← Home AI in the product
Oracle NetSuite · Director of Experience Design · 2024–26

Making AI coherent across an ERP

RoleDirector of Experience Design
Team3 senior managers · ~20 people
ScopeIA · design system · patterns · visual design · AI patterns
ProductERP used by 43,000+ businesses in 220 countries
The problem

NetSuite is one suite covering finance, inventory, CRM, ecommerce, analytics and more, used by over 43,000 businesses in 220 countries, and shaped differently for a dozen industries.

Oracle's commitment for NetSuite Next was that AI would be built in at every layer, not bolted on.

Read that as a design brief and it's brutal. Every module, every role, every industry gets AI at once. Either the way AI behaves is decided once, or it gets invented twelve times and the suite stops being one product.

And this is a system of record. An assistant that's confidently wrong in a notes app is an annoyance. In a general ledger it's an audit problem.

A NetSuite business insight card surfacing a detected increase in returns, over a dense ledger screen
The same AI pattern appearing across another NetSuite module
What I led

I led the foundations: the information architecture, design system, components, patterns and visual language every product team builds on. At this size, foundations aren't documentation. They're the only place a decision can be made once and actually land everywhere.

Three tensions had to hold at the same time.

Consistent without being uniform

"One source of truth" and "shaped by your industry" pull in opposite directions. AI had to feel native to a manufacturer's inventory workflow and a nonprofit's reporting, while staying recognisably the same product.

We fixed the behaviour and left the content free. How AI announces itself, hands back an answer, and takes a correction: identical everywhere. What it talks about, and the vocabulary it uses, belongs to the industry. Teams got no say in the grammar and full say in the language.

Conversation on top of density

ERP users are experts. They live in dense forms, keyboard paths and saved searches, and they measure a release by whether it slowed them down. A conversational interface is the opposite: linear, low-density, patient. Deciding when AI leads and when it stays out of the way was the core interaction question, not a feature.

We landed on two modes rather than one. AI surfaces where you're already looking: insights arriving in the dashboard you'd have opened anyway, no conversation required. Ask Oracle is there when you choose it, not in front of you when you don't.

That was the unglamorous choice. A single conversational front door demos better and reads more like the future. But an accountant closing the books doesn't want a chat window between them and the ledger, and the people who pay for an ERP are the people who already know where everything is.

Trust as an interface, not a disclaimer

Source citations on every response is a public commitment, and it lands in the foundations. Provenance had to be visible enough to verify and quiet enough not to wreck a screen already carrying a hundred numbers.

Provenance rides with the answer instead of interrupting it. Every response carries its sources, one step from the number, so verification is available and never compulsory. Trust isn't a warning you dismiss, it's the ability to check without leaving what you're doing.

Underneath all three, we restructured the system onto tokens, which turned density and theme (including dark mode) into properties the system resolves rather than decisions every team remakes. Fewer decisions in the modules is what coherence actually costs.

And we didn't govern from a distance. My designers sat inside the AI product sprints, so patterns arrived with the people who understood why they existed.

I stayed in the work too. Through the SuiteWorld run-up I was in Figma designing the home dashboard you see at 1:04, and coordinating program managers and translation teams through the launch.

AI output arriving inside a dense NetSuite screen
Source citations shown in context with an AI response
The NetSuite Next home dashboard
What shipped

NetSuite Next launched with Ask Oracle at its centre: one conversational way to search, analyse and act across the system, with source citations on every response.

It's public: Oracle offers a free three-day test drive of the real product. netsuite.com ↗

We held a flagship launch date while the product changed shape underneath us.

Under NDA. Everything shown here is Oracle's own published material. The thinking is mine; the confidential detail stays confidential.