The retail revolution won’t wait – and Dataiku is making sure it doesn’t miss a beat. At NYSE Wired: AI & Retail Trailblazers, Jed Dougherty, head of AI architecture at Dataiku, joins theCUBE’s John Furrier to unpack how AI is accelerating from back-office buzzword to front-office reality.
Dougherty shares candid insight on why fewer than 10% of retailers are truly AI-ready – and what’s holding the rest back. From GenAI-powered customer interfaces to rethinking governance beyond the “Copilot for all” strategy, he outlines how trust, process alignment and strategic focus separate experimenters from AI-native enterprises. With examples ranging from Stitch Fix to Rolex, Dougherty explains why data maturity is fast becoming the new retail differentiator.
In a wide-ranging discussion, he demystifies the architecture required to support agentic AI, warns against the pitfalls of “vibe coding” on unverified datasets and makes the case for visual interfaces that empower business users. Dougherty also previews Dataiku’s investments in open source and tight integrations with players like NVIDIA and Snowflake, highlighting how the company aims to be the indispensable human layer atop modern AI stacks.
Forgot Password
Almost there!
We just sent you a verification email. Please verify your account to gain access to
theCUBE + NYSE Wired: AI & Retail Trailblazers. If you don’t think you received an email check your
spam folder.
Sign in to theCUBE + NYSE Wired: AI & Retail Trailblazers.
In order to sign in, enter the email address you used to registered for the event. Once completed, you will receive an email with a verification link. Open this link to automatically sign into the site.
Register For theCUBE + NYSE Wired: AI & Retail Trailblazers
Please fill out the information below. You will recieve an email with a verification link confirming your registration. Click the link to automatically sign into the site.
You’re almost there!
We just sent you a verification email. Please click the verification button in the email. Once your email address is verified, you will have full access to all event content for theCUBE + NYSE Wired: AI & Retail Trailblazers.
I want my badge and interests to be visible to all attendees.
Checking this box will display your presense on the attendees list, view your profile and allow other attendees to contact you via 1-1 chat. Read the Privacy Policy. At any time, you can choose to disable this preference.
Select your Interests!
add
Upload your photo
Uploading..
OR
Connect via Twitter
Connect via Linkedin
EDIT PASSWORD
Share
Forgot Password
Almost there!
We just sent you a verification email. Please verify your account to gain access to
theCUBE + NYSE Wired: AI & Retail Trailblazers. If you don’t think you received an email check your
spam folder.
Sign in to theCUBE + NYSE Wired: AI & Retail Trailblazers.
In order to sign in, enter the email address you used to registered for the event. Once completed, you will receive an email with a verification link. Open this link to automatically sign into the site.
Sign in to gain access to theCUBE + NYSE Wired: AI & Retail Trailblazers
Please sign in with LinkedIn to continue to theCUBE + NYSE Wired: AI & Retail Trailblazers. Signing in with LinkedIn ensures a professional environment.
The retail revolution won’t wait – and Dataiku is making sure it doesn’t miss a beat. At NYSE Wired: AI & Retail Trailblazers, Jed Dougherty, head of AI architecture at Dataiku, joins theCUBE’s John Furrier to unpack how AI is accelerating from back-office buzzword to front-office reality.
Dougherty shares candid insight on why fewer than 10% of retailers are truly AI-ready – and what’s holding the rest back. From GenAI-powered customer interfaces to rethinking governance beyond the “Copilot for all” strategy, he outlines how trust, process alignment a...Read more
exploreKeep Exploring
What insights have been observed about the impact of companies making big bets on generative AI at NRF?add
What are the implications of integrating generative AI into consumer-facing interactions for businesses?add
What is the importance of governance in the decision-making process for AI projects in successful companies?add
What is Dataiku's value proposition and how does it implement its enterprise AI platform?add
What factors should be considered when making architecture decisions regarding IT budget allocation for Generative AI?add
>> Welcome back everyone. I'm John Furrier, host of The Cube here at our NYSE Cube Studios, of course. We have our Palo Alto studio connecting Silicon Valley and Wall Street. This is part of our NYSE Wired series, a Cube original program we've kicked off about a half a year ago where we feature all the leaders. This is Retail Week. NRF has been kicked off. We've got all the leaders coming in. You had CES the week before, NRF, One-Two Punch, physical AI. Now you got retail AI, which has a little bit of both. It's all about agents. Jed Dougherty's here. He's the Head of AI architecture at Dataiku. Jed, great to have you on again on the queue. Appreciate you coming on.
