How we turned our biggest problem into our best user onboarding flow
How do you help someone find something, when they themselves don't know what they're looking for?
That was, in essence, the problem we faced at Art in Res. We're an online art marketplace, and our target demographic is new art buyers looking for affordable, original art.
With high bounce rates, we agonized over how best to tailor the art browsing experience for our users. Our mental model was something along these lines:
- People have limited attention spans, especially online.
- We have a big collection of artworks (thousands), which takes a long time to browse.
- If we don't show people artworks that they like, and quickly(!), they'll just leave.
We experimented and we conducted user research. We bought people coffee and watched them browse on their phones. We built features, we used third party tools. We found out that if you showed someone an artwork, they could tell if you if they liked it or not. That didn't help much – at first.
But then when we had a major breakthrough. Wait, we can do something with that. If you tell us "I like this," and "I don't like that," that's super useful information. With a little creativity (and some machine learning), we can use that information to show you more things you'll like.
One of my co-founders, John, got to work on building a recommendations engine using a collaborative filtering algorithm. We had lots of "artwork likes" to work with – “liking” artworks was one of our first features on the site. After some tinkering we had artwork recommendations that worked pretty darn well. We used pathgather – big shout out to the contributors on that project.
And that was great for existing users – we could use their likes to recommend new artworks. But what about our new users? They would still hit the website and bounce if we didn't make a good first impression, and we didn't have their like and dislike data yet.
Another co-founder, Dan, got to work on a "taste quiz," a big blue button that greets you when you hit our homepage. Even in its most basic form – like or dislike 20 artworks, sign up, and we'll give you recommendations – it worked wonders. Account creation spiked, and more sales followed.
I'm not sure what the moral of the story is. Maybe it's "Work with what you've got," like we did with our like and dislike data. Maybe it's "Talk to your users," which seems to help in nearly every situation. Would love to hear your thoughts in the comments. Anyway, this was a fun little saga that hopefully other indie hackers can learn from!
