Retailgentic | Consumer Behavior & Retail Trend

This one is a build deep dive. Daniel has spent eight years on the same problem, and the answer has changed shape three times along the way: a consumer app in 2018, a shoppable recipe platform in 2023, and now a conversational shopping assistant running inside grocersโ€™ own sites. 
We go deep on the plumbing here: product catalogs, ontologies, eval systems, model selection, retail media, and what an agent does when a shopper types โ€œtaco night.โ€ If you like knowing how these things actually work under the hood, this is your episode.
Never miss the latest Retailgentic newsโ†’

Highlights:
  • ๐—ง๐—ต๐—ฒ ๐—ป๐˜‚๐—บ๐—ฏ๐—ฒ๐—ฟ๐˜€: Live in over 700 stores, reaching over 10 million shoppers, with the number of locations set to double before the end of the year. The consumer app has over a million users. 
  • ๐Ÿฐ๐Ÿญ ๐—ฑ๐—ฎ๐˜†๐˜€ ๐—ฐ๐—ผ๐—ป๐˜๐—ฟ๐—ฎ๐—ฐ๐˜ ๐˜๐—ผ ๐—น๐—ถ๐˜ƒ๐—ฒ: A one-line JavaScript install that reuses the retailerโ€™s existing catalog, add-to-cart, analytics, and retail media. National scale in six weeks.
  • ๐—ฆ๐—ฒ๐—ฎ๐—ฟ๐—ฐ๐—ต ๐—ฎ๐—ป๐—ฑ ๐˜€๐—ฒ๐—น๐—ฒ๐—ฐ๐˜ โ†’ ๐—ฟ๐—ฒ๐˜ƒ๐—ถ๐—ฒ๐˜„ ๐—ฎ๐—ป๐—ฑ ๐—ฎ๐—ฐ๐—ฐ๐—ฒ๐—ฝ๐˜: Danielโ€™s framing for the whole shift. The old era: you search one thing, you select it. The new one: the agent assembles a solution and you approve it. 
  • ๐—ง๐—ต๐—ฒ ๐—ฒ๐˜ƒ๐—ฎ๐—น ๐˜€๐˜†๐˜€๐˜๐—ฒ๐—บ: Thousands of test scenarios with grading criteria aligned to each retailerโ€™s merchandising principles, run tens or hundreds of times to get a distribution of responses. 
  • ๐—ง๐—ต๐—ฒ ๐—ณ๐—ถ๐˜€๐—ต ๐—ฟ๐˜‚๐—น๐—ฒ: Retailers write procedures the agent pulls when a scenario matches. One of Danielโ€™s favorites: if youโ€™re cooking fish today or tomorrow, it recommends fresh. If itโ€™s later in the week, frozen.
  • ๐—ง๐—ฟ๐—ฎ๐—ถ๐—ป๐—ถ๐—ป๐—ด ๐—ฎ ๐—ป๐—ฒ๐˜„ ๐—ฑ๐—ถ๐—ด๐—ถ๐˜๐—ฎ๐—น ๐—ฒ๐—บ๐—ฝ๐—น๐—ผ๐˜†๐—ฒ๐—ฒ: Onboarding is essentially an interview with the business. What are your values? How do you respond? What are your merchandising principles? That becomes the agentโ€™s identity and personality. 
  • ๐Ÿญ,๐Ÿฌ๐Ÿฌ๐Ÿฌ ๐—™๐—ผ๐˜‚๐—ฟ๐˜๐—ต ๐—ผ๐—ณ ๐—๐˜‚๐—น๐˜† ๐—ฏ๐˜‚๐—ป๐—ฑ๐—น๐—ฒ๐˜€: Instead of one holiday carousel every shopper sees, merchant teams define ~20 axes โ€” price, dietary needs, brand preference โ€” and an agentic process generates permutations personalized against transaction history.
  • ๐—”๐—œ ๐—–๐—ต๐—ฒ๐—ณ๐˜€ ๐—ผ๐—ป ๐—ง๐—ถ๐—ธ๐—ง๐—ผ๐—ธ: Long before ChatGPT, Cooklist ran a TikTok series generating recipes with GPT-2 and early GPT-3. The models got good at basic recipes so fast that the fun moved to challenging them to be creative.
Daniel is one of those founders who has clearly built the thing himself, and it shows in the answers. We learned about a thousand things from this one. Enjoy the conversation.

Timestamps:
02:57 Welcome Daniel Vitiello, CEO & co-founder of Cooklist
03:10 Grocery, CPG, and retailer-hosted agents: the pod's favorite topics
04:46 The $1 check and the handwritten letter for loyalty data
05:46 Why 2017 was the moment
06:12 Inside the Cooklist app: home inventory and the AI chef
07:35 AI as the thread: Kurzweil, GPT-2, and the AI Chefs TikTok series
09:37 B2C to B2B, and the synergies of running both
12:05 The cat food problem
13:34 The core identity system: hummus, three ways
15:58 Search and select vs. review and accept
16:23 The spark for B2B: Target, Techstars, and saying no
18:45 Recipe commerce and meal-pairing merchandising
20:19 Funding: RevTech, Mercury Fund, and profitability
21:56 Freemium, pro, and advertising in the app
22:48 700 stores, 10 million shoppers
24:28 The flagship deployment: live in 41 days
25:56 Ontology, context engineering, and interviewing the retailer
27:03 Non-determinism and the eval system
29:06 Entry points: sparkles, suggested prompts, and PDP workflows
31:01 How the catalog and inventory get in
33:41 Procedures: the weekly flyer and the fish rule
35:55 Cost, speed, and the token-efficient stream
37:38 Model agnostic: scoring Opus against Sonnet
38:15 Instacart, Storefront Pro, and winning head-to-head
39:47 Retail media and Criteo
40:22 Agentic search and the "taco night" problem
42:02 Agentic merchandising: 1,000 Fourth of July bundles
42:55 Proactive carts, and the 50% who finish in the store
43:43 Smart fridges and forecasting demand
44:44 Where to find Daniel

