In this interview series, I talk to ecommerce experts, industry professionals, and thought leaders with in-depth knowledge and experience in online shopping trends. As a part of this series, I had the pleasure of interviewing Eli Finkelshteyn.
Can you tell us a bit about your backstory and how you grew up?
I was born in Ukraine. My family immigrated here when I was a child, and I grew up in Tennessee.
I’ve long been interested in data science and engineering, and when I was growing up, I taught myself how to build and fix computers and then, eventually, how to code. I worked at Walgreens in high school, and during my breaks, I’d go through—little by little—a big book on computer engineering. Of course, all that knowledge is obsolete now. Still, at the time, it sparked my interest in the field, helped me get started, and eventually led to a love of data science and solving problems relying heavily on statistics.
I was also very interested in languages—a passion that’s stayed with me. Growing up, I spoke Russian at home and English in school while getting to study some Hebrew and Aramaic on the side. Then, in high school and college, I got especially interested in more ancient languages. It’s fascinating to read texts that people wrote thousands of years ago. Those people were very different from you and me, but the idea of still being able to understand them—and still finding jokes they were telling thousands of years ago funny—struck me as really cool. Later on, I picked up some Spanish, Croatian, and Vietnamese, getting interested in the sound and structure of different languages.
Both of these interests—data science and languages—connect to what I’m doing today. To engineer really good product discovery experiences, you have to understand people who are very different from you: their speech, their likes, their needs, and their intent. And you need to understand them on a massive, zoomed-out level that requires heavy reliance on statistics because we’re talking about billions of queries a week. My belief is the better we can do that, the better an experience we can create for our customers and their customers.
What led you to this specific career path?
I didn’t always know I was going to help brands connect shoppers to the products they love. But the field is a great fit for me; I really enjoy working in this area.
As for how I got here—I wound up getting my master’s in computational linguistics and did my thesis on speaker sentiment analysis in ancient languages. After that, I was looking for something that combined all my interests. That is, I really liked the machine learning part of data science, and a lot of my interest in languages was rooted in the psychology of how words can affect people. So, tackling issues related to product search and user intent made a lot of sense.
Still, it took some trial and error to find the right path. I spent a small amount of time as a data engineer at Tumblr when it was still a startup, working on a spam and phishing detection system, and realized it wasn’t for me. And then a fortuitous thing happened: A stock photography company reached out. They were looking for help on the search side—something I hadn’t given a lot of thought to at that point, assuming Google had already solved search issues across industries.
But I quickly realized that to help shoppers find what they need, ecommerce companies had a different problem that Google wasn’t solving and that they were all building their own solutions on top of search engines built for just matching on keywords (the same engines most companies still rely on today).
So, I took a job as a search engineer at Shutterstock in its early days, working on its search engine and search algorithms. I saw how even the very small changes we implemented impacted the bottom line. For example, when I helped build a way to compare what words sounded like—to account for and address spelling errors in search and autosuggest—that increased conversion rates by 9.8% when Shutterstock A/B tested it.
Knowing I could have that kind of impact on the company by doing work I found fascinating started a lifelong love for the product discovery and search space. From that point on, I knew the field of ecommerce product discovery, and search was something I could spend my career on. I later started a company, Constructor, to help retailers successfully address these issues.
Can you share the most exciting story that has happened to you since you began at your company?
The most exciting time for me at Constructor was when we went live with our first enterprise customer: Jet.com, an online marketplace that Walmart acquired and later wound down.
It was a huge milestone for our business. We only had four employees at the time; I was doing most of my work out of cafes and a free AWS coworking space near my house. And we didn’t have our full platform for product search at that point either: only Constructor’s autosuggest functionality.
So, having a big company take a shot on us was both thrilling and gratifying. Jet.com was competing directly with Amazon and had huge amounts of traffic—thousands of queries per second—that we were handling. When they agreed to A/B test us, and we were able to show that our autosuggest positively impacted their business, it felt like winning the Super Bowl, and it was a feeling I wanted to replicate as many times as I could.
Even amidst the excitement, we knew that we needed to finish building out our full platform for product discovery and search. We did a couple of years later—and when early customers such as Sephora and Bonobos signed on, it was another momentous time. They also saw through A/B tests that the revenue lift from our platform far exceeded what they were paying us, bought into our vision, and are still customers today. We knew we were ready for “prime time,” and the business took off.