Jed Dougherty
>> Thanks very much, John. Happy to be here.
John Furrier
>> So honestly, you're talking off camera, you at a pharmaceutical conference, now it's a retail conference. The themes are rhyming.
Jed Dougherty
>> Oh, yeah. Oh, yeah.
John Furrier
>> Let me guess, agents and AI.
Jed Dougherty
>> Shocker, right?
John Furrier
>> What's your take on NRF this week? I want to get into some of the AI architecture stuff, but I really want to get your views on NRF first.
Jed Dougherty
>> So yeah, NRF has been fascinating over the last few days. What I've really seen is that there's a few companies who are betting big, and a lot of companies that are hoping they can get away with staying pretty even. There's definitely a situation in which these big bets are scary to existing workflows, to existing revenue streams. And if you're big enough, maybe you're not willing to make that big bet. To me, the way these big bets are revealing themselves, a lot of it is pushing gen AI and AI out from the back office and in front of the end consumers. So how are you going to have your end consumers interact with you through gen AI is the question everybody's trying to answer.
John Furrier
>> Which is the data.
Jed Dougherty
>> Yeah. It's-
John Furrier
>> The back office migration to the front office. User touches.
Jed Dougherty
>> Exactly. It's easy to say for back office like, oh, we're going to enhance productivity by having our coders do more stuff using gen AI. It's a lot scarier to say we are going to provide some type of gen AI interface for our end consumers.
John Furrier
>> Based upon the data that's in the back office.
Jed Dougherty
>> Yes, exactly. Yeah.
John Furrier
>> Because the data that surfaces in the front, that's generative, so it's going to be ready to go at a moment's notice.
Jed Dougherty
>> Yeah. The use of back office data to curate a customer facing front of product experience is very powerful. I was just talking to the chief product officer at Stitch Fix, great company, who they were saying that they'd spent the last 15 years... They have an advantage over maybe some of the bigger retailers. They've spent the last 15 years building AI products essentially. They've been trying to decide which piece of clothing they should be shipping to their customers. It's been very easy for them to layer that gen AI interface on top of their prior huge stack.
John Furrier
>> Because they did the work.
Jed Dougherty
>> They did the work already.
John Furrier
>> So foundationally they were prepped?
Jed Dougherty
>> Exactly. Yes.
John Furrier
>> All right. What would you say, and you're based on... I mean, statistically not... I mean, close, order of magnitude. The people ready and not ready, how would you scope the retail market percentage-wise?
Jed Dougherty
>> I would say-
John Furrier
>> When I say ready, I mean like really ready. Projects are rolling. Somewhat ready would be, okay, we could get up and running.
Jed Dougherty
>> I think really ready is under 10%.
John Furrier
>> Wow, that's low.
Jed Dougherty
>> Getting ready is maybe another 30% and folks spinning their wheels, and having no idea what to do is 60.
John Furrier
>> That's a huge number.
Jed Dougherty
>> Yeah.
John Furrier
>> What's the issue? People? Is it infrastructure?
Jed Dougherty
>> We've got clients all across retail. We've got Love's travel stops. We've got LEMH. You got your gas pump and your heel pumps competing with each other there. We got clients like Rolex, all of these different organizations, all within retail, all kind of struggling with what to do right now. The ones that are able to move forward and the ones that are able to have a lot of success are the ones that have built trust and governance into the way they choose projects. I think there was this big move... Well, I know there was this big move over the last couple of years where it's like, "Oh my God, we need a GenTech AI. We need Gen AI. What are we going to do? What are we going to do? All right, let's just give 40,000 Copilot licenses to our entire organization and hope somebody builds something cool."