๐Ÿ‘‰ Connect with Daniel: https://www.linkedin.com/in/dkvitiello/
๐Ÿ‘‰ Learn more about Cooklist: https://cooklist.com/

๐Ÿง  Want to stay ahead in AI commerce? Subscribe and follow along:
๐Ÿ“ฐ Subscribe to the free Substack: retailgentic.ai
๐Ÿ“บ Watch episodes on YouTube & Subscribe for updates: https://www.youtube.com/@Retailgentic

Creators and Guests

Host
Scot Wingo
CEO and Co-founder, ReFiBuy.ai

What is Retailgentic | Consumer Behavior & Retail Trend?

Retailgentic, presented by the ReFiBuy Media Network and hosted by Scot Wingo, is where Agentic Commerce meets retail innovation. We help retailers and brands prepare for the future of agent-driven shopping with actionable news, expert insights, dxis, forward-looking predictions, and practical guidance on Agentic Commerce Optimization (ACO).

Speaker 1:

We're coming in as a technology partner to a grocer to allow them to have an experience that is at least on par, or if not in some ways better, than what Amazon and Walmart are doing, and do it extremely quickly. We build both a front end kind of a little JavaScript app that can be inserted into an existing platform. So it's kind of like a one line install that reuses most of the retailer's existing infrastructure for their product catalog, the add to cart capabilities, you know, tying that in with analytics or retail media.

Speaker 2:

Welcome to the Retailgentic podcast, where we explore the intersection of retail, ecommerce, and AI agents, which we call Retailgentic. Our mission is simple. We are shining a bright light on the new agentic AI shopping megatrend which will disrupt the trillion dollar digital retail market. The stakes are high. Google, OpenAI, Microsoft, Perplexity, Amazon, Apple, Meta, and Anthropic are all fighting to be the Consumer AI Shopping agent winner.

Speaker 2:

There are gonna be some tough decisions for online brands and retailers to make. What happens to checkout? How should payments work? Could you continue to invest in retail media networks? What does the world look like when over half of your traffic is not from humans but from AI agents?

Speaker 2:

These existential questions are the tip of the iceberg. On this podcast, we're talking to industry experts, analysts, and influencers to get answers to these questions, identify trends, and best practices. This is Retailgentic, your podcast partner in navigating our AI shopping agent future. And now here's your host, Wingo.

Speaker 3:

Welcome to this episode of Retailgentic. My guest today is Daniel Vitiello. He is CEO and cofounder of Cooklist. Cooklist's journey is a very interesting entrepreneurial story that starts with a b to c consumer app in 2018 and ends here in 2026 with a B2B Conversational Agent custom made for grocers. It is currently available in 700 stores and available to 10,000,000 consumers, which is amazing story.

Speaker 3:

We'll talk about that journey, how the whole system works. And then in typical Retailgentic fashion, we're gonna go very deep on this. We'll talk about product catalogs, how CPGs tie into this, what's the Instacart story on this, all the things. And then finally, Daniel is gonna give us some hints into his vision for where he sees Cooklist going. This is my conversation with Daniel Vitiello, CEO and cofounder of Cooklist.

Speaker 3:

I hope you enjoy it, and happy Agentic conversing. Welcome to this episode of Retailgentic. Today, I am joined by CEO and co founder of Cooklist, Daniel Vitiello. Daniel, welcome to the show.

Speaker 1:

Thanks for having me, Scot. Excited to chat today.

Speaker 3:

Yeah. Yeah. I was just, as I was telling you in our, in the green room, one of our most popular topics, is our two most popular topics on the pod are anything around grocery and CPG. So, so yeah, spoiler alert, you're going to check that one for the audience that has no idea what you do, but it's right there in the name Cook. So, and then number two, anything about retailer hosted agents is always interesting.

Speaker 3:

So the big guys are obviously, Amazon, the artist previously known as Rufus, which is now Alexa for shopping, the much more worry answer, and then, Sparky on Walmart. So, you guys are in both buckets. So this is gonna be a fun conversation. Appreciate you coming on the pod.

Speaker 1:

Yeah. Yeah. Thanks. There's there's a lot of cool stuff happening in the space and excited to get into it.

Speaker 3:

Let's start with a little bit of background. You're coming to us live from Austin, Texas, where it's almost as hot as it is here in North Carolina. How did you get into the startup journey that led you to be an entrepreneur?

Speaker 1:

Well, I've been an entrepreneur since I was a little kid, selling popcorn door to door with Boy Scouts and making various crafts and selling them out in front of my house. Had a lawn and landscape business in high school. Hired a bunch of my friends to work at that. And then have really been an entrepreneur my whole life. Last company before Cooklist, we were manufacturing coffee grinders.

Speaker 1:

And so we were in a physical product business, bootstrapped. And then, the origin of Cooklist, though, actually was even before the coffee grinder. It really be... Came from this, interest in insights that can come from a lot of data. And I landed on, for some reason, loyalty card data is...