Today, ecommerce companies worldwide use Constructor to give shoppers highly personalized experiences across channels. So when someone searches for “shirts” on a retail site, for instance, they don’t see just any old shirt (or pages and pages of shirts that don’t interest them), but, rather, ones that map to their preferences: favorite brand, colors, price point, etc. Or, if someone on a grocery site is looking for milk, and that person tends to buy organic, we’ll display organic milk first.
We don’t personalize for personalization’s sake, so it’s not always obvious where it happens, but the A/B tests speak for themselves. Retailers can also optimize our platform’s results and recommendations for their KPIs, including conversions, gross profit, and so on.
To deliver this high-impact product discovery and personalization, our clickstream-based AI factors each shopper’s history and behaviors across the retailer’s digital channels (e.g., what did the person click on? spend the most active time viewing? add to cart? purchase? etc.).
We also apply information from “Quizzes” that retailers give their shoppers and pull in data from retailers’ other data streams, such as loyalty program information, to get a well-rounded and actionable picture of each shopper and their intent. Our platform continually refines results and experiences hundreds of millions of times per day based on shoppers’ search queries and how shoppers interact with the results and recommendations they see.
What are some of the most interesting or exciting projects you are working on now? How do you think that might help people?
There are a number of projects at Constructor I’m really excited about right now, some on the shopper side and some on the merchandising side.
Oftentimes, shoppers have product discovery needs that aren’t addressed by traditional search-and-browse technologies. To find the items they’re looking for, shoppers need to be able to engage more deeply—doing more than entering terse phrases and keywords into an engine—but they’re constrained by the technology that exists.
So, one of our newer products, Quizzes, addresses this. It lets retailers pose a series of brief questions to their online shoppers, much like an associate would in-store. The retailer can immediately and dynamically serve up personalized recommendations and guides based on the individual's responses. So, let’s say a shopper is looking for toys for their nephew but has no idea what to purchase. After answering a couple of questions about their nephew’s age, hobbies, interests, favorite franchises, and so on, the shopper will get highly personalized and unique recommendations to purchase with confidence.
As for other projects and ideas: applying ChatGPT and its underlying technologies (large language models—LLMs—and transformers) to ecommerce product discovery is something we and our retail partners are extremely excited about. In the business landscape nowadays, nothing is more transformative and talked about than ChatGPT. But rather than jump on the ChatGPT bandwagon because it’s “cool,” we think it’s important to use the new frontiers in AI in a way that’s truly useful to retailers and shoppers alike.
To that end, I’m especially enthusiastic about the potential for ChatGPT to bring valuable changes to online search. The succinct way that people today interact with ecommerce search engines works in some cases—but it’s not always the best way to search. There are times when shoppers have a more involved question or a long-form ask, like: “What’s a good dessert I can make for someone who likes strawberries and blueberries and is gluten-free?” Or “I’m going to a friend’s backyard wedding in Southern California in May. What should I wear?”
ChatGPT can understand and answer questions like these. Still, there are limitations: It doesn’t know all the items in a retailer’s product catalog, and it also doesn’t know each shopper’s preferences and history with the brand. So, we’re seeing a great opportunity to augment and integrate ChatGPT with Constructor’s product discovery platform and personalization technology—so that the responses shoppers receive don’t just make sense but they’re also highly personalized and reflective of retailers’ KPIs. We have a beta product ready in this space that a number of our retail customers are integrating with and testing. You can see a video with examples of how it’s used below.
And then, as I mentioned, on the ecommerce merchandising side, we’re also doing a lot of cool and useful things—giving merchandisers tools that free them up for more strategic work. Many of the merchandisers who use our platform previously spent lots of cycles on repetitive and manual work: Researching and specifying synonyms for search terms, correcting bad results, creating fixes for typos, manually implementing page redirects, and so on.
Constructor’s AI automates many of these tasks so that merchandisers can focus cycles elsewhere. We’ve heard from customers like Birkenstock that with Constructor, they’ve been able to reduce manual work by 20%. Now, merchandisers have greater flexibility to experiment with new campaigns and ideas.
