Design of AI Products & Services (by John Zimmer)
April. 2019 (1 month)
Ethan Ye / Ja Young Lee / Qian Wang
My Role
UX Designer
Venmo is a popular mobile transaction App. By December 2018, it has ~27 million users globally . About 66 % of young Americans are using this app.
How can we improve Venmo experience with AI technology?
Millions of transactions are made on Venmo everyday. Many of the transactions are re-occurrences. How can we utilize AI technology to improve the experience?
A more efficient experience through smart suggestions
We designed two types of suggestions: one is the list view, the other is the pop-up view. Pop-up view is used for more certain suggestions, which can be location-based, or user-configured. We also designed suggestions at two stages: one is transaction-level, the other is more detailed on the transaction page.
We first did a few user interviews with primarily university students along with some secondary research
Insight 1
Transaction flow can be tiresome especially when you are doing it repeatedly
Through user interviews we learned that people often have recurrent transactions such as monthly rent, phone bills etc. The repeated transactions can be payments or requests. So we concluded the typical payment/request workflow as shown below, and decided to use AI to make the process more efficient.
Payment Flow
Request Flow
Insight 2
People often use the same message for similar transactions
Interestingly, we discovered that people often use the same comment message for transactions, especially emojis. And they find typing the same amount and description once and once again annoying. This inspired us to design the suggestions for amount and comments.
Insight 3
Many students have regular transactions
Through the interviews, we discovered that people often have transactions like rents, phone bills, utilities etc. on a regular basis. This inspired us to allow user set reminders on their own.
Insight 4
Tolerance for wrong requests are higher than for payments
AI has uncertainty, so it's important to understand users' tolerance for mistakes. We learned that people has higher tolerance for wrongly suggested requests than payments. This inspired us to design different forms showing different levels of intrusiveness while making suggestions.
Insight 5
Food and drinks seems to be where people use Venmo the most
We don't have the actual log data to identify the most common purpose for transactions on Venmo. But based on our interviews, the No.1 reason for Venmo transactions is paying friends for food. This align with a study we found online, which shows more than half of the top 15 emojis in Venmo are food or drinks. This inspired us to specifically design for payments at restaurants.
top 14 emojis on Venmo
Suggest transaction
On the "Add Recipients" page, we designed a section showing suggested transactions based time, date, location, history transactions etc. The amount and comment message will be automatically filled.
Error recovery
Because the workflow is extremely fast, users can accidentally make a transaction, we also designed a recall button for error recovery.

Key iterations

Location-based suggestion
When detected you are at a restaurant, user will be prompt a quicker way to pay or request nearby friends who also had their Venmo opened. This can also be used for stores that support Venmo
Suggest transaction detail
Because people often use emojis or messages repeatedly, sometimes even the amount will be repeated for regular payments. On the transaction page, we also added some suggestions for users to quickly choose from.

Key iterations

Request reminder
We used the pop-up window to show suggested requests with higher confidence. User can also set reminders themselves to pay friends the rent, utilities, phone bills etc periodically.

Key iterations

Privacy Settings
User might be concerned with their data being used, so all the adaptations are opt-in, and can be configured in the privacy setting. User can also specify which data to grand access for.
Things we did correctly 😉

This project is about practicing a method to apply AI on simplifying workflows. We successfully identified the workflow with potentials and iterated to find the right way for adaptation.

Things we can improve 🤔

Due to the short time of this project and the educational purpose, we couldn't do thorough research to validate the user value of our idea. We also couldn't do enough user testing on the adapted workflow. So if given more time, I'd focus on these two part more.