We know that when we put relevant, interesting designs in front of customers they engage, and the AI work will help us do this in a much better way.
In another reflection of the changing nature of the e-commerce sector in the Vaccine Economy, customers at online designer marketplace Redbubble are increasingly “value-oriented”, according to CEO Michael Ilczynski. What that means in practice is that while revenues are still growing, helped by an expanding US t-shirt business, the company is having to make some tough decisions around its own costs, announcing a 14% reduction in headcount this week.
So is reaching for AI tech the way to turn things around? There have already been transformative gambits in the past, such as the 2018 acquisition of TeePublic, which Ilcynski points out had an impact in terms of targeted demographics:
TeePublic's target consumer market is a bit different to Redbubble, primarily Gen Y who are 26 to 40, and their revenue is much more concentrated around apparel sales to US consumers.
Redbubble's target market is Gen Z households, so 12-25 year olds and their parents…Having the two marketplaces has proved beneficial for the group in recent years in a changing consumer landscape as each marketplace is at a period where it has outperformed the other.
During the COVID-19 pandemic, there was strong demand for homewares, which benefited Redbubble. More recently, we have seen demand for apparel in the US remains strong, benefiting TeePublic. Their relative performance provides valuable insight for the group and increasingly, we're implementing successful initiatives from one marketplace onto the other.
Both marketplaces operate a flywheel, he goes on, where improving one element creates a positive impact on another:
The content that artists sell attracts customers and as customers purchase, that enables the fulfillment network to scale, lowering costs and attracting additional customers. This increase in customers attracts small artists, creates more artist revenue, encouraging new artists to the platform. They add more content, and thus more customers, and the cycle continues. For many years, we have spoken about the benefit of the flywheel. We continue to believe that ensuring the flywheel is operating efficiently will enable us to deliver long-term growth.
That said, there’s room for improvement in key areas and that calls for tech investment in the right places, particularly in terms of improving the content experience. This is where the AI aspect comes into play. Ilcynski explains:
Historically, we have spoken about how Redbubble's growing content library was a key competitive element for the marketplace. A core advantage of the content library is having something that can appeal to the most niche tastes. And historically, we've seen a strong positive correlation between more content and revenue growth.
However, around 18 months ago, this relationship started to break down as we had a surge in the volume of new content uploaded to the Redbubble marketplace. Unfortunately, a lot of this content was non-additive. It wasn't particularly unique or creative.
Given the volume, and the way it was described by the artists, it impacted the perceived quality of search results and the overall customer experience. Often Page 2 had better search results than Page 1, which is not ideal for a search-driven experience. As a result, we've seen a drop in on-site conversion as potential customers, people who have never visited the site before decreased in their engagement in add-to-cart rates.
Another factor here was that the pandemic-driven boost in the e-commerce sector as a whole masked some of these issues. But actions are now being taken, including the deliberate introduction of more friction. Ilcynski says:
We've added much more friction when artists are uploading content to the Redbubble marketplace to promote the uploading of additive content only. We've also introduced new technology, as well as changed processes, within our group artist team. That team is built out of TeePublic's artist acquisition team and is now taking a lead role in artists acquisition across both marketplaces.
It’s working, he says - from October last year through to the end of January, Redbubble saw a 35% drop in new content being added, while uploads from more established artists increased. AI may boost this initiative even further, suggests Ilcynski:
While we're pleased with the improvement in reducing non-additive uploads, we are most excited about the opportunity to step-change our understanding of artists' content through the use of AI. This will enable improvement in search and discovery across both marketplaces, both on and off-site, and solve a core, long-standing challenge of how to build an objective, accurate understanding of each piece of content available in the marketplace, so we can place the right design in front of the right consumer every time.
This last point is incredibly important, he adds:
Our search functionality is reliant on data provided by the artist, the title, the description and the tags. This means that, up until now, our understanding of each piece of content has been dependent on the data the artist provides. Unfortunately, this data is not always accurate, sometimes accidentally and sometimes deliberately. This can mean that relevant designs appear when a potential customer searches.
As there are currently millions of designs on the marketplace, solving this problem at scale has been an ongoing challenge for Redbubble. It is just not possible with human review - more than 60 million different individual designs and every piece of information an artist has provided about each one of them. However, the advances in AI in the last 12 months hold the key to addressing this issue where we will be able to remove our sole reliance on artist supplied information as we can use AI to objectively enrich the data on every piece of content.
AI has provided accurate text information for content on the site. This will significantly increase the relevancy of search results. AI can also provide an alternate search engine in terms of surfacing relevant images for a given search, and it can detect highly similar content that evades our current duplicate detection.
And there are other potential benefits around surfacing relevant content to customers:
We are really excited about the potential of this technology and saying it could revolutionize search and discovery on the Redbubble marketplace is not an overstatement. And this is not a pipe dream. We have already tested this on several thousand designs, and we have just recently produced AI-generated data - or tags - across the entire Redbubble content library. Now it is all about testing and learning, and they're applying this at scale into the marketplace. This will take some time, but we expect to be at scale in production during this calendar year.
This isn’t something that’s going to be hit customers at once, he cautions:
Just to be crystal clear, we have not yet tested this on our platforms yet out to consumers. So I'm not going to point you towards specific metric uplift yet. That is the exact step we're moving into, literally over the next few weeks. What we have done, though, is we have applied this technology not just to a small corps, but across the whole marketplace. So we know it can be done, as we've done it. We've also human-reviewed samples of this to test for accuracy. We are very confident in the accuracy and what it can add, particularly in terms of increased relevancy and reduced irrelevancy, which we know is a real bugbear for customers, for artists etc.
Overall, while he acknowledges that the current macro-economic pressures are not going to ease off any time soon, Ilczynski argues that underling long-term trends are still in favor of the marketplace:
Over the past decade, we have seen a steady increase in the penetration of online shopping, which was accelerated by the COVID-19 pandemic, with many individuals trying online shopping for the first time.
I, like many others, expected to see a plateau in online demand following this acceleration. Instead, in many categories, we have seen a pullback, and that impacted our ability to generate enough of a return from the investments we've made over the past two years.
However, buying goods online has become a more accepted practice among a much broader cohort of society, and we expect e-commerce penetration will normalize at a higher level than before the pandemic in the near future.
This is something that we're already well, well into.
While some siren voices will mutter about silver bullets here, this use of AI tech is an interesting real-world application that appears to fit a clear business/operational need. This is an initiative that’s going to be well worth keeping an eye on as the testing/learning phase moves into customer-facing delivery and, it is to be hoped, an uplift in metrics.