That doesn't work.
John Furrier
>> That's not a strategy.
Jed Dougherty
>> That's not a strategy. That's just a-
John Furrier
>> That's hope.
Jed Dougherty
>> Yeah.
John Furrier
>> It's prayer.
Jed Dougherty
>> Throwing stuff against the wall until it sticks.
John Furrier
>> What a prayer.
Jed Dougherty
>> The companies that have succeeded have identified the decisions that are most critical to their success and then have built in a governance layer so they can choose which AI projects are actually going to greatest help those decisions. Our customers who are working on that type of thing, I think they have been helped by what we offer that really helps them with that governance layer. When I say governance, I mean process governance. Who's going to sign off on this? Who is regulating this? Who's documenting this? How do we ensure that it's actually going to drive value? All of the above.
John Furrier
>> I want to get into some of the architecture things you guys are working on because you're the Head of AI Architecture, which is awesome. But first, take a step back. Context. Give a quick summary of what you guys are doing, the current strategy, products. Set the table for your value proposition, how you guys roll it out, and what it does.
Jed Dougherty
>> Right. Dataiku presents itself as the enterprise AI platform. However, platform, a lot of people talk about platform.
John Furrier
>> Everyone has it.
Jed Dougherty
>> Yeah. When we think about platform, you can think about a few different layers. Maybe at the base layer you have data. So the folks providing your data are your Databricks, your Snowflake, your AWS of the world. On top of that, you have compute. So how am I interacting with that data? How am I processing it? How am I turning it into something useful? So there you can obviously think of NVIDIA, organizations like that. Then the layer on top of that that maybe we forget about a little bit with these two, is people. Who is going to be building things with this if data and compute are all about how do I run these AI platforms? Dataiku believes in how do I build these platforms and how do I give people the right tools to build these platforms?
John Furrier
>> Basically, interface with the data as a development product.
Jed Dougherty
>> Exactly.
John Furrier
>> That's what you guys do.
Jed Dougherty
>> Absolutely.
John Furrier
>> Data is the new software in that weird way.
Jed Dougherty
>> Yes. Yeah.
John Furrier
>> It's a data development kit. Go code.
Jed Dougherty
>> I love it. Yes. Thank you.
John Furrier
>> Not a kit. I mean, I'm dating myself when... Developer kit is an old con store. "Here, here's some software development kit, go build." That's essentially what you're saying. I think your point about the back office to front office is interesting because what I've heard from other interviews, not just here, but other interviews is that the gen AI is interfacing with our data because that's what it is. The prompts and the answers are the mechanism of how it's being delivered to some software involved, but ultimately it's the data.
Jed Dougherty
>> Yeah. There's some nuance around that. When we think about vibe coding, which is kind of the hot thing right now, everybody wants to vibe code. I vibe code all the time for the record.
John Furrier
>> Yeah. It's fun.
Jed Dougherty
>> Yeah, it's really fun. There's a real-
John Furrier
>> Until you can connect five APIs together in there. Wait a minute, where's that database? Putting up a post, Chris, is pretty easy.
Jed Dougherty
>> Right. I want .
John Furrier
>> With six parameters.
Jed Dougherty
>> I think there's a danger with vibe coding on data. The challenge there is really, if I'm vibe coding of front end, I build a nice website. If I click a button and the button works, then I know the code was correctly written. If I'm vibe coding on top of a database, I can ask the question, what's my mail to female distribution of buyers in France? And that's a relatively simple SQL query. It's going to go write that query. It's going to hit my database. It's going to give me a number back. But I don't have any way of confirming whether that number is correct without me reading all the code, which nobody's doing anymore. People are vibe coding without even knowing how to write code. And so, how do we transition the data world into one in which we can be comfortable, maybe not vibe coding, but vibe talking to our data? A lot of this text to SQL stuff, I don't think solves that problem, which is where a company like Dataiku that has a full visual interface for describing how data gets broken down... If I can converse with Dataiku, it builds me a visual interface. I can check, make sure that it's working well, that it's answering the question correctly. That allows me to apply this vibe coding paradigm to big data. There's a big gap there that I think most people are filling right now.
