The brief
The brief
Getting a company's finance systems running means defining its structure: legal entities, ledgers, chart of accounts, cost centers and more, often separately in each application. Today that work leans on systems integrators and produces brittle mappings and multiple sources of truth.
Leadership wanted an AI-driven platform that could take a company's own documents and data, propose the setup, and keep it consistent across applications. They asked for the vision first, with a tight scope that did one thing well. I was the only designer, working with a product lead and reviewing regularly with a senior executive.
What I found
An implementation admin applies the configuration. The finance teams, from planning to consolidation, are the ones who know whether it reflects how the company actually runs its money. In a consultant-led setup, the consultant carries that knowledge between them. Remove the consultant and that gap is exactly where the agent has to sit.
So the design question became: how do the people who know check what the agent proposed, before the person with the button applies it?
How I found it: persona working sessions with finance, consolidation and implementation roles, and before-and-after workflows for the administrator whose job the product would change most.
What I designed
Challenge: Leadership wanted the full vision first. The team also needed something buildable soon, and nobody had agreed on what working with this agent should feel like.
What I chose: I compared three models against the jobs people had to do: monitor progress, audit decisions, pick up where they left off, interrupt the agent, see its reasoning, and learn over time. The three were chat-first, a hybrid of chat with a persistent workspace, and a task flow driven by agent cards. My recommendation: ship chat-first with document output as the MVP, move to the hybrid workspace, then add task-flow progress tracking. All three shared one foundation: a versioned document library, an audit log of prompts and decisions, suggested responses, and controls to interrupt.
What I rejected: Building the full workspace vision first. It would have delayed learning from real use.
Result: A path from MVP to vision the team could plan against, without rebuilding the foundation at each step.
Challenge: The planned onboarding had the admin pick a pre-written template for their type of business, download it, pass it around to subject-matter experts, and enter their answers. Nothing useful could happen until that first round of feedback came back.
What I chose: Don't wait for it. The agent drafts a full finance configuration from publicly available company information, so the admin is met at onboarding with a ready-to-review finance blueprint, and the experts react to a draft instead of filling in a blank template. To prove it was feasible, I had the AI research a few public companies and generate their blueprints, then vetted the results with seasoned implementation consultants on our team. I also created a fictional company that sells wildfire alert systems to test other parts of the model. Once the core company information was in place, free-form prompts worked well for refining the draft.
What I rejected: The download-and-circulate template as the starting point. It left the admin with nothing to review until every expert had replied.
Result: A draft-first onboarding, tested on real public companies and vetted by experienced implementation consultants, so review can start the day a customer onboards.
Challenge: The plan already included a fallback to the existing manual setup screens for anything the agent couldn't handle. As planned, that meant leaving the agent and finishing the job somewhere else.
What I chose: For the MVP, bring the existing screens into the agent's interface as a frame, only when they're needed. People stay in one place, the agent keeps its context, and engineering reuses screens that already work.
What I rejected: Sending people out to a separate manual interface, which breaks the flow at exactly the moment the agent needs help.
Result: It was part of my MVP recommendation. The team hadn't built it by the time I left.
How it earns trust
What happened
In about four months, working as the only designer, I delivered personas, before-and-after workflows, an MVP and vision specification, a recommended interaction model, and a tested draft-first approach to onboarding, all reviewed with the senior executive sponsor. The 30–60% figure is the projection stakeholders used to plan the product; it was never measured. I left Oracle before the product launched.
Alongside the design, I wrote a UX strategy that framed design as part of how the product is built, with lifecycle ownership, shared artifacts the team could learn from, and visible return on investment. Its practical form was a toolkit of primers and models that helped engineers and product managers reason about agents, personas and workflows.
What I'd change: I'd test the drafted blueprint against a real customer's internal details much earlier. Public information gets the broad structure right, but internal specifics are where the agent's limits show up first.
"Chris brings big-picture thinking and strong execution, and he has a real talent for communicating ideas in a way that gets everyone aligned and energized."
— Yuwei Fu, Director of UX