Speaker 1:

You know, I knew my parents used it every week when they went to the grocery store and had heard some things about how the retailers and the food brands were able to create a lot of value from that data. And that kind of piqued my curiosity. And in the time, this was back in probably 2015, I actually called the number on the grocery store loyalty card and asked them if I could get a copy of all of my purchases. So it was kind of before there was online ordering with Walmart and Kroger and all the big retailers. And they told me that I had to mail a $1 check and write them a handwritten letter asking for my transaction history.

Speaker 1:

And then sure enough, a few weeks later, they mailed it to me and realized that was probably a little bit too early to be building a business around getting data through the mail. But around 2017, following the Amazon acquisition of Whole Foods and all the big retailers moving for online ordering for pickup and delivery, That's when we realized the timing was right to build a business around this new e commerce space that's happening with all the grocery assortment available in a digital format, as well as your transaction history is available digitally. And so that was the origin of Cooklist.

Speaker 3:

Are you... Do you have a technical background, or you're kind of the business guy and you bring in the the tech talent to fulfill your vision?

Speaker 1:

Yeah. So I I lead all of our technical teams. And really, it was just me kind of coding away for the the first couple years, building out the initial platform, the integration with a lot of the retailers to be able to link your loyalty cards. For anyone who doesn't know, the core feature and kind of niche of the Cooklist app is a home inventory system. And so you connect all the grocery stores you shop at.

Speaker 1:

It downloads your transactions into a digital inventory and then can show you recipes you can cook with the groceries you buy. We have an AI Chef in there now that you can chat to, and it has that context of everything you have in your kitchen, as well as help you place an order for a meal plan you've put together in the app. And it kind of does that reconciliation of here's what you need for the recipes you want to make, and then here's what you currently already have in your home.

Speaker 3:

Yeah, so let's walk through that history. So you started in 2018 as an app, so kind of on the App Store type of an app, right? And then what we're going to talk to today is kind of a different delivery of the technology. So walk through the transition. And I assume, maybe it's not even a transition, but the things you've added as you've kind of built along.

Speaker 3:

I think that'd be a cool next thing to hear about.

Speaker 1:

The core theme that's the thread that connects everything is AI. So you know, starting back even, like, reading some of Ray Kurzweil's early books, talking about, you know, following Moore's Law and seeing what's gonna be happening around the mid twenty twenties. You know, he was talking about in the early nineteen nineties. And so I I spent a lot of time learning AI and thought that that was a fundamental skill that could then be applied to anything else. And in the early days with Cooklist, we were using AI to turn unstructured data, so like the universe of product content and recipe content, using AI to classify it, bring structure to it, and then to understand, for this retailer's product catalog that has 100,000 products, which of those can be used in each scenario for the universe of recipe content?

Speaker 1:

So make sure we're not naively trying to use cucumber scented lotion when we're making a salad. Or if you if you have this kind of basic string matching, you'll run into all these issues. And then as AI progressed from doing classification, we started doing content generation. So we actually had a TikTok series called AI Chefs where we were using early versions of GPT-two. This was around, yeah, that 2018, 2019 era.

Speaker 1:

And then the early GPT-three models, before ChatGPT launched, we were having it create and generate recipe content and then, you know, trying it out for better or worse. But the model quickly got good enough that, it was more about challenging it to do something creative, because it was so good at all of the basic recipes. And, today, what we're doing is we've just stayed on the forefront of the capabilities with AI. So today, we're in the AgenTic era, and we're building the AI shopping agents or an AI shopping assistant for some of our retail partners, as well as capabilities delivered in the Cooklist app. This was really the core transition, was going from pure B2C to now serving retailers with B2B capabilities.

Speaker 3:

And is the Consumer app still out there and popular, or have you just made a jump?

Speaker 1:

Yeah. Yeah. The Consumer app is still available. We've got over a million people on it and, you know, works with all the major retailers in The US. The B2B side has the...

Speaker 1:

You know, there's been a lot of excitement and demand from the market this last year or two, so it's definitely taken more of our team's attention. But, we we do still support the consumer app and actually have what we think are gonna be some pretty cool plans, coming forward soon.

Speaker 3:

And in my world of entrepreneurs, b to b is hard. B2C is a thousand times harder, and doing both is like 10x harder than the thousand times. So you're definitely chosen the hardest path possible. And you're having success at both, which is exceedingly rare, so congrats on There's companies that have had... Companies that have pulled it off, there ends up being synergies in there, right?

Speaker 3:

So consumers may go to one of your grocery situations and love the experience and then download the app. And then they may be in the app and they, because you're working on the grocery side, you can improve that experience as they go cart to cart or whatever's happening in there. So there's like a, there's a, there's a bit of a flywheel that some companies have been able to build. So I have a feeling you've already thought about all that. I'm not telling you anything you don't know.

Speaker 3:

But, just for the audience, it's kind of interesting to think about both sides of that equation.

Speaker 1:

Absolutely. I mean, there's synergies. One, with the Consumer app, we're allowed to kinda have the freedom to experiment and move extremely fast to try new things out. We can then take those learnings and then bring them to our our retail partners. And then there's also synergies, right, where, you know, people shop at multiple stores.

Speaker 1:

So, they may, use the experience with a retail partner, for... Make their shopping easier there. But then they also use our consumer app to connect all of the places that they shop at, and have that, you know, that context of everything that they buy in one place. So there's absolutely, you know, synergies there and learnings that can be transferred both ways.