John Furrier
>> I think vibe coding is just a great way. It's like the first version of ChatGPT. It was elementary, but groundbreaking.
Jed Dougherty
>> Yeah.
John Furrier
>> And if you look back now at ChatGPT's first rev, it kind of like is that Netscape browser mode. "Oh, I can't believe their fonts are so terrible." But it just got incrementally better. I guess my question is, as agents come on board, that's a tailwind for you guys because now the agents can fill in the gaps where my API comment gets solved.
Jed Dougherty
>> Yeah.
John Furrier
>> Okay. He wants to connect to a database that's not in the vibe coding maybe kit, but it's an enterprise secured with identity.
Jed Dougherty
>> Yeah.
John Furrier
>> Okay. That's interesting.
Jed Dougherty
>> I don't know if this is going to be too nerdy of a statement here, but the evolution of agents, I think what we're very quickly going to be seeing is exactly what you're saying, is that the agents are going to write code on behalf of the user to solve whatever the question that user has is. So you don't have to have 10,000 tools. The agent is going to write the tool on the fly to do the thing to give you the information back. I'm very excited.
John Furrier
>> All right. What is the biggest AI architecture best practice? Thinking architecturally, we're in a systems architects mode right now. You've got developers and system architects. Those are the two hottest areas in my mind that are most active on the long game, setting the table, setting the foundation because that's what's happening. You got to look at AI infrastructure. It's all about architecture. How am I dealing with the memory? It's all the internals, if you will. Then devs, that's what can devs do? What's in your mind a good way to think about the different categories of companies, small, medium, and large, that are thinking through... Again, budgets aren't massively growing. I mean, no one has NVIDIA money.
Jed Dougherty
>> No.
John Furrier
>> I mean, unless you're JP Morgan Chase with $10 billion of IT budget a year. Not everyone has that kind of cabbage.
Jed Dougherty
>> I think I'm going to twist this one a little bit to talk about how you should be making your architecture decisions. I think there's a real risk of saying, "Okay, we're going to have our developers over here and we're going to have our architects over here, and they're going to be making all of the decisions on how we're going to be spending our IT budget on Gen AI."
Meanwhile, you have the business users over here, or the people over here who may actually be deciding which applications and which interfaces you should be building. What do I show to a customer? The IT guy doesn't know. This business person over here knows. If they're making decisions in isolation from the actual business users, from the people who know the business, you're going to have a massive wasted of money.
John Furrier
>> That's problematic.
Jed Dougherty
>> It's a huge problem.
John Furrier
>> Yeah.
Jed Dougherty
>> I think that-
John Furrier
>> That's why I like your vibe coding crossover to reliable pre-production grade thinking because when I did my vibe coding, it was all the projects that I wanted my team to build, but I didn't have time to even go to the meetings to describe it.
Jed Dougherty
>> Totally.
John Furrier
>> I just did it in a weekend. I was hanging around, "Oh, this is cool. Look at this great app I just built. This is what I want," almost like it's like my PowerPoint. I'd love to finish it without involving the team. Get on Slack, "Hey, I need credentials for that server." This is where you start to see kind of like the Lego block thinking.
Jed Dougherty
>> Totally.
John Furrier
>> That's a business user mindset.
Jed Dougherty
>> Yeah, it's absolutely... It's very similar to the data science problem of 10 years ago of like, "Oh, I have all these-"
John Furrier
>> Explain that.
Jed Dougherty
>> Okay. The data science problem, which anybody who's worked with data scientists, and I used to be one, knows-
John Furrier
>> When really kind of came on the scene.