They can also use our platform to get feedback: “How will this change conversion rates?” “What are the tradeoffs for implementing these kinds of campaigns?” These capabilities didn’t really exist for merchandisers before. Still, in my opinion, they’re critical to the future of ecommerce and to making sure companies are creating delightful experiences for their shoppers that also work for their bottom line.
What are three traits about yourself that you feel helped fuel your success?
Honestly, I think I have a lot to learn, and a lot of what got me to where I am is luck. I wish I could take credit for it, but I think the component of luck is true for most business leaders. We work incredibly hard, of course—but there’s that essential and intangible element of being at the right place at the right time and meeting the right people, to truly help your business take off.
So for me, I was incredibly lucky to meet Dan McCormick back when I worked at Shutterstock. At the time, he was the CTO there and my boss’s boss’s boss. Dan took a big risk launching a startup with me; he and I co-founded Constructor.
I was also very lucky to meet incredibly kind and visionary people who were willing to trust me in Constructor’s early days. Back then, customers like Sephora and Bonobos took a chance to work with us as an early-stage start-up, and we were very grateful they did. They certainly didn’t have to do that, and business leaders at those companies likely put their jobs on the line by taking a chance on us. Those people know who they are, and they’ll always have a special place in my heart. We wouldn’t have a business without them.
Humility is another trait that I think is important. There will always be things you don’t know, and it’s important to be comfortable with that. It’s OK to say, “I don’t know the answer, but I’m going to find someone who can help. I’m going to read a lot and look into it.” That’s how you learn and get better.
And finally, innate curiosity is often an asset. I find lots of different areas interesting, and I think that’s helped me in my work. When we needed to figure out things for the business that I didn’t know, it was an excuse for me to learn something new, which I found very exciting.
What was the original vision for your ecommerce business? What pain point(s) were you trying to solve for your customers?
So, as I mentioned earlier, my first experience with product discovery was when I was a search engineer at Shutterstock. I very much enjoyed the work, but at the same time, I wondered: Did it matter? Was I working on a “science project” or something that made a difference?
I saw through A/B tests that our efforts truly did move the needle, improving both the customer experience and the bottom line. So, I wanted to effect that kind of change on a larger scale for many more ecommerce companies—helping them better turn search results into revenue.
And I wanted to do it with search technology that was tailor-made for ecommerce. The open-source search and product discovery technology we used at Shutterstock (used by many other ecommerce companies at the time) was limited—it wasn’t specific to ecommerce and wasn’t oriented to conversion optimization.
So when Dan and I founded Constructor, we wanted to build an ecommerce product search and discovery platform once. We wanted to build it well—essentially to build the system of our dreams that we knew we would never have been able to build at an individual ecommerce company because of how risky, time-intensive, and capital-intensive it would have been. We reasoned that if we did it right and if we could drive the results we were thinking about, no ecommerce company would ever need to build this thing again.
Plus, building a successful product discovery platform internally isn’t easy or feasible for most companies. Their own teams are often resource-constrained, which ends up driving a lot of decisions. At Shutterstock, when we undertook search and discovery projects, we’d be extremely proud of our results—but when we compared ourselves to Amazon, which has many more resources, it was like we’d invented arithmetic when they were already doing calculus.
So with Constructor, we also wanted to democratize product search and discovery and “level the playing field”—because more competition and better customer experiences are good for retailers and shoppers alike.
How do you see the ecommerce industry evolving in the coming years?
The way people search for products online—relying on keywords, not being able to use complete sentences to explain their needs, hoping for the best—and the way they discover them haven’t changed much in 30+ years. But as consumers continue to expect better digital experiences—and as automation, transformers (the technology ChatGPT is built on and that we’ve started to use at Constructor over the last couple of years), and other AI technologies pick up steam—we’re going to see rapid, transformative changes across ecommerce (no pun intended).
Retailers will have more flexible and better ways to showcase their inventory online, and shoppers will have new ways to find the right products and interact with search functionality. In addition, retailers’ ecommerce operations will become more tightly intertwined with, and more interdependent on, their in-store ones, so the traditional separation between brick-and-mortar and digital will be much more indistinct.
