Next up in our Smart Kitchen Summit speaker preview series is Spencer Price, the founder of Halla.
Halla has built an AI personalization and recommendation platform for grocery store providers. According to Price, the turning point for his company and the broader grocery store industry was when Amazon acquired Whole Foods.
“When Amazon acquired Whole Foods in 2017, it sent grocers into this innovation frenzy,” said Price. “I think the main driving force for grocers to want to look at this type of tech back then was that Amazon generates over one-third of all of its product sales revenue from their recommendation with the ‘You may also like’ and ‘customers also bought’ type product suggestions. Grocers do not have a passive piece of AI that drives a third of their sales, and that is what we enable grocers to do. We give them that competitive weapon to fight back in this World War grocery.”
Price thinks that while the grocery industry is lagging behind other industries, such as entertainment, when it comes to personalization, they are looking to AI to make up ground.
“Netflix isn’t just on a one-account basis. Within an account, you have a handful of profiles in your household, and each profile sees a completely different set of suggested categories, titles within those categories, and even different cover art for each one of those titles that’s likely to resonate with you as a specific end user.
“Grocers are a little bit behind these content platforms, but I think in 10 years time, we will see a very similar thing, and it’s going to be even more exciting because if you can give every single shopper their own unique grocery store, that’s going to make for both the fastest and most efficient and of course, most inspiring shopping experience. Grocers want to move quickly.”
Price’s company was acquired by Wynshop in March. Price says the company brought over his entire team and that Halla remains an independent business unit within Wynshop.
You can hear Price speak at Smart Kitchen Summit on June 4-5th in Seattle. Get your ticket today!
You can read the full transcript of our conversation below:
Michael Wolf: All right, I have with me Spencer Price, the CEO of Halla, now a part of Wynshop. It’s been a while since we caught up. We wrote a first article about you guys since then, and you guys have changed a lot since then.
Spencer Price: We have changed a lot since then, yes.
Michael Wolf: At that point, you were very much focused on being a personalized recommendation platform based on a lot of different data. I still think that’s a lot of what’s pretty true, but you guys did evolve since then.
Spencer Price: Yeah, so 2018 was a transitional period. We had developed, as you said, a personalized recommendation engine centered on food and beverage products. And we had a mobile app that would recommend restaurants and even specific dishes from those menus to users or groups of users with varying taste preferences, dietary restrictions, et cetera. And that was 2017 to 2018.
When Amazon acquired Whole Foods in 2017, it sent grocers into this innovation frenzy. There was a demand for us to gut the tech from the app, license it B2B, and we ended up sunsetting the mobile app, which feels like a lifetime ago now. And all we’ve done is deploy personalized recommendations, search and substitutions for online grocers ever since.
Michael Wolf: I didn’t know that that had such a big impact. It makes sense, in retrospect, the acquisition of Whole Foods by Amazon. But like you said, there was this frenzy and a wake up call to existing grocers, and that sent you in a completely different direction.
Spencer Price: Exactly. We had some innovative nimble online grocers as well as some legacy retailers that knew they needed to step up. I think the main driving force for grocers to want to look at this type of tech back then was that Amazon generates over one-third of all of its product sales revenue from their recommendation with the ‘You may also like’ and ‘customers also bought’ type product suggestions. Grocers do not have a passive piece of AI that drives a third of their sales, and that is what we enable grocers to do. We give them that competitive weapon to fight back in this World War grocery.
Michael Wolf: I love that; World War grocery sounds like a movie, starring you guys apparently. But I mean, look at the last 18 months, right? I think the world’s woken up to AI. It’s permeated all the press and the pop or consciousness largely due to the exposure of things like ChatGPT and generative AI to everyone. It seemed like like six, seven years ago, a lot of people were building ontologies and had a custom code, to make their AI to get certain outputs. But now, with generative AI, you can basically do prompts and get a lot of the same results. And these large language models just keep getting bigger. Can you talk about how your business has changed by incorporating larger langue models and generative AI?
Spencer Price: Yeah, so the way that generative AI has taken shape thus far has, of course, been through chat bots. One of the things that those, at least from a consumer-facing standpoint, one of the things that chat bot ask technology with LLMs, Gen. AI, et cetera, plays into e-commerce at large and potentially grocery down the line is conversational commerce.
We don’t see that as being a particularly exciting use case, particularly in this category where people are adding usually a couple dozen items to their basket. They’re not saying, you know, I need help finding the right sweater that matches these pants. It’s a household you’re shopping for with different dietary restrictions, taste preferences. And that’s where language models don’t necessarily perform best. That’s where recommender systems have decades of tried and true proven methods.