Jed Dougherty
>> Yeah, yeah, it was hot. Everybody was like, "We got to go hire some data scientists." Sexiest job in America, yada, yada. Those data scientists built out models and algorithms that ran on their laptops and worked once. They were like, "Our job here is done." Then there was this huge, painful gap of productionalizing what the data scientists had built so that enterprises could actually get value out of it.
John Furrier
>> Because they were math nerds, they were statistics, they were building the dashboards.
Jed Dougherty
>> I want it to work once, and if it works once, that's good enough for me.
John Furrier
>> Go scale it.
Jed Dougherty
>> Why do I need to worry about my database access credentials, or if it's going to respond in 30 milliseconds versus 200?
John Furrier
>> Not my job. Not my job is there.
Jed Dougherty
>> Yeah. And then whose job was it? I don't know. Then these projects fell by the wayside. If that's what happens with vibe coding all over again, then we've really learned nothing in-
John Furrier
>> It's kind of happening.
Jed Dougherty
>> Yeah, I know.
John Furrier
>> It is kind of happening right now.
Jed Dougherty
>> Yeah. It's a real risk for-
John Furrier
>> All right. And so, the solution is... You guys have a solution for this?
Jed Dougherty
>> Yeah. It's called DataIku. The companies-
John Furrier
>> Give an example. Give an example.
Jed Dougherty
>> Sure. Sure.
John Furrier
>> Picture a working example because I think this is where the innovation could turn into a toy view versus a real productivity game. Explain how you guys cross that over.
Jed Dougherty
>> Sure. Let's say you're working in insurance claims. I know it's a very boring example, but it's one that everybody understands.
John Furrier
>> Yeah, and it's an easy way to improve something.
Jed Dougherty
>> Right, sure. Claim comes through, any different decisions need to be made on it. You probably need to go pull information about it. There's an approval or a deny situation, and then you have to keep track of what happens to that claim afterwards. Now, right now, in today's world, that multi-step process is being performed by some combination of actuarial work by claims processors, by intake workers, maybe by a legal team. You have a bunch of humans in the loop across this entire process. Now, we're going to be able to start identifying aspects of this thing, but the actual building of each one of these agents cannot be done by IT.
It needs to be done with the help of the experts who are currently working on that flow. With Dataiku, these experts are going to be able to work directly in an interface with IT, build out the things they need, and then instead of shipping it off to somewhere else and it gets programmed and returned six months later, it's ready to go. It's production ready. It has-
John Furrier
>> I mean, IT is curating, not building.
Jed Dougherty
>> Yes. Exactly.
John Furrier
>> They're just verifying connectors, delegation, credentials.
Jed Dougherty
>> Yeah. And so if it's not IT building, who is building? It has to be the business users.
John Furrier
>> Yeah. Yeah. Okay. What's on your agenda for this year? How do you see the retail and other verticals? Because clearly when you get into domain-specific expertise, that's where this shines. But you also have the horizontal platform needs too. You got to have horizontal scale, but vertical specialization. That's where the innovation is. What's the focus?
Jed Dougherty
>> I think there's... I don't know if this is the ultimate focus, but it's one that I'm very familiar with, so maybe it's worth me speaking to, is the decision from an architectural level across all of these different organizations between am I going to run my own model or am I going to continue paying the API services? Am I going to spin my own stuff up? Am I going to buy a bunch of NVIDIA GPUs myself, or am I going to continue to trust Anthropic and OpenAI to be worth the per token investment under the assumption that they're going to keep rolling things out faster than I could on my own? And like, okay, what are my data privacy policies-
John Furrier
>> I have an opinion on that....
Jed Dougherty
>> and everything like that? Please.
John Furrier
>> I want you to, well, share your data. My opinion on that is I think it's going to be a hybrid. I think if you're serious about being competitive in the future, you have to have your own model. If you have data, you have to have your own model. I'm as encouraged by Amazon Web Services Frontier Agent and the Nova thing that they had where they said, "Oh, here's a half-baked, distilled... Go take this and make it your own."