Speaker 3:

This is this is kind of a... This will make sense to you in a second, but there's... Doing what you're doing is hard. I've never tried it, but I I can just, like, feel it's gonna be hard. So so I was reading an Adweek article before we got on, just totally unrelated, but they they were they were talking about what happened at Conn.

Speaker 3:

All the ad people are at Conn. I'm sure you've it in your your timeline too. And they were talking about, a grocery store was putting ads into a chat. And then they... Then that caused them to try the chat.

Speaker 3:

And then they had a... The grocery store, who is not one of your customers or related to you and and other than being in the grocery, also had a a consumer app. And the the chat was was was... They had a great experience. So then this tester went and this person has two pescatarians, and they asked the app the grocery store gives them about building a recipe, and it kept injecting cat food as the ingredient that you should have.

Speaker 3:

Because it was just looking at, I'm sure the AI was looking at things that are fish based, right? Cat food is largely fish based. I'm not a cat person, but I know that I watch the commercials, right? So I thought that was, I think the Consumer app has given you the ability to harden a lot of these capabilities. And then you pull them into the b two b piece, it it feels further along than you would get from just the out of the...

Speaker 3:

You know, even out of the box OpenAI or or Gemini or one of the other models, like, you... You've got this layer, you kind know, of think of it as a harness. You've got this recipe harness on here where you know how to make sure that you don't, for instance, recommend cat food.

Speaker 1:

Yeah. No. We we actually... It's called our core identity system. And so something that we've developed proprietary in house to solve this problem.

Speaker 1:

And there's a lot of secondary benefits that come from it as well. So an example of a core identity that I like to talk about is hummus. So you could have a product that is like a plain hummus. You can also have a recipe, which is maybe like a salad wrap kind of sandwich that calls for hummus as a recipe ingredient. But then the third thing you can have is you can actually have a recipe to make hummus from scratch, just like blending up some garbanzo beans and olive oil.

Speaker 1:

And so we're able to represent kind of this trio of, you could have chicken enchiladas, which is both a product you can buy from the store, as well as a recipe where you make chicken enchiladas from scratch. And we take all the data a retailer has and classify it with that core identity system, which then allows us to if if you're looking at the recipe, we can show you a meal that you could just heat and eat instead. And this kind of also goes into the core thing that I think makes Cooklist different than most of the other recipe and meal planning experiences, is the problem we've always focused on is how do we, assemble the groceries you need in a shopping cart as efficiently as possible? Where we don't... We didn't really come from, like, a chef or foodie background where they're trying to push...

Speaker 1:

Cook every meal from scratch in this meal plan. We're we're really much more aligned with, we wanna help you get everything you need, including snacks and beverages and paper products, and really just try to, you know, get as close as we can to the end state, which is you just open the experience, and your card is filled, and it's perfect. Right? That's like the dream that we're working towards. And we...

Speaker 1:

You know, slowly making progress. I mean, I guess it's been a little faster than slow, but, you know, iteratively, right, where initially it was kind of tools that someone is using to make the process easier and remove some of that friction, like doing the ingredient reconciliation. And then today, it's the Assistant where you can actually offload some of that research and decisions and work to, and it can operate on the context of your previous purchases, what you've told it in the past, what you did last week, and then kind of assemble that together. And the the core thing that we say we're doing is removing from the era of search and select, where you go to the search bar, you search one thing, you select it, to now it's review and accept, where the agent or the assistant puts together a solution for you, and then you just review it. Maybe you need to make an edit.

Speaker 1:

But for most of the time, you're just gonna review and accept it and cut down the time and work that's required to do it.

Speaker 3:

Going back to the, so you started with this app. And then what led you to have this idea to do the B2B piece to kind of say, Hey, everything we've built in the app, or we have a lot of the parts, the Lego blocks you would need to power an on-site assistant for a grocer. What was that spark? Was it an inbound from a grocer who said, We need this? Or was it you looking at the grocers' experiences and saying, These kind of stink.

Speaker 3:

I think we can help them do this better. Like, There's always kind of like a thing in there. What was the spark that got you down that path?

Speaker 1:

When we first started the company, we went through the Techstars Retail Accelerator Program in partnership with Target. And even during that, Target was like, Hey, we really wanna white label, some of this technology and bring it onto our platform. And we were very, you know, maybe just naive. We were like, no, we're gonna be only focused on consumer. We're just gonna do the consumer app and make it work.

Speaker 1:

And really, for the first few years, we just had our heads down trying to build the best experience possible. And then it was inbound. So we had nothing on our website that said we did b to b. But, you know, some of the product teams within retailers were just going out and looking at the App Store to look for inspiration for meal planning experiences, shoppable recipe kind of experiences. And they saw our app ranking as one of the top apps in the App Store for those kind of situations when people...

Speaker 1:

You just search Meal Planner in the App Store, right, and we come up as one of the top ones. And they, yeah, reached out to us, had us participate in some RFP processes. And we've kind of expanded rapidly in the amount of capabilities that we're offering with the retailers. Originally, it was shoppable recipe, or we called it like a recipe commerce platform. But then they saw what we were doing with AI, and that's when we realized there was an opportunity to bring some of these conversational based AI experiences to retailers as well.

Speaker 3:

Yeah, so good. You had the inbound. Yeah. Yeah. And you didn't say no, so that's a lot of controversy.

Speaker 3:

You said no the first time, but then you finally were you know?

Speaker 1:

We said no to a lot of companies and, you know, think, like, appliance companies, other sort of

Speaker 3:

Having the fridge the fridge order. Yeah. That doesn't make any sense. Yeah.