Not every change will be for the better. As we’ve seen recently with new, hot technology like ChatGPT, there’s often a rush to do anything to capitalize on the buzz. But creating something AI-based for the sake of using AI is a miss. Ecommerce companies need to look at: What problem or pain point are we trying to solve? How can we use ChatGPT, other AI, or non-AI-based technology to offer a better way? When you can blend what’s new and cool with what’s genuinely useful—creating something that shoppers will use repeatedly and that drives value for the business—you’ve set yourself up for success.
How do you balance the need for innovation and experimentation with maintaining a stable, reliable ecommerce infrastructure?
That’s a really good question, and it requires a well-thought-out and balanced approach. You have to know where it’s ok to experiment and where it’s best not to do anything risky.
As an example, when it comes to our core services at Constructor, we would never put out anything that wasn’t thoroughly tested. Any updates or enhancements go through a battery of unit and stability tests: running multiple days of previous traffic against the new code to make sure it holds up. We also require multiple people to thoroughly examine new code prior to production. In short, we’re very careful, and it’s why we never have downtime, and the core system has never gone down in the history of the company.
As for driving innovation, we often partner with customers who want to be on the leading edge and incorporate the hottest, newest tech out there. So, we strive to develop useful and innovative ideas and solutions, and we’re clear about what’s still an experiment.
Other clients of ours take a more measured approach, though. Rather than being the first to try something out, they’ll wait until more ecommerce companies who work with us have gone live and seen measurable growth, and then they’ll ask to try it. Both approaches have merit and underscore the different ways ecommerce companies operate their businesses.
What five emerging trends do you believe will have the biggest impact on the future of online shopping?
1. Totally new forms of product discovery. Honestly, this is the thing I’m most excited about. And it’s not just pie in the sky. The way people discover products when they’re shopping online is going to expand and change. Consumers won’t be limited to searching, browsing, and viewing recommendations. There will be completely new forms of ecommerce product discovery that don’t exist yet.
We’re starting to see what this future could look like with technologies such as ChatGPT. When people search for products today, they have to do it in a way that search engines can understand: with succinct search terms, speaking almost like a caveperson. Sometimes, that’s all shoppers need to do to find what they need.
But for the times when you need to search in a longer, involved, or more open-ended way—like “What ingredients do I need for a lemon meringue pie?” or “I live in Florida and am taking up jogging, and want practical but stylish running clothes”—ChatGPT can help. By training transformers (the “T” in ChatGPT) on ecommerce data—and integrating personalization technology—there’s an opportunity to create transformative and useful product discovery experiences.
But this is still just scratching the surface of what’s possible. Ecommerce companies recognize that search and browse functionalities are useful but imperfect: So, the future will be search and browse, along with other ways of product discovery, to more fully address shoppers’ far-reaching needs.
2. Displaying search results in fresh, new ways. When it comes to displaying search results, we’re still stuck in the 1990s. There hasn’t been a lot of investment in experimentation with the user interface (UI), and we’re long overdue for a change.
What I mean by that is: When you search for a product in a search engine, the results are typically spit back in a grid that’s a few columns wide, with multiple rows. Is this because it’s the optimal way to display products? No—but it is an easy way to display information pulled from a database. Disrupting this status quo will happen, and it will pay off for those companies who do it well because they’ll be able to give their shoppers a better experience and they’ll be able to better define their brand identity.
This is something they’ve already mastered in-store. When you search in a retailer’s brick-and-mortar location, the effort they’ve put into curation and displaying their products is evident: from what goes on the mannequin, where the mannequin is placed, how displays are positioned, what associates are wearing, and so on. These details set retailers apart from each other and give you a feel for their brand. So, when it comes to online product discovery, it’s possible to get creative like this too.
Think about the difference between an Apple store and a Walmart and a high-end clothing store. It’s a completely different experience walking into those three places. Now go to each of their websites and search for something. All of the UIs and displays are basically the same. There’s so much opportunity to differentiate in the UI, and I think most people aren’t thinking about it.