And so that’s still a foundational component of our science. However, for one of our solutions, search, LLMs allow for a much more robust level of understanding natural language. So we had our own raw sort of NLP models that we developed in -house a few years ago, that we’ve been fine tuning, and now we can incorporate some of these open source transformers and LLMs to take our vertical eyes, rather than a generalist sort of assistant, our vertical eyes recommender systems and layer them with this cutting edge technology that allows for the generation of synonym lists and a better understanding of things like typos. But the risk with using just generative AI to try to develop these highly specialized models in a category that’s clearly so nuanced and personal is the hallucinations. I was recommended a beef and banana soup from chat GPT. And I got to tell you that that feels a ways away. I did not. It was terrifying to be honest.
Michael Wolf: Did you make it?
Spencer Price: I did not. It was terrifying, to be honest.
Michael Wolf: Well, I’ve been talking to a lot of folks who are in this area of food and beverage that are trying to deploy AI centric solutions. And like you said, a lot of the LLMs have this problem with hallucination. They’re oftentimes, they’re ingesting the world of the broader internet, but they don’t necessarily go deep on things like food and beverage. So I’ve heard companies that are building special, small language models that can couple into large language models. They’re doing kind of these transformers that provide the intelligence. Sounds like you guys have your own kind of approach to that. And you’re using LLMs as the conversational smart interface that is just so much more savvy than it would have been in the past. And then diving deep into your knowledge set.
Spencer Price: Precisely. We are using these new state of the art technologies, both as sort of a research platform to understand what we can benefit from and leverage and also where the watch outs are, like the example I just shared. One thing that you’d imagine might be really nice, whether it’s with a small language model specific to what we’re doing, or using the best of these large language models.
One use case that probably strikes you as obvious is groceries have a notoriously dirty data problem. And so maybe there’s a way to clean up these product catalogs and inventories and descriptions and attributes. The challenge is you can’t run the risk of things like health claims, nutrition facts, or marketing descriptions being completely wrong. And we’ve seen a lot of inaccuracies in using it for that.
So everything we do with LLMs has a human in the loop to make sure that none of those inaccuracies end up facing a user. But by and large, what sets us apart is layering in, as you said, our knowledge base, which is an ontology of every single product, but more than that, the essence of each product, knowing that orange, for example, is a distinct flavor. It is a product and it’s also a color. And LLMs are not built to have those nuances at play to the level of sophistication that you need them to be. Does that make sense?
Michael Wolf: Yeah, it does. What are you most excited about if, 10 years down the road, you’re building systems that use technology like AI in terms of the grocery shopping experience? What do you think will change the most?
Spencer Price: So I think that personalization historically took a lot of different shapes, and then they all kind of converged five to 10 years ago by having truly individualized browsing experiences on content platforms. Netflix isn’t just on an account basis, but within an account, you have a handful of profiles in your household, and each profile sees a completely different set of suggested categories, titles within those categories, and even different cover art for each one of those titles that’s likely to resonate with you as a specific end user. Spotify acquired Echonest, and they were able to map out all the different attributes down to subjective metrics like the danceability of every single track in their library, now they have the most robust music recommendation engine in the world, and people love them for that, and I’ve never left as a result.
In online shopping, we’re talking about products now, not content. We’re a little bit behind these content platforms, but I think in 10 years’ time, we will see a very similar thing, and it’s going to be even more exciting because if you can give every single shopper their own unique grocery store, that’s going to make for both the fastest and most efficient and of course, most inspiring shopping experience. And we’re not there yet, but we have all the rails to get there in a lot less than 10 years. Depends how much. Grocers want to move quickly.
Michael Wolf: That’s exciting, getting Mike’s grocery store tailored towards me. That’s perfect. Tell us about the Wynshop deal. You guys got acquired, which is exciting news for you. What does that mean?
Spencer Price: So our biggest channel partner to reach retailers and have our personalization technology directly embedded into an e-commerce platform was with Wynshop. And they’re the leading provider of e com platform technology on a white label basis to grocers all over the continent and a handful of international accounts as well. And we’ve been working with them for a few years. We love the team. We think they have a clearly differentiated product and they got to know us, our team and our tech. And it was just a pretty perfect match, to be honest, to have what we’ve developed baked in as more of a base level set of functionality, as well as being able to offer premium levels of functionality for these grocers that they can opt into if they want.
So yeah, about six weeks ago, we joined the team. They brought on all the day to day, all the personnel, we remain an independent business unit within Wynshop, but obviously it’s not like there’s any walls up. We work with everybody there very well. They put some resources behind us and yeah, the goal is both to service their existing accounts and future customers as well with the tech we’ve built and the new stuff we’re building.
Michael Wolf: All right, well, Spencer, congratulations. You worked hard for years to build the product and then create a opportunity for you. So I’m looking forward to talking more with you in Seattle in June at the Smart Kitchen Summit. How can people find out more about Halla and Wynshop?
Spencer Price: Yeah, well, thank you so much for the opportunity and the congratulations. You can still find us even though we don’t go by holla .io, we’re just holla now, at halla .io and winshop .com, W -Y -N, shop.
Michael Wolf: Cool. Hey, well, Spencer, thanks so much for spending time with me, man.
Spencer Price: Thank you so much, Mike. Look forward to seeing you in June.
