It Takes Two to Tango
Bits of love for the other 90% of a payment
As I’ve previously written, one major contributor to Venmo’s success is its recognition that payments fall into the category of social interactions that are enabled by software, alongside conversations (chat, email), storytelling (Instagram, Snapchat, Facebook), networking (LinkedIn), etc., and deserve the same level of love and care, which it has not historically received. There are always two (or more) human beings involved in a person to person payment, and the interactions between them comprise the other 90% of the payment — i.e., everything besides the transaction itself.
In my first year and a half at Venmo, we were focused on the small but important details that improve the experience of that Other 90%. There were three areas in particular that we felt were key to those interactions and therefore materially impacted individual and group engagement with Venmo:
- Connecting with the right people
- Involving and notifying friends
- Communication and self-expression
Working with PMs, engineers, and other designers, I designed a few improvements and features in each area.
Connecting with the right people
Features: usernames, improved profiles, improved search
Venmo is a community, and finding and connecting with the right people are a central part of a good experience — it’s a requisite part of transacting as well as managing your content and privacy preferences.
We worked with our support team and conducted usability studies and user interviews to uncover areas in which we could improve. We also looked at Mixpanel analytics to get a sense of their scale and scope. After gathering enough data, some themes stood out:
- It can be hard to know if you’re paying the right person — lots of people have similar names and unclear or missing photos
- I don’t see people I expect to see in Search results
- When I’m on someone’s profile, it’s not clear what I need to do to pay them and the requirements are unclear. Do they need to be my friend?
Finding and identifying with people on Venmo was hard.
Most people liked to use the Pay or Request screen to find a recipient. However, we also saw that people also liked to check a person’s profile in order to verify that they’ve found the right person before transacting with them. Since they couldn’t do this from the Pay or Request screen, they’d end up going to Search. In either case, the issues above led to uncertainty and sometimes mistakes. At best, people would back out when they were unsure and look for another way to make the payment. At worst, they’d pay the wrong person, which meant they had to contact support. We knew it was crucial to solve these problems — people shouldn’t have to worry about these things when they use Venmo.
Goals and Principles
With an understanding of the problems, the team agreed that we were looking for a solution that makes it easier and faster to be sure you’ve found the right person, and clearer how to pay them once you have. More specifically, that meant
- Showing more relevant search results
- Making it easier to identify people
- Simplifying and improving calls to action
We tracked support tickets and search/profile analytics to see if we’re making progress towards that goal.
The PMs and I wrote up a few project docs to help everyone think through possible approaches and nuances. Then, the team came up with some potential solutions without constraining the possibilities. Some of them were:
- Rank search results by relevance and clearly indicate top results
- Improve search/friend graph algorithm
- Show relevant hints, like mutual friends, in results and on profile pages
- Indicate whether a person is a Venmo user and/or a phone or Facebook contact
- Add an easy way to preview profiles from the Pay and Request screen
- Use Nearby to prioritize people close by
- More prominent profile pictures
- Add profile photo and username setup during onboarding
- Improve information hierarchy of profile page
- Add scannable QR codes as a shortcut for people who are together in person
- Emphasize reliable forms of ID, like usernames or phone numbers
We rated our ideas on likely impact and technical effort and decided to start by experimenting with low-effort ideas that could help us learn about which heuristics were more valuable and which interactions were more natural. We also favored ideas that we could prototype and test without requiring production level platform work.
We focused on usernames and photos, improved profiles, clearer search results, and profile previews.
Accordingly, we focused on usernames and photos, improved profiles, clearer search results, and profile previews. Meanwhile, our platform team started work on improving the search algorithm and updating the architecture such that it’d be easy to experiment with sort order and retrieve data like mutual friends.
I mocked up concepts and built some interactive prototypes that the team could play with and test in user studies. On concepts that touched onboarding, I paired with the indomitable Keith (a fellow designer focusing on new user experience at that time).
