Artificial Hype? eBay's History Of Unfulfilled AI Promises
eBay's Chief AI Officer Nitzan Mekel-Bobrov has been making the media rounds trumpeting the company's AI efforts, saying this emerging tech will bring a paradigm shift that will "completely transform ecommerce" in the next 3-5 years.
Nitzan Mekel-Bobrov joined eBay in 2021 as the company's first chief AI officer. The role was an evolution of the chief-scientist position and a signal of how seriously the e-commerce marketplace company took artificial intelligence.
The company is bringing AI to both sides of the e-commerce equation, innovating for both buyers and sellers on its platform. Its initiatives have included generative-AI tools to make it easier to list products, a 3D-imaging tool, natural-language search to improve search results for buyers, and AI-powered expert content for product categories such as electronics and motor parts...
...Mekel-Bobrov told Insider he viewed the arrival of AI as a "paradigm shift" that "will completely transform e-commerce" and outlined how he saw that transformation taking place over the next three to five years.
It's interesting that Mekel-Bobrov is still touting the 3D image experience for sneakers as an example of eBay's supposed AI prowess.
It's really just bolted on white labeled functionality developed by third party RestAR/Unity and so far in the 2 years since the initial launch, eBay has failed to expand it to any other categories on the platform - not exactly a rousing success.
eBay has been making similar predictions about AI's "transformational" abilities for years before ChatGPT and other Large Language Models came on the scene, but many of those previously promised paradigm shifts never panned out.
In 2016, then CEO Devin Wenig told CNBC that eBay in the future would use AI to "find you [buyers] where you are" - a somewhat unsettling prospect for privacy conscious consumers, especially considering the troubling details that have since emerged of the cyberstalking and harassment of journalists by eBay security personnel during Wenig's tenure.
Someday, eBay wants to be able to find its more than 162 million active buyers wherever they are online, its chief executive said.
Devin Wenig, eBay president and CEO, told CNBC’s “Fast Money: Halftime Report” on Thursday that he has a vision for how artificial intelligence will help eBay ahead of the commerce field in the next three to four years.
“There’ll always be an eBay destination, or an app, but there’s a fabric that lives beyond eBay,” Wenig said. “It’ll find you where you are. You may be in a social network, you may be in a messaging application, you may be living in other digital properties. As long as you express intent, I’d like eBay to come to you and bring you the right item.”
Wenig expanded on his thoughts about AI at the 2016 Code Conference (where he also ironically positioned himself as a proponent of free speech and a free press when asked his opinions on then brewing controversy between Gawker Media and PayPal co-founder Peter Thiel).
It's [AI] really important to me and I think it's important to our company. Predictive modeling is not new for ecommerce. For ten years we and all of our competitors have been trying to say hey people have bought this also bought this but it was pretty crude and it was based around inventory clustering...
...and I think it's a combination of computing power GPUs being brought online to model big data sets and the sophistication of algorithms which now inventory clustering is only one factor. We're now looking at demographics and other data and we're layering all of that on top of each other in very sophisticated predictive models...
...What's the result to the customer? Relevance. It's all about consumer relevance. I go on eBay I want it to feel like it's my eBay I don't want it to feel like it's your eBay I want it to feel like you know me. You know whether I want a $7 scarf or a $70 scarf, you know whether I like blue or red, you know that you already know the car so you know the bumper.
It's very personal and not intrusive personal seamless personal and I actually think we are getting really close to that.
As part of those AI-powered personalization efforts, eBay introduced ShopBot Beta in October 2016.
This smart personal shopping assistant could only be accessed via Facebook Messenger, which was an interesting choice considering at the same time Facebook Marketplace was launching with an initial product that competed on a local level but would later expand to directly challenge eBay with full ecommerce shopping capabilities.
In addition to major enhancements we’ve been making to our core platform, we’ve also been hard at work on a new commerce experience powered by artificial intelligence (AI), cloud computing, and human judgment. We’re starting to bring this experience to life with eBay ShopBot, which debuts today in beta on Facebook Messenger.
eBay ShopBot is a personalized shopping assistant that helps people find the best deals from eBay’s one billion listings. Our vision is to make shopping with eBay as easy as talking to a friend, whether you are looking for something specific or just browsing for inspiration...
...The science of AI provides contextual understanding, predictive modeling, and machine learning abilities. Combining AI with eBay’s breadth of inventory and unique selection will enable us to create a radically better and more personal shopping experience for virtually anyone that owns a mobile phone.
You can beta test the early beginnings of eBay ShopBot by texting, talking or snapping a picture to tell eBay ShopBot what you are looking for. The bot will then ask you questions to better understand your intent, and make personalized recommendations.
