Empowering Finance Workers with AI

Workday is an HR software provider. They had decided to make a large push into Finance and were looking for new ideas to gain market share. Managing expenses is a key function of any corporate finance team, supported by teams of Expense Partners who review expense reports for policy violations and fraudulent activities. 

We were tasked with finding a feasible solution to enable better management of their work, one that would layer ML based insights in to determine the risk of a report needing attention. 

My Role

I was the lead designer, responsible for guiding a team that included a junor designer and a researcher, running meetings, and leading product direction. Our team was distributed across the Bay Area but would meet in person weekly. 

Audience and Stakeholders

Audience : Expense partners

• Financial professionals with (typically) no formal training
• Review hundreds of incoming expense reports for policy violations and fraud,
• Front line workers protecting financial integrity by enforcing policies

Stakeholders 

Product managers from the Finance product organization. Many of whom were accountants or finance professionals in a previous life. As a result we were contending with strong opinions base on personal experience. 

Using task flows to understand Expense Partners' process

We interviewed expense partners at Workday as well as doing desk research into the industry. We built task and user flows to understand an expense partner's routine and pain points. Some key points we learned

• There is a core repetitive loop of identify / analyze / take action
• Partners depend on domain knowledge to make decisions, and usually focus on one subset of financial data
• Partners depend on colloboration and follow up both within their team and within the larger org

The simplified view of an expense partners process. 

The whole process, laid out end to end. 

Building a Persona

To help our stakeholders and people new to finance understand the expense partner we created a simple persona. This deliverable helped bring our stakeholders who couldn't speak corporate finance along as we developed concepts and kept us focused on meeting specific user needs. 

Understanding the Machine Learning Process through Collaboration

In order to understand the capabilities of the ML models, I lead workshops to understand the model training process and what we could present to users. From this we created a diverse catalog of potential solutions we could later insert into our mockups, speeding our creativity.

Heuristic Audit of the Workday Inbox

Our stakeholders asked us to graft our experience on to the existing Workday inbox. The Workday inbox is an odd duck, used for both job related messaging and to handle an Expense Partner's professional workload. In order to illustrate potential hiccups we conducted a heuristic audit in addition to collecting feedback from users and customers. 

We found 

• Important actions weren’t able to be surfaced on a enough high level to useful
• A user’s workload couldn’t be saved, forcing them to go through the process of setting up filters each time
• Delegation and collaboration wasn't efficient. Expense partners were sending emails or messages in other systems to coordinate their work

Putting it together : Testing two directions with users

Our work got us clearance to test building a custom interface. We qualitatively tested 

• A modified inbox or a custom view based on standard Workday patterns
• language around model confidence and granularity of message content

What we found

• The table view was vastly preferred over the inbox
• Simple was better for surfacing insights, but needed to be customizable 
• Configurability and customization was high on the list of needs, no two expense partners work exactly the same

Customer feedback sent a strong message to stakeholders and got us buy in to ditch the inbox. 

Our two concepts side, by side. 

Leveraging in-development features to improve the experience

I helped our team seek out in development features we could use to extend the core experience and meet the goals of our users. In particular, we sought to use collaboration and note taking features in our app. 

What we found is that corporate finance has strict legal and compliance requirements, and getting the features incorporated meant making the business case to take on the work to bring things up to our standards. 

Building and refining our MVP

We fleshed out and prototyped the full product. We were able to find an existing filterable table to build a custom view without heavy development overhead. We added in collaboration and audit features teams asked for, based on other features developed in the company. 

The main dashboard.

Digging into the expense lines. 

Our MVP went into development and maxed out early adopters. Customer satisfaction was high and our MVP aligned with the business goals of getting high volume financial professionals to use Workday's tools. The Expenses Back Office was a crucial first step for the Finance product organization into machine learning and automation.

Outcome and Lessons Learned

Our MVP succeeded.

• Launched to overwhelmingly positive reviews and a maxed out beta customer list
• A successful core experience and coherent backlog of vetted future improvements

But we had worked faster than devs could absorb the work.

We were so far ahead of development by this point my team was stuck in a loop of designing ancillary features while customers waited to get a basic version of the tool. Meanwhile, we had plenty of other priorities we could focus on. Working to assuage stakeholder fear was an important lesson in management. I referred to the project plan, which he had written together. Our stated goals included 

• Build a functional mvp
• Gain buy in from customer finance teams
• Prove the financial viability of the product

With these goals accomplished, we could afford to roll off as development caught up. In order to make this happen, I brokered an agreement between my director and the finance team that we would come back for a phase 2. 

Finance data is a whole other level of scrutiny

Features in a financial products are subject to increased legal and regulatory requirements, even above and beyond what a company that stores personal data would normally be subjected to. Because of this, we had to temper our expectations of what consumer features we could integrate until said features had updated privacy and audit functionality.