We took a hacksaw to the concepts and refined them to their most essential components so that we could build and test quickly. After a few internal iterations to tune usability, we felt that we were ready to ship the experiments — since they were relatively incremental changes, we felt we’d learn the most from getting them into the hands of real users, getting feedback, and watching our KPIs.
What we shipped
Here’s a before and after of the iOS app. An analogous version was shipped for Android.
Results and lessons
After about 4 months,
- Weekly support cases related to paying the wrong person went down significantly, by two thirds
After about 8 months (in June 2015),
- The % of users with custom usernames had increased 5x
- A bit over half of all users had a profile picture, a modest increase from before
Support cases related to paying the wrong person went down by two thirds.
We went into the project interested in learning:
- What should be weighed strongly when calculating relevance?
- What kinds or combinations of information are the most useful aids in correctly identifying someone?
- What’s the right info hierarchy on the profile page?
We found that profile info and shared payment history were the best indicators that you knew someone
Profile info and shared payment activity were most useful
- Larger and more profile pictures helped
- Payment history on the profile page also helped
- High frequency and most recent payment partners were more relevant
- Profile preview on the payment screen was useful but hard to discover
Network connections were another strong signal
- Overlap between Venmo friends and address book contacts (if you had granted access) was another useful signal.
- Mutual friend count was also a solid tiebreaker
Usernames needed a more reinforcement.
- Phone number continues to be the most popular fallback method for reliably paying someone. We’d made a bet on usernames with moderate traction, but they weren’t yet a universal identifier.
- People had trouble remembering usernames because they didn’t seem them very often.
Profile pages should emphasize action
- The large Pay or Request CTA worked well to reinforce the primary purpose of the profile page
Taking the lessons above, we continued to tune the search results sort order to show Top Friends first, then friends on Venmo, then others on Venmo. We added mutual friend count to non-friend results. We removed address book contacts not on Venmo from Pay or Request search results because they created complexity and confusion. We’ve made profile previews more discoverable in upcoming releases.
We also felt that we could push usernames further to really put them to the test, so we doubled down on them with the Mentions project (keep reading!).
We had put a dent in the problem, but there was more work to be done.
There’s still a lot of work to be done, but I’m proud of what the team accomplished and continues to improve — this is the type of problem for which success is people not noticing what you’ve done because things are working as they should, and it’s become much less of a hot topic lately.
Involving and notifying friends
At around the same time as the work on connecting with people, another interesting opportunity presented itself. Venmo was growing bigger and bigger, and one trend we noticed was that people often split their communication about a payment so that some of it happened inside of Venmo and other parts outside. For example, when a payment involved several people — such as when one person fronts the cost for a dinner or group hotel booking — the group would sometimes coordinate through group texts or Facebook events in addition to sending requests and reminders through Venmo. This was especially true of users who were natural organizers (we called them “Coordinators”) and liked to plan and document things meticulously.
People still needed to go outside of Venmo for to organize conversations about payments.
In other words, people still needed to go outside of Venmo to fulfill some of their payment interaction needs. Was there an opportunity to help Coordinators and others by centralizing some tasks and better complementing existing workflows for others? We thought there might be.
Goals and Principles
There was a lot that we could do in this area. Eventually, the team would conduct more in-depth research, collaborate with Stanford design student teams to uncover all the possibilities, and launch experiments like Venmo for Groups. But in our first stab at it, we looked for a more lightweight and flexible way — formal groups, while a natural eventuality, would be a big change for our users and a larger technical investment. So, we began experimenting with a battle-tested interaction that leveraged our recent work around usernames and was true to our brand philosophy of distilling out the fun part of payments: Mentions.
Mentions should work the way everyone is accustomed to in other services.