I think deep seq pointed to that, that you don't really need the entire web. You need to have a distillation of that here. I think there's going to be fusion between domain-specific models that are highly accurate, vetted, bulletproof, and then integrating in with some sort of parameter pass, I don't know what to call it, but like integrated neural connection, I'll just say.
Jed Dougherty
>> Yeah.
John Furrier
>> I just don't see any other way because... Those guys just cannot... Unless they have a service that comes out that says... Unless OpenAI and cloud say, "Hey, we will distill for you." That's possible. I just don't see that. I think if I'm a customer, I'm like, "Crown Jewels."
Jed Dougherty
>> Yeah, I want my own stuff.
John Furrier
>> I might not need a huge-
Jed Dougherty
>> Just mine. That's my market differentiator.
John Furrier
>> I might need a small GPU box from NVIDIA and some compute.
Jed Dougherty
>> Yep. Yeah. I think that everybody we talk to whose big enough is looking at some type of hybrid.
John Furrier
>> What do you think about that?
Jed Dougherty
>> Yeah.
John Furrier
>> Do you agree?
Jed Dougherty
>> Yeah, yeah, definitely. We've been working heavily with NVIDIA. They have something called NVIDIA AI Enterprise in which they're rolling out their own models that are optimized to work faster and cheaper on NVIDIA hardware. We have seen a big shift to that. Nobody's moving entirely away from the API-based services, but for specific use cases, for situations where you need to inject a whole bunch of your own information or your own private company data, this does become a more sensible, more effective solution.
John Furrier
>> All right. What's your priority for the year? What's on your agenda? What events are you going to be at? What are you optimizing for? What's your focus?
Jed Dougherty
>> A lot of my-
John Furrier
>> That's like five questions in there. Pick one.
Jed Dougherty
>> Yeah, yeah. Oh, my.
John Furrier
>> I'd love to find out what events you guys going at because we're going to hit a bunch of events. I mean, as you guys develop this, you're on the front end of this wave. Again, every vertical is beautiful for you because you can basically bring data to the front end.
Jed Dougherty
>> Yeah. A lot of my focus this year is on further cementing our relationships with the other players in the space. We know that we're never going to be the only piece of software inside of somebody's data stack, and we want to make sure that we are the best piece of software on top of the NVIDIAs, the Databricks, the Snowflakes, the OpenAIs, the Anthropics of the world, because that's why people choose us. A lot of what I'm doing is talking to those other companies. It means a lot of flying to San Francisco, to be honest. The other big thing that we are investing heavily on, and you're actually the first people I've ever told this to, is we're opening our own open source office. We've just hired a Director of Open Source who's going to be reporting to me. We're making additional hires in the Bay around open source projects. We're going to be investing heavily in open source applications that are specifically around gen AI governance, trust, security, control.
John Furrier
>> Yeah, that's a good call. I think one of the things that you're pointing to, and we've seen certainly in the semiconductor side, tightly integrated products, engineered, not just bolt-ons. Engineered is the new ecosystem. The old ecosystem was, "Hey, API, we're in partner. Hey, we sell together, all good." I'm talking about tightly coupled integrations. Is that something that you guys see?
Jed Dougherty
>> Yes, absolutely.
John Furrier
>> And that's what you're doing?
Jed Dougherty
>> Yeah. Well, and if you think about when I was talking to you earlier-
John Furrier
>> If you want to be that layer, you've got to be the best.
Jed Dougherty
>> If we're the people layer, the compute layer and the data layer wants us to be as tightly integrated into as many of their capabilities as possible, because that encourages them to bring us .
John Furrier
>> Yeah, and your performance has to be right next to the next best token. Jed, thanks for coming on. I appreciate it.
Jed Dougherty
>> Absolutely.
John Furrier
>> Wrap up the day on Retail Trailblazers. Retail Trailblazers is our new series in AI featuring the leaders from AI architecture to the front office of retail. The user experience will ultimately be determined by the people who can make the effect to change the best. That's the domain experts. Of course, you've got to have the scale and the business operations all going to be tightened up with AI. We're doing our best to bring that to you. Thanks for watching.
John Furrier