Speaker 2:

When the

Speaker 1:

when the... Yeah. When when it really... We saw what they were looking for and we're like, this is exactly what we do. This You is a great opportunity.

Speaker 1:

Let's explore it.

Speaker 3:

And then how long... Like, given the idea of the time frame, how long have you been working on the b two b part? Is that, like, just in the last year, or or has this been a couple years in the making?

Speaker 1:

It's probably been going for, I would say, about three years overall. So yeah, maybe like 2023, we first started doing the Choppable Recipe technology. And then, you know, has expanded. We also do even, like, product merchandising, so on on websites using the core identity system. Because we have this deep understanding of food and how a a retailer's product catalog works together.

Speaker 1:

If you're looking at a main dish, we can, do a merchandising solution where, think of the product carousel below a product detail page, and you have main dish. Well, we can do it, others also purchased, and kind of use the cart co occurrence purchase data, to show items that you're likely to buy. But we can also kind of do a more, you know, something, that typically isn't done, where we show a side dish, and then we show... And it's a side dish that pairs well with that meal type. Right?

Speaker 1:

So if we're looking at the chicken enchiladas, right, we could have a Mexican rice, and then we could have a salad that goes with it, a beer, a wine. And Retail should kind of define these meal components. And then we can use our core identity system to understand the course and cuisine that's going to pair, and then do the real time calculations based on the inventory that's available at that store for the fulfillment method that the shopper is using.

Speaker 3:

So you did a fundraise. Walk through a little bit of the funding history. I looked up on Crunchbase and it said you had raised about $3,000,000 so far in your journey. Did you do like, You probably got the Techstars kind of a thing. Is that probably part of it?

Speaker 3:

But walk through kind of where you are on the funding side.

Speaker 1:

So Techstars was 2018 and we did our initial kind of like pre seed round, unpriced. And right after Techstars, yeah, we had a couple other. RevTech, a Dallas based retail technology firm also participated, as well as some friends and family. And then in 2021, we did our first institutional round, priced round, that was led by Mercury Fund out of Houston, Texas, and had some other great investors come in with that round as well. And then shortly after that, we actually achieved profitability and haven't needed to raise since then.

Speaker 1:

So we're growing. We were actually just looking at this the other day, and it's, yeah, I mean, I think we've had five years of over 100 year over year revenue growth. We just run a really lean team, applied AI to use the building the technology that we build. It's allowed us to be very capital efficient.

Speaker 3:

So then the app, does it have a subscription component or like a in app kind of... What's the monetization on the app?

Speaker 1:

Yep. It's a free... I don't need

Speaker 3:

to know details, but like just the...

Speaker 1:

Yeah, it's freemium. So there's a fully featured... You can do pretty much everything throughout the experience with the free version of the Cooklist app. And then we have a pro subscription, which has kind of like AI powered features. So importing recipes via photo, kind of using that AI assistant, the AI chef in there, And then as well as advanced dietary and meal planning requirements, so allowing you to filter by allergens, diets, things like that.

Speaker 1:

So it's basically just extra capabilities, but the whole core flow of There's not like a limit on how many items you can store in your pantry or things like that. Basically power user features for the Pro version.

Speaker 3:

I know. I mentioned everyone's at Con. So so the CPG people love to throw ads into things. Do you have an ad piece in there as well that you monetize?

Speaker 1:

There is. We work with AdAdoptin to... They, you know, are kind of a a source, the CPG demand on that side, and, yeah, they're our partner for advertising within the Cooklist app.

Speaker 3:

So three years ago, you decided to do this B2B thing based on an inbound. And now here we are, 2026. We're recording this kind of basically July. Give us an update. What's the scope and how much success have you had on the B2B side?

Speaker 1:

We're live at over 700 stores, reaching over 10,000,000 shoppers. I like to say that we started with the biggest and the best, grocer. You know, we're very fortunate to have them come on as our initial,

Speaker 3:

b two b partners. And... Is that the one that originally inbounded? Or Yeah. Yeah.

Speaker 3:

It worked out.

Speaker 1:

And, it's it's worked out extremely well. So... And we're we're expanding. You know, that that that reach right now, the number of locations that our assistant is live at, is going to double before the end of the year.

Speaker 3:

How many grocery stores are there? I should know this, but I'm a I'm a general merchandise guy.

Speaker 1:

Like, total number of doors in The US? Yeah. I mean, if you count the big box retailers, like Walmart and and Target and everyone, it depends really if you count Dollar General. I mean, they alone have, you

Speaker 3:

know, 1,000 plus. Grocery.

Speaker 1:

I don't know. They're they're making a push. You know? They're

Speaker 3:

I've seen they've got some freezers in the ones nearest, but it's not everyone.

Speaker 1:

But, yeah, it's it's probably on the order of, you know, that 20,000, I would say, is the total number of

Speaker 3:

So you still have a lot of wood to chop. 700. You got, you know, 19,300 to go.

Speaker 1:

Yep. Yeah. By my estimate. Lots of opportunity ahead of us, for sure.

Speaker 3:

Yeah. Yeah. Not that your progress isn't amazing already, already, but, yeah. The 10,000,000 shoppers is pretty... That's pretty solid.

Speaker 1:

But again, there's, you know, 300,000,000 people in The US, so, we're we're just at the starting line, really.

Speaker 3:

Okay. So I go to my favorite grocer, you know? And so is this kind of like a Rufus thing? So so I see the little agent there and, you know, what's the usual... I know I know the customer can probably choose the deployment and where this thing's in Jax, but what's your kind of like...

