customers, closer collaboration, and faster feedback so we refine sooner.

I define the scope and test before I design.

Clarity first saved us a second rebuild.

I let data decide, and write everything down.

Every decision above has a number behind it.

I give my team a free hand.

And build tools that make everyone faster.

Thanks for reading.

Anuj Bhargava, Lead Product Designer at Intuit, Bangalore.

Next case studyIztri