3. Empowerment of online merchandisers. As I mentioned, brick-and-mortar stores today are really good at curation—to the point that shoppers often don’t realize the effort that goes into making displays look as good as they do. These stores and their associates need the right tools to order and display products in a way that’s inviting and attractive to shoppers. In an online setting, retailers also need the right tools to showcase their inventory and product curation: making the way they display products a reflection of their brand. This gets a bit more complicated online, where you have the added and necessary dimension of being able to personalize for each visitor.
So, the number of tools available to online merchandisers, and the sophistication of those tools, will increase. And even though many tools will be AI-based, they won’t replace human merchandisers, who largely define the brand’s identity, which is as much art as it is science. Those merchandisers need to make strategic decisions about the best way to convey brand identity. Completely turning over everything to AI would result in too many identical online storefronts, so human decision-making elements will always be important. Humans know best how to make delightful experiences for other humans.
What’s more, AI tools that effectively empower merchandisers can’t be “black-box” solutions. It’s important that underlying algorithms be transparent to merchandisers, so they can understand what the system is doing and that they can override the AI’s recommendations when they deem it appropriate.
4. More useful personalization: “A digital home.” In its early days, personalization was often gimmicky, annoying, or intrusive—like an ad stalking you across the web, showing a product you checked out once on an ecommerce site. We’re starting to see the beginnings of better, more useful personalization on retail sites—in part due to better and more judicious use of customer data and less use of third-party data to drive personalization.
Think about Spotify and Netflix as examples of what’s possible: They use what customers tell them directly (called “zero-party data” because it’s data the customer actively wants the site to know about them) to power individualized experiences. And users want to share their data with them; I’ll go out of my way to “like” songs that I enjoy on Spotify because I want them to give me similar songs I’ll enjoy. For consumers, the value of sharing data and engaging in this way is clear, and they see immediate benefits.
Personalization in retail will follow this path, too, relying more on explicit information that shoppers give willingly and showing the value of this trade-off. This is what Constructor does with our Quizzes product—immediately providing personalized recommendations based on how shoppers answer brief surveys. Using zero-party data in this way—alongside first-party behavioral or clickstream data—will yield highly relevant, highly personalized experiences that respect shoppers’ privacy.
Spotify and Netflix also excel at creating a “digital home” for users: an account or destination that’s familiar, easy to navigate, and tailored and personalized to them. For retailers to successfully compete with the likes of Amazon, they’ll need to create this type of personalized digital home, too, enabling online shoppers to feel like they’ve entered their favorite store or favorite bar, where the owner knows them and is attuned to their needs.
To make shoppers feel more at home, personalization must continue its journey: going from gimmicky and intrusive to empowered and empathetic. We’re just seeing the beginnings of this progression, and the companies that figure it out the fastest and best will have the most loyal users.
5. The continued melding of digital and brick-and-mortar experiences. These domains had traditionally been totally separate: separate teams, siloed data, and different experiences. But to consumers, it’s all the same brand, right? Separate experiences can cause confusion and frustration, while the convenience of omnichannel experiences is good for shoppers and retailers alike, and it’s another place to continue that connected, personalized shopping journey from one touchpoint to the next.
So, expect to see a continued blurring of the line between digital and physical stores—with more “buy online, pick up in-store” options, the use of mobile apps and beacon technology to help you locate products in-store, and real-time personalization driven by cross-channel data. For retailers, campaigns on the digital side will help campaigns on the brick-and-mortar side and vice versa.
When it comes to consistently providing holistic and connected customer journeys, are we there yet, as an industry? Not even close. But it’s the way of the future: one that’s both technologically possible and will create a far better customer experience. Retailers who get there first will reap the rewards.
Is there a past trend that’s now common practice in ecommerce that you would have spent 50% more time focusing on? Which one and why?
A year or so ago, if you tried to promote or implement new types or forms of product discovery, it was really difficult—very much an uphill battle. It would take a lot of time and convincing to get people to try out new ideas. This was something we did spend time on, but less than I would have liked, simply because the appetite wasn’t there in the market.
But that culture and reluctance seemed to change overnight when ChatGPT came on the scene. It opened everyone’s eyes to the fact that there will be new forms of product discovery—and significantly increased companies’ desire to try and pioneer them. So, what used to be a challenge has become a much easier problem to address, as brands now approach us proactively to ask if we have anything new around ChatGPT.