Speaker 3:

If we thought about your flagship experience, walk us through the flagship B2B experience at thegrocer.com.

Speaker 1:

It's essentially a shopping copilot. So, yeah, very similar to Rufus or Sparky. Essentially, like, what we're doing is, we're coming in as a technology partner to a grocer to allow them to have an experience that is at least on par, or if not in some ways better, than what Amazon and Walmart are doing, and do it extremely quickly. So we've been able to go from starting a contract to having a live deployment in forty one days with national scale. So...

Speaker 1:

And and the way that we do that is we build both a a front end kind of a little JavaScript app that can be inserted into an existing platform. So it's kind of like a one line install that reuses most of the retailer's existing infrastructure for their product catalog, that add to cart capabilities, tying that in with analytics or retail media. So the build requirements on the retailer is really low. And then we do all the back end work of essentially, we're not using a third party agent platform or service. We're just directly consuming raw language model APIs.

Speaker 1:

But what we build is the the platform and the scaffolding, the context engineering, which allows a...

Speaker 3:

Palantir would call it the ontology. You you you have to, like, build this whole ontology on top of there.

Speaker 1:

Exactly. Yeah. So part of that's this core identity system, which we can automatically apply to their whole product and recipe catalog. But then also just gathering the business requirements. We kinda act...

Speaker 1:

It's almost like an interview process where we're having them train a new digital employee to say, right, what are your values? What is the way that you respond? What are the merchandising principles? And how do you apply those when you're having a conversation with a customer? And we take all that information back from the business, and then we turn it into essentially the agent's identity and its personality.

Speaker 1:

And the tricky thing with these sorts of systems is they don't do the same thing every time you ask them. You can ask them the same question multiple times, and you'll get different answers each time. It's a nondeterministic system. And so to be able to have confidence before you put this live to your customers, what we also build and where we spend a lot of time is developing an eval system, which is essentially think of it as thousands of sort of test scenarios, where a customer may submit a question, and then the assistant is going to generate a response. And then we basically have grading criteria that are aligned with the retailer's principles for what that response should look like.

Speaker 1:

And then we can run it tens, hundreds of times, and kind of get a distribution of all the responses the assistant is gonna generate. And also understand, like, when it does fail or do something that's outside of one of the requirements, what did that look like, and how often is it happening statistically? And that has been really helpful for building confidence before putting systems live.

Speaker 3:

So you go through this interview process, you kind of show them all the things, and you go live. Where do you usually like it to be invoked? Is it kind of like a little persistent thing with a brand? You have magic or you have AI Chef over in your app. But over here, does it usually have a persona, like Rufus or Sparky or I think there's Magic Apron now.

Speaker 3:

They or Magic apron and red vest. What are the Milos one has a little vest.

Speaker 1:

Yeah. Milo. Yeah, yeah.

Speaker 3:

Yeah, Milo. Yeah. And then so that's probably a best practice. But then what's interesting is Amazon increasingly is doing you know, they're they're putting it on the PDP. So you could be looking at an artichoke or something and have it actionable there.

Speaker 3:

And then maybe even Rufus is increasingly sending some search flow. If they if they detect this is a query search is gonna kinda stink at, we might as well send it to Rufus. That's my language, not theirs. What kind of entry points do you recommend in this kind of flagship experience to the grocer?

Speaker 1:

It's a great point because this is such a new capability. It's one of those things where customers don't even know what to do with it in some ways. And so what we say is we have to inspire and educate shoppers on the ways that the Assistant can help them. Typically, there's always gonna be some kind of persistent button in the top right corner of the site that has the classic AI sparkles. But then as you're talking about with the artichoke, we...

Speaker 1:

When you open the Assistant, we have a few suggested prompts to kind of give some of that inspiration and education for what to ask. But we also have workflows which we can programmatically generate. So using AI to look at a product and then kind of working with the retailer and say, like, what types of questions do you want to inspire customers with about your products? Do you wanna inspire them with prompts about details? Just questions like, you know, what ingredients are in this, or is it gluten free, and kind of things like that.

Speaker 1:

Do we wanna inspire them with prompts that lead to using that product as a component of a bigger meal or a trip kind of, objective? And then we we... So we certainly have, like, several categories like that. And then we... Once we have those identified, we can say, okay.

Speaker 1:

For each product detail page, you know, generate prompts on these axes and then, programmatically generate them for the whole product catalog, and then they can show on the product detail page. And that's kind of where... As customers are browsing, they just see those. And, yeah, I think Amazon kind of does the best job of it today of, you know, really putting them in a lot of places so you can't miss it.

Speaker 3:

That's how I knew that it was converting better, because Amazon knows the monetization of every pixel down to six decimal points. And then once they started putting Rufus everywhere, I was like, oh, boy, it's off to the races now. You won't be able to move your mouse without Rufus popping up. I'm a product catalog nerd, and that's kind of we draw a lot of people to the show that really like to get into the weeds on this. How do you get their inventory?

Speaker 3:

Because you're just dropping a JavaScript on there, you're just kind of sitting on top of the e commerce experience and relying on it to have good inventory, and you're pulling the catalog from that. Is that kinda... Is that a fair assumption? Or do you have to... For a better experience, do you go pull it from other places or anything?

Speaker 1:

If a retailer wants, they can actually give us, like, a periodic sync and give us all of the catalog and inventory information across all of their locations. And then we will ingest that. And then when the agent is... For the agent, we build tools that it can use. So it has a search products tool, a search recipe tool, can have an update memory tool to save and store memories to use in future conversations.

