Reinventing how Deloitte Audit teams collect and format data

the process

The solution: direct connections to data sources

Of the many explorations we tested to solve the problem of receiving wrong or incorrect data, we settled on connecting directly to client databases. While this approach required installing software on client servers, it reduced miscommunications, kept the process secure, and covered most problem areas:

• It made database browsing accessible to less technical users.
• Using a “what-you-see-is-what-you-get” made database browsing accessible to all users, bringing the rate of incorrect data requests close to zero.
• The new approach made data requests repeatable and consistent, which was critical to enable formatting automation.

‍

Minimizing delays and setbacks

Human interactions were the root cause of delays. Not involving critical personnel such as Database admins or approvers, miscommunications of data requirements contributed to delays in delivering the required data.

‍

To minimize the impact of human neglect, I ensured that approvers could approve requests at any time, from anywhere. I made this process as easy as possible, including all the information required to approve a request.

‍

Automating data formatting

Once the data has arrived, the next step is converting it into the desired format to be able to conduct the audit. I designed the Data formatting app with these goals in mind:

• The formatting app's main goal is to increase efficiency and enable reusability of work.
• It should be easy to use and accessible to less technical users.
• It should be future-proof and able to process millions of rows (big data).

‍

The process: small and frequent tests with few users

To build such a complex app, my approach was to iterate as much as possible. So I ran small tests once or twice per week with 4–5 users. During tests, I asked users to think out loud so I could know what they were thinking, despite contaminating task completion times.

‍

Dozens of design explorations

The final Data Formatting app

After testing the initial ideas, I selected the final layout based on test performance. I wanted the UI to feel familiar enough that users could get an idea of how it works at a glance, but not so familiar that they would fall into their default behaviors. See the final design below:

‍

Helping users see data as a whole, not single cells

Imagine you had to change a million date entries from this pattern: "12302022 16:12:44" to "12/30/2022". Rather than changing entries one at a time, you want to identify the underlying pattern and change it to the desired format. When faced with this test task, about half of the tester's first instinct was to edit individual cells. I wanted users to stir away from this behavior, so I did a few things.

‍

I disabled single cell selection entirely and turned the click-to-select interaction into column selection.

‍

I emphasized rows over individual cells by styling rows using a zebra pattern.

‍

Adding data distribution diagrams to column headers was the most impactful change. Users loved using the graph and statistic charts to find inconsistencies in the data.

Making the app accessible to less technical users

During early tests, users seemed afraid of trying things in the app and using the UI. To help them feel comfortable using the app, I pushed several initiatives.

To create the array of tools, I looked at apps that users were already familiar with. The common denominator was the MS Office Ribbon, so I decided to style the toolbar using this pattern. The idea was to help new users feel comfortable and reduce the feeling of overwhelm when confronting such a complex app for the first time.

‍

Some testers still struggled finding some of the tools. So I included a search option that parsed through tool descriptions and offered quick access to documentation.

‍

While tools were active, users had to decide on the formatting operation they would make but hesitated to commit. When prompted, users mentioned that they didn't know what the result would look like. To address this, I created formatting previews, which helped users understand the consequences of their actions.

Despite all the efforts to make the toolbar attractive to users, most testers preferred using tools directly from the column headers. So instead of correcting the behavior, I decided to lean into it and increase the number of quick-launch options in the dialog.

With great power comes... slower execution?

As we were refining the app, our early testers started complaining about long loading times (8+ seconds to load even a small dataset), which seemed excessive compared to Excel.

After examining the problem and talking to engineers, I learned that this "slow" performance was expected with the tech we were using. But, unfortunately, engineers couldn't reduce loading times any further.

‍

How I created “busy” time to make up for the wait

As an initial "fix, " I replaced the default spinner with a progress bar, giving users a better sense of progress while waiting.

Then I timed the loading times of 50 data samples to understand how fast data loaded overall. Then, I used the insights to test different timing distributions and found that users perceived loading times as shorter when the first 90% of the load went faster, slowing down at the end.

See the best-performing timings below.

To add some delight while users waited, I included randomized messages, and inserted Audit and Accounting puns amidst the messages.

‍

‍

What happened? Complaints stopped and auditors loved it!

Rolling out the updated loading experience not only stopped complaints but also resulted in a massive increase in sign-up requests. It turns out auditors were showing the new loading experience to their peers, who loved it and wanted to try the app themselves!

Making data formatting repeatable

I designed the steps to feel like a "recipe": a sequence of repeatable actions to get the data into the proper format. The idea was to save every tool used as a "step", allowing you to re-do the formatting with the press of a button. This is also why we called it the "Recipe".

‍

While designing the Steps, I wanted users to understand how the data changed quickly. So I used natural language to describe each step.

Additionally, saving each tool as a step in the sequence freed users from the fear of breaking their perfectly formatted dataset and encouraged experimentation.

‍

Finding a better way to explain data connections

During tests, I noticed that 40 to 60% of testers struggled to understand how the Data sidebar worked consistently. I tried to solve the problem by iterating on the labeling, information architecture, and UI, but the problem persisted.

‍

‍

A Flow map to show the relationships between the data

I approached the problem from two angles. First, I got rid of the Data sidebar. Then, I designed a UI layer on top, the Data Flows, to give users a clear picture of how the data was being processed. Users found this representation of the data connections to be self-explanatory.

‍

‍

‍

The Data Flows just made sense

By replacing the old sidebar with the new Data Flows, the overall experience  improved on many levels:

• According to 70% of the testers, the new Data flows were easy to understand, accurately depicting the data relationships.
• Removing the sidebar gave users room to focus on a single dataset simultaneously, reducing the cognitive load.
• Removing the Data sidebar reduced the user's ability to trigger loading events, which improved perceived performance overall.

‍

‍

Not only did Data Flows allow users to reuse the recipe of steps over and over, but they also made automated formatting features more evident to users.

‍

‍

At the same time, Data Flows enabled the app to scale exponentially by affording the option to add new tools and integrations. And it was able to achieve this without cluttering the UI.

business impact

The Outcomes

Overall, Audit and other teams at Deloitte received the app very positively. As a result, we reduced the time spent collecting and formatting data so much that the app earned Deloitte 2 Audit of the Year Awards in 2018 and 2020, respectively.

‍

THANKS FOR READING!

Want to see more work?

Check out this case study!

‍