Unfortunately, the eBay ShopBot did not gain wide adoption and was eventually shelved.
In 2017, eBay turned their sights to the promise of image recognition technology to enhance search.
Not sure how to describe what you’re looking for? eBay is making it as simple as snapping a picture.
Starting today, our new image-recognition features are live on eBay’s mobile apps -- Find It On eBay and Image Search. These features allow shoppers to use pictures instead of words to search eBay’s 1.1 billion listings, spanning the new and truly unique. Enabled by artificial intelligence and machine learning, these innovative ways to shop will help you quickly find the things you love, making the entire internet shoppable on your mobile device.
Using AI to Transform the Way People Are Shopping
With Find It On eBay, you can start your search on any social platform, like Facebook, or while browsing your favorite blog or website. Simply “Share” the image with eBay and the eBay mobile app will surface similar listings.
However, eBay's search by image functionality has always been a bit iffy in practice and was removed from the app for several months in 2021 because there were "inconsistencies" in results.
When they finally re-enabled search by image 3 months later, the results were even less useful than before and the feature often struggled to identify truly relevant or similar items in search.
Wenig told CNET at the time that AI is "not a feature, it's the next major computing platform shift."
You've said eBay wants to democratize AI for small and medium-sized businesses. How?
Wenig:... We're doing a lot of work on image recognition. Computers are getting really good at recognizing things. We're not that far away from a future where you'll just do this. [Wenig picks up his phone and acts out taking photos of a pen in front of him.]
You'll just take a picture of that and we'll match it through AI to our catalogue, we'll know exactly what that pen is. And we'll say, do you wanna sell it? We can get you five bucks right now for that pen.
6 years later, current CEO Jamie Iannone is singing a similar tune, promising a Magic Listing experience that will allow sellers to simply take a picture and let AI do the rest is "coming soon" - though eBay has given no timeline for when this "employee beta" will actually make it into the hands of sellers.
Iannone announced the new feature at Goldman Sachs Communacopia & Technology Conference 2023, providing a mocked up demo video for investors.
So, the example that you saw there, which we call the magical listing flow, that went from like a concept to idea in a matter of like two months -- sorry, from concept to being a working version of it in two months...
...The version we showed there is a product that's currently an employee beta, which is, how do I figure out what a product is and do all the hard work so the seller doesn't have to do anything?
And I'll just tell you a story. We're cleaning up my garage with my wife, and she finds this old remote control from a stereo I had 25 years ago. And she's like, "Okay, throw this out. Like, Jamie, we don't have the stereo. It's been 25 years." And I'm like, "Okay, better idea, let me sell it on eBay."
So, I literally just hold up my phone to it, and it figures out what this remote control is, it figures out what stereos it's compatible with, it writes the description for this remote control, and I listed on the platform. It says to price it at $30. I priced it at $30. And guess what, a week-and-a-half later, this remote control sells for $30. That would have absolutely been in the trash without this technology.
However, the gap between that magical vision and the current reality on eBay may be wider than Jamie would like to admit.
The AI description generator eBay introduced a few months ago did not have a smooth launch - in fact they had to temporarily pull it from the app due to a bug which caused "unintended consequences that exposed a broken experience."
And just last week, the introduction of the AI description generator in the Desktop Listing Experience had similar results, being removed from production just days after launch.
What about those image recognition capabilities Wenig was so sure would be a game changer?
In addition to the often problematic search by image, eBay also built a "scan to list" tool in 2021 for collectible trading card games like Pokémon and Yu-Gi-Oh! but for some reason were never able to expand the tool to include sports trading cards or other collectible card games.
More recently eBay has been testing a "find visually similar" feature on the View Item page that supposedly uses AI image recognition to suggest similar items to the buyer.
However, the results show eBay still has a long way to go before their image recognition tech can reliably return actually visually similar results.
At eBay Open 2023, VP Buyer Experience and Focus Categories Blair Ethington sayid eBay has focused on improving search this year using AI.
This past year, we've focused on creating a simple design across the site that helps buyers purchase the items they love on eBay. We've also made significant changes to our search experience, utilizing AI and machine learning to enable buyers to find your items...
...This year, with the help of AI, we've improved our visual search results to be more relevant, with image similarity predictions closer to real-world expectations. This is more convenient for buyers as it does not require the use of specific keywords.
However, in addition to the questionable performance of visual search, other so called "improvements" have caused massive problems with how search functions on the site.
For example, in February, June, and August of this year, eBay had multiple events which caused the search engine to be completely non-functional, returning 0 results for search terms that should have had 10s of thousand of results.