We were all quite familiar with tagging in other services. The team knew that Venmo’s implementation should
- Work the way everyone expects and is accustomed to
- Let me involve and notify friends both during and after making a payment
- Helps me bring friends into relevant or interesting conversations around payments that I didn’t make
- Keep me in control of my payment and post privacy
- Leverage and reinforce the then-recent aforementioned bet on usernames as a central form of ID
We were interested in seeing if people would respond and rely on Venmo more for group coordination and discussion. To see how we were doing, we tracked direct feature usage as well as overall engagement (comments and likes on payments). Related payments involving mentions to the same person was another signal.
Since tagging interactions are fairly ubiquitous, the process was straightforward. It was clear from our goals that mentions should be enabled in both the payment note and comments section. I mocked up a couple screens and our engineers quickly built out a prototype.
How could mentions be triggered? Two obvious ways were 1) a button on the payment note screen, and 2) by typing an “@”. But since this was a new behavior, it was likely that users would miss or forget to use them. To solve for this, the team built an automatic trigger that showed up when a enough letters of a matching name was typed.
We experimented with the prototype to find parameters that weren’t overaggressive for the automatic case. We also added analytics to keep track of how frequently each trigger was used over time.
A big question was how mentions would affect payment privacy. It’s really important that people can trust Venmo and feel in control of their privacy, so we took a conservative approach. Venmo already let users control whether payment partners can change a post’s privacy settings and deferred to the more restrictive default between the two; the team applied the same rules to mentions. This meant that mentioning someone who could not normally see the payment would do nothing.
Privacy was really important, so the team took a conservative approach.
Another question was whether mentions should appear as real names or usernames. The former was more recognizable in the short term, but the latter would reinforce our bet on usernames and was more scalable as a unique Venmo ID in the long run. We created mockups to help us visualize how each felt in the Feed and comment threads and ultimately went with usernames.
Finally, our users needed a way to learn about the feature. I created concepts for two complementary approaches that the team then tested: an introduction screen and an in-context tooltip. There were a few copy variants we tried.
In the search project described earlier, we’d tested username setup in onboarding for new users, and this was a great opportunity to do the same for existing users; it’d help them claim a personalized username they’d be happy with.
What we shipped
Here’s the iOS version. The same thing shipped on Android.
Click the gear in the lower-right corner if the video is blurry.
Results and lessons
A few days after launch,
- Almost half of the active user base had completed the onboarding modal
- Out of those, a third had changed their default username, which meant the % of users with custom usernames had doubled.
Our questions going in:
- Which triggers did people tend to use?
- How would people respond to the overall concept? How would they use it?
- Would we see a pickup in custom username usage?
- What about general engagement?
- Early on, a majority of users were using the button on the pay screen — probably thanks to the tooltip.
- As time went on, this tapered down and automatic mentions became the majority, indicating that most people weren’t actively thinking about tagging friends, but were open to doing so when reminded.
- People seemed to be using it to notify other friends involved in a payment activity, as expected:
- As mentioned above, we did see a significant increase in custom username usage
- There was a moderate bump in like/comment engagement (as a % of payments) in the 4–6 months immediately after launch, but it wasn’t significant in the long term
Overall, usage didn’t stick after the initial rush.
- Overall, usage didn’t stick after the initial rush. We spoke to some users who said that the feature was still too barebones — it’d need to do more than pushing a temporary notification to be a meaningful replacement for texting and other ways to communicate about a payment outside of Venmo.
It was clear: while useful in some cases, Mentions would need to do a lot more to become a sticky behavior.
There were a couple options: experiment with ways to make Mentions more engaging and useful, or leverage it to build something else. Some obvious ideas for the former included a permanent notifications center, so it’d be easier to find payments I’ve been tagged in, or a shortcut to use a completed payment as a template for others by tagging them (e.g. I paid Eric for a hotel and can tag Jesse and Mike to do the same) — if Mentions improved the core task of making payments on Venmo, maybe it’d become more top of mind.
After weighing the engineering effort and likely impact on our goals, the team ultimately decided to go with the latter. We’d return to group interactions on Venmo in due time; in the meantime, the autosuggest framework we’d built for mentions paved the way for a hack project that became one of Venmo’s more iconic features.