Speaker 1:

We also have other tools for accessing procedures, like standard operating procedures when certain scenarios are encountered, or references that could be... Think of like FAQs or questions customers may have. But for the search product tool, we need to access what products are available at this location the user is shopping at. And so we can either ingest that periodic dump and run the query on our end, essentially that search query, or we can hit a retailer's external search endpoint. So what happens...

Speaker 1:

The easiest thing is they already have a search endpoint that their clients, their app, and their website will hit whenever someone is using the website to execute a search. And we'll essentially just make that same search query but directly from our server. And so then the agent, can actually, do this kind of iterative process where it may search the retailer using one term. But if the results aren't what it's looking for, it'll actually execute another search term with more detail, or it'll do four or five terms, to really try to find the specific thing the customer is looking for. And just like you as a user of a website, you can apply different filtering options or things like that to narrow down the results.

Speaker 1:

We also give the agent the ability, when it's using the search tool, to do things like filter the results to only items the customer has purchased before. If they're looking for something that the intent from their prompt is something we can tell that they bought before. Or if they want something new, we can filter out the things that they bought before, and then be sure that we're only showing them new items to try. But yeah, essentially, just from the product catalog standpoint, it's ingesting it from either search endpoint or a bulk transfer to our system.

Speaker 3:

It's pretty impressive the number of tools you're giving the grocer to manipulate this experience. Imagine that's been where a lot of the effort has gone. What are, are there any here, any of those tools we haven't mentioned yet that are, so I like the memory one. That's pretty neat. How about, you know, even today, grocers are big on their flyer.

Speaker 3:

Can they put their flyer in and have it kind of like know that, oh my gosh, I've got a big promotion on barbecue on Aisle 6 or something like that?

Speaker 1:

Yeah, exactly. So we give the merchandising teams that capability. We have an internal admin system, which the grocers team can use to access all their product content, recipe content. Some teams even develop recipes and have the nutritionists and the testers and everyone kind of doing a whole recipe development workflows within our admin system. But everything related to the shopping assistant is managed in there as well.

Speaker 1:

So, they can go in in seasonal, different holidays, right, kind of put together their merchandising campaign, for fourth of July, or, you know, we just had Father's Day. Whatever those products are that they're wanting to promote both in store and throughout their digital experiences, the teams can input that information. And then we have ways of linking the suggested prompts directly to a procedure to basically guarantee that it's always followed. Otherwise, the procedures kind of have sort of like a title in a scenario. And then the agent makes a determination based on the customer's response, should I grab that procedure?

Speaker 1:

One funny one is actually a fish merchandising procedure. And so if the customer is talking about making fish, it'll pull the fish procedure. And then if they're gonna make the meal either that day or the next day, the agent recommends fresh fish. But if they're planning to make this fish tacos or whatever later in the week as they're developing their meal plan, then it'll recommend frozen fish. And so this is something that was important to the retailer to make sure that behavior was there.

Speaker 1:

And we were able to develop a procedure to have the assistant do that reliably without, I think actually a big thing, anyone who's considering something like this, is cost and speed. So the LLMs can be fairly expensive, especially if you're running one of the flagship high intelligence models. But then also speed, right? Customers are used to search, and search returns extremely fast. And so we don't want to have to wait.

Speaker 1:

You see the ChatGPT thinking, and it takes two minutes, three minutes for a response. And so we have a lot of optimizations with our platform where the agent does that search tool, and it sees the phish product that it wants to return and show to the customer. But instead of it having to transcribe out the title, the image, the description, the reviews, all that information that we wanna return to the client, We have a very optimized system where it just returns the ID. And then we have a stream processor that's processing the token stream going from the server to the client. And it basically sees that ID.

Speaker 1:

It it then checks to make sure that that ID is not a hallucination. So we actually do reconciliation between what the search tool...

Speaker 3:

You're eval ing on the fly.

Speaker 1:

Yep. And then then we just populate all the results from the database. So it's extremely fast and extremely token efficient. There's lots of little optimizations like that that just allow this to be

Speaker 3:

And imagine you're calling all kinds of models in there because you know that for this, I'm going do that. And then like, Anthropic's always down right now. So you have to have like a plan. That's never a good consumer experience. So you have to...

Speaker 3:

For these consumer type, I know it's B2B, but it's B2B to C, consumer is going to just bail on it if you... It doesn't love saying, sorry, the cloud's busy right now. Come back later.

Speaker 1:

And so that's where we're model agnostic, so we can plug in any LLN provider on the back end. And then with that eval system, this allows you to make a good business decision where you could say, okay. If we pay Anthropic for Opus 4.8 to run it, what does it cost and what does it score? But then if we drop down and go to a smaller model like Sonnet or even an open source model, you can really just have a very objective comparison on the cost and performance with running different models, which is helpful from a business decision as well.

Speaker 3:

So it's hard to talk about grocery without talking about Instacart. How do you view, they're kind of over here doing their thing. Grocers kind of see them, this is my own interpretation. I think they liked them at first, but I think now they would rather just kind of do it. They would rather control.

Speaker 3:

They don't like someone coming between them and the customer. I think that benefits you. But you see them... I know they try to then... There's part of Instacart, and I don't know much about this.

Speaker 3:

I see them talk a lot about how they're trying to sell things to grocers too. But I imagine it almost like Amazon's tried this before and not got a lot of uptake with retailers because it's kind of like, why would I use my competitor's technology? Is that... Like, let... I don't even know if they...