At other times, search has repeated page one results over and over and autocorrect continues to show wildly inaccurate and irrelevant results to buyers - is that the magical AI component taking over?
Image recognition and augmented reality were top of eBay's priority list when they lured Jan Pedersen away from Twitter and named him Chief Scientist, Artificial Intelligence in February 2018.
eBay has hired Jan Pedersen as vice president and chief scientist, Artificial Intelligence (AI). Pedersen will lead the company’s AI strategy, including computer vision, natural language understanding and machine learning, to deliver new customer experiences across the eBay platform.
“Jan is a true pioneer in the industry, with over thirty years developing search, deep learning, machine learning and AI technologies at scale,” said Devin Wenig, president and CEO of eBay Inc. “He joins us at a pivotal moment when AI sciences including computer vision and deep learning are now capable of transforming personalized, immersive shopping experiences. eBay runs on AI and our continued innovation under Jan’s leadership will transform each meaningful moment we have with our buyers and sellers.”
Pedersen will be responsible for charting eBay’s technology strategy as it relates to AI, including enhanced ways in which eBay interacts with its consumers. He will set the agenda for AI R&D investments, lead a team of research, data and applied scientists and further our leadership in AI-enabled commerce.
We've already looked at how the image recognition initiative has turned out, but what about augmented reality?
eBay introduced an AR feature to help sellers identify which USPS Flat Rate shipping box will best fit their items in March 2018.
When shipping an item, finding the right box can be a hassle. Today, eBay unveiled a clever solution using Augmented Reality (AR) that enables millions of sellers to quickly select the best USPS flat rate box for items they need to ship.
Built on Google's ARCore platform, which enables high-quality AR experiences on Android, eBay is taking advantage of motion tracking and environmental understanding to place a real-world item inside virtual shipping boxes of various sizes
In the new Android feature, users select a box size, place the virtual box over their sold item, and move the box around the item to see a complete view of the product inside the box. This allows sellers to virtually see if an item fits inside any USPS Flat Rate shipping box before they choose the physical box or try to package the item.
Unfortunately, this augmented reality shipping feature was also eventually shelved, though it's not clear if it was due to lack of performance or if sellers simply did not find it a useful addition to their workflow.
Pedersen and then SVP Americas Scott Cutler presented a keynote about AI at ShopTalk 2018, including the AR shipping tool, but the examples showing how it could be used in the sneaker category were interesting to look back on as well since Cutler jumped ship to lead rival marketplace StockX in 2019, taking a chunk of the sneaker resale market with him.
Pedersen's stint at eBay was notably short lived. After only 9 months on the job, he abruptly left to head up Uber's AI efforts from 2018-2020 and is now VP Search and Personalization Technologies at Walmart.
In early 2019, Wenig was very publicly called out by activist investor Elliott Management for ongoing technical failures and management misexecution in a letter to the Board of Directors calling for substantial changes at the company (emphasis mine).
Today eBay suffers from an inefficient organizational structure, wasteful spend and a misallocation of resources. By increasing operational efficiency, eBay can free up capital to invest in capability- and revenue-enhancing activities.
...As an online marketplace that provides a critical forum for millions of buyers and sellers, the efficient and effective functioning of the platform is paramount. Unfortunately,eBay has been plagued by technical problems and operational challenges for years...
...Fast forward to recent years and the platform still faces issues. In 2018, eBay sellers complained about countless technical issues including incorrect billing, lost photos, warped titles and many others. On this month’s end of year podcast, eBay senior management apologized to sellers and admitted, “This is a 2018 that we don’t want to repeat on a number of levels. And the technology issues that we have had with the platform is top of the list.”
We agree: The consistent reliability of the platform is central to eBay’s success, and management must do all that it takes to achieve it.
While innovative endeavors in new pursuits like machine learning and augmented reality are promising future technologies,eBay’s publicly touted initiatives in these areas will add little value if the core platform continues to have critical functionality failures.
While Elliott has since divested its eBay shares, it's surprising just how much of the 2019 letter still rings true today.
The site is still plagued by ongoing, near daily business impacting technical issues and operational challenges and current management is clearly not doing all that it takes to achieve consistent reliability of the platform.
Just like machine learning and AR before, AI may be a promising future technology, but it will add little value if the core infrastructure continues to falter.
Will AI really "completely transform ecommerce" as Mekel-Bobrov predicts?
It will take a true paradigm shift to operational and technical excellence (and a Board of Directors willing to hold executive leadership accountable) for eBay to pull Iannone's promised magical AI rabbit out of a hat.