Communication: The Venmo Language
Feature: Emoji on Venmo
Emoji were becoming an unofficial language on Venmo. People, including ourselves, were scrolling the iOS keyboard a whole bunch to find the right ones to use. Why not make it a whole lot easier?
Emoji started as a weekend hack project by Daz.
Emoji actually started as a weekend hack project by Daz, one of our iOS engineers. He related the story of Venmoing his friend when out for late night NYC pizza and finding himself scrolling endlessly to find the right emoji to us. Being the stud that he is, he sent a working prototype built on the Mentions framework out to the team. We’d fancifully discussed similar ideas in the past, so it resonated strongly with everyone. It had to become an official feature.
I paired with Daz and Aditya, a PM, to iron out the nuances. The prototype was already functionally complete and identical to Mentions in terms of interaction, so my role was primarily to think about how to tune it uniquely for emoji usage.
How could we tune the Mentions interaction uniquely for emoji usage?
Daz and I talked about the autosuggest interaction: unlike Mentions, which matched names, emoji names matched with many everyday words. That meant that the rules had to be tighter to ensure that the list didn’t annoy people by popping up for every word they typed. We settled on showing suggestions after 4 characters had been typed, but with an exceptions list for common ones like “cab.” In solving for this, though, we uncovered a more exciting idea.
For scalability, Daz had built a couple JSON dictionaries mapping emoji names to their respective characters and enumerating the exceptions. As we went through the initial list to ensure the names were search-friendly, we had an idea: why not add custom strings for common Venmo payments like “rent”? That would truly be a step towards defining a Venmo language with emoji at the center.
In Venmo language, you don’t “pay rent,” you “🏡💸”
Best of all, anyone with a great idea could easily open a pull request and get it added. The Venmo language, like all real languages, is crowdsourced.
We asked the team at large for their help, and it came in droves. Soon, we had an ample set for initial release.
What we shipped
Emoji shipped in May 2015.
Results and lessons
1 out of every 2 payments today contain at least one emoji.
About seven months after Emoji launched, 1 out of every 2 payments on Venmo had at least one emoji in it — a 2x increase.
Unlike with Mentions, autosuggestions had a huge impact in reminding people to use Emoji and establishing it as habitual behavior — there was no reason not to replace boring text with fun pictographs. People saw more and more emoji in their friends’ feeds and began to follow suit.
Emoji caught on thanks to autosuggestions and the feed.
A secret language
We noticed people starting to create their own secret codes using Emoji strings that were cryptic yet interesting to a third party. A fascinatingly appropriate outcome of a platform full of social media users that revolves around sharing traditionally private financial information.
Emoji are now commonly associated with Venmo. That’s great, since our brand is all about distilling out the far part of payments.
Emoji had started as a hack project, but we still were curious about whether our solution matched the way people wanted to use emoji. Remember, the motivating problem was that we were tired of scrolling and looking for emoji in the iOS keyboard, so our solution was decidedly search-driven — type what you’re thinking or looking for and we’ll suggest emoji that fits.
As it turns out, this wasn’t enough! People do like scrolling through an array of because you don’t always know what you’re looking for. Plus, it’s nice to be able to spam a whole bunch of them at once 🔥🔥🔥🍆🔥🔥👌😁.
The optimal solution combines search with a curated set of options to choose from
The optimal solution combines both approaches: search for when I need it, autosuggestions for when I’m just typing, and a contextual, curated set of options for when I’m looking for inspiration. Our subsequent work in Pay with Venmo drew on some of these insights, and there are a few more ideas cooking.
In addition to the above, we continue to add new emoji strings to the the dictionary, including limited-time ones for events like March Madness and holidays.
The amount of evolution that the social experience on Venmo has seen over the past two and a half years is staggering and sometimes easy to forget. “Social” isn’t just sharing what you paid for — a payment is inherently social; it’s an interaction between one or more people. The team’s done some awesome work on this front and I feel lucky to have been a part of it.