And we do have lots of ideas—layering personalization and product catalog information on top of generative AI, as I mentioned. We’re also working on using LLMs and transformers to provide better product retrieval and filter out irrelevant search results. This work may have an even greater impact on customers’ businesses than the “conversational commerce” aspects of ChatGPT; we’ll see as we continue to experiment and test.
So, on the one hand, we have a great opportunity that I’m thankful for. But the opportunity also comes with an obligation to develop products responsibly and add value. We’re in the middle of a hype cycle now because customer excitement creates new opportunities. Unfortunately, a bunch of companies jump into the mix with “snake oil.” And that snake oil is what turns a hype cycle into a trough of disappointment.
That’s why we’re very clear when we’re working on something experimental. We don’t know if it’s going to be the best thing ever, but we’re excited to find out. And when customers want to get on board to beta test some of these innovations, we love that. They know we’re committed to balancing excitement with humility and doing right by them and their shoppers.
Looking ahead, what are the biggest opportunities and challenges facing the ecommerce industry, and how do you plan to address them in the coming years?
There are many opportunities to improve and expand the way shoppers interact with ecommerce sites. As I’ve mentioned, with so much functionality still stuck using concepts invented in the 1990s, it’s time for a change. And as technology continues to advance rapidly, we’re on the precipice of that change. It’s an exciting time and an opportunity to improve customer experiences and optimize results for ecommerce companies: a win-win.
The challenge nowadays, across all facets of ecommerce, is combining what’s possible with what’s ethical and drives true utility. I’m a firm believer in the ethical and responsible use of AI and in using customer data in ways that respect privacy and deliver value.
Plus, it’s easy to get dazzled by the next “shiny new thing” in the world of technology today. But when companies rush to incorporate a new technology because it’s cool—or simply so they can claim they’re using it—they often create something consumers will try once or twice, then abandon. We’ve seen, and we’re going to continue to see, a lot of gimmicks fizzle out.
In particular, we’re still in the early days of what’s possible with generative AI. At Constructor, we’re truly excited about our experiments in this space. As we embark on new projects and activities, we remain rooted in a commitment to utility: it’s at the heart of what Constructor does.
Among our guiding principles is the promise to ask ourselves: Is what we’re creating beneficial for the shopper and retailer? So, as we forge ahead, we’re committed to using technology to create the best possible experiences for shoppers and retailers alike.
If you could start a movement that would bring the most good to the most people, what would that be?
This is a hard one. I spend a lot of my time in the business world. There are far smarter people than me thinking about the more important problems like world hunger, so I hope it’s okay to stick to the business world for my answer, just because I think I can help more there. More organizations should be committed to creating genuine partnerships with the people they do business with. There should be an expectation that trust needs to be earned, and it’s the vendor’s responsibility to go earn that trust.
When you look at the way traditional vendors are set up, there’s a salesperson who sells you a product—then once you buy it, you’re passed off to someone else. You’re not talking to the same person you bought the product from initially. And that person, at any rate, wasn’t involved in building the product, so there’s a limit to how much they can help you. But there would be more trusting and happier people in the business world if there were a more pervasive culture of accountability, better and more transparent handoffs and a commitment to driving true partnerships.
There is a ton of time and money spent on evaluating vendors today, and for a good reason—so many of them have broken their customers’ trust that those same customers know they have to evaluate vendors thoroughly. I just think it would be a much better world if more businesses said, “I want to earn your trust before you pay me,” and then really prioritized building that spark of trust—not just into maintaining a customer, but going above and beyond and building a real partnership.
How can our readers further follow your work online?
Don't miss out on the latest tips, insights, software recommendations, and expert advice from The Ecomm Manager. Subscribe to our newsletter today!
More great ECM content:
- How to Use AI in Ecommerce: Tools, Tips, and Strategies
- Ecommerce + Retail: A Guide To Omnichannel Commerce For Businesses
- Why Ecommerce Personalization Matters: 5 Tactics (With Examples)
- What Is An Ecommerce Chatbot: Types, Ways To Use & Benefits For Your Online Store
- How To Collect Customer Data
- 8 Ecommerce Trends Driving The Future Of Online Shopping in 2023
- 8 Ecommerce Personalization Trends for 2023 and Beyond