Speaker 3:

Do they have a competing offer? And, you know, do you do you... Am I right that people would choose you because, you know, it's kind of letting the hen... The the fox in the hen house?

Speaker 1:

No. They they do. I mean, they've actually been talking about it since last year. I haven't been able to try it yet. I've seen a lot of press about it, but I haven't seen it live.

Speaker 1:

But, they do have a AI assistant that is kind of a a direct competitor, that's mostly part of their Storefront Pro offering, which is kind of their e commerce platform for grocers. And we have, you know, done some, bid processes where we're up directly against Instacart, and, you know, we've we've won out because of some of the reasons you mentioned, you know, we would would assume.

Speaker 3:

Seems like a heavier lift too, because I imagine it's not uncoupled from the entire platform, whereas you can be... Just be drop in. You're you're a a la carte type situation.

Speaker 1:

Exactly. Yeah. We have lots of different capabilities. And another one that's, you know, more and more important to retailers today is the retail media, right? Instacart is very wanting.

Speaker 1:

If you're gonna use one of their systems like that, they're gonna want you to be using their retail media platform. So we can also integrate with, Criteo or any other retail media platforms that a retailer is working with. And just, you know, we kinda come in as the bespoke white glove a la carte service that is really here to to serve the retailers and just try to make it easier for customers to get the groceries they need.

Speaker 3:

Where do you see the future of the company? Obviously, you've got to go from 700 to 220,000, so I'm sure that's on your task list. Are some... Any other kind of forward things you're thinking about as the... As CEOs and founders, we have to spend some, you know, non 0% of our time thinking about the future.

Speaker 3:

So so what's your... How do you see the future of the company, and then where do you see Agenda Commerce going?

Speaker 1:

We see a big opportunity, one, to keep growing the conversational assistant. We're also bringing those AI capabilities to other parts of the experience. So Agentic Search, where we're able to handle kinda like, you know, you see Amazon doing already, where it's now a search or ask bar. We've developed a similar but actually completely separate approach, that's highly optimized for search that uses high speed inference providers like Cerebras, where we can still return search results in one hundred milliseconds to one second using LLMs to handle the difficult queries. If you type in taco night into a retailer's website, it's either gonna key on the keyword taco or night.

Speaker 1:

And we've seen some experiences where you have underwear, adult underwear essentially showing up because of the word night or sleep supplements and things like that. When really searching taco night, you want taco shells or tortillas, probably some ground beef and cheese and sour cream. And so we actually use an LLM to break down that intent into a bunch of subcomponents and then also use another set of LLMs to curate the results that are showed. And so we think there's a really big opportunity there, just meeting customers where they already are in the search bar, as well as agentic merchandising. So for occasions, a merchant team may put together a fourth of July home page, of think of it like an end cap, right, of the, here's the five or 10 products we're gonna kind of feature.

Speaker 1:

But every shopper now gets that same carousel for fourth of July. What we've done is kind of trying to give the merchant teams superpowers where they can say, hey, we want to make a fourth of July carousel on the site, but we want to have kind of permutations based on these 20 different axes. Think like price and dietary requirements and just preferences on brand and things like that. And so we can actually then run that agentic process to make one thousand fourth of July bundles and then personalize them to the shopper based on their transaction history and other interactions with the site. And so I think the end state, right, that we're working towards is, ideally, your your shopping assistant with a retailer, maybe even proactively reaches out to you and says, hey, Scot.

Speaker 1:

I've got everything that you're gonna need from the store this week. Click this link to check it out, and then you can order it for pickup or delivery with one click. But on the other hand, we we still see 50% of the time where people are adding items to the shopping cart with an AI assistant. They're going in the store and completing the transaction in the store. So there really is this you know, omnichannel shopper that's kinda doing that planning and and some of the, you know, cons...

Speaker 1:

Decisions and considerations of what they're gonna buy in the digital experience. But then they're still going in the store to fulfill it. I think for us with Cooklist, we're thinking of how do we make the synergy between the retailers and then the consumers using our Consumer Home Inventory app, even with helping things like forecasting demand, where we also work with some appliance partners. Think of smart fridges and kind of really trying to make the context of what's in home and bring some of that to retail partners. And, you know, all all at the end of the day, to just make the weekly grocery trip a whole lot easier and faster.

Speaker 3:

That's a lot to work on. So you've got a full... The the backlog is always full when you're you're... Even though AI lets us chew through it, then it just more comes behind the Right.

Speaker 1:

Holding. It's like AI was supposed to make it, like, I mean, we've never been working more.

Speaker 3:

Especially I'm totally in the camp that this creates more jobs than than takes them for sure. Yeah. Because we can just, like, do... We can... You know, your imagination is basically the boundary now, which is a map, but you know, allows you to build these just incredible experiences.

Speaker 3:

Appreciate you coming on. If people want to learn more about Cooklist, obviously you can go to the website and then you're pretty active on LinkedIn. Any other places people could follow to kind of learn more about what you're doing and see your thought leadership, or is that the the best places to go?

Speaker 1:

Yeah. Cooklist.com. Or, you know, you can send me an email, danielcooklist dot com. And, yeah, would love to to connect and always open to talking about ideas in this space. I think it's super fun.

Speaker 3:

Really appreciate you taking time, out of your busy schedule to come on, and I learned a thousand things from this. So it was, really, really amazing. Thank you.

Speaker 1:

Awesome. Yeah. Thanks for the time, Scot. Really enjoyed it too.

Speaker 2:

Follow, rate, and share to stay up to date with Retailgentic, your guide to the future of AI shopping agents.