Brent Peterson (00:02.463)
All right, welcome to this episode of Talk Commerce. Today I have Rochelle Phelan. Phelan, sorry. She is the CEO of Traject Data. Rochelle, go ahead. I know I've just screwed up all the names. I got CEO, right? Go ahead, do an introduction for yourself. Tell us your day-to-day role, one of your passions.
Rochelle Thielen (00:21.358)
Okay, no worries. You got it. So Rochelle Thielen, like you said, CEO of Traject Data, Day to Day, we are really at its essence pulling together the data that's powering a lot of the AI, ML, LLM type solutions for the retail industry.
Brent Peterson (00:40.095)
That's awesome. a passion, do you have a passion life?
Rochelle Thielen (00:42.926)
Passion and life, ooh, this goes way back for me. I actually originally started flying helicopters when I was younger and still living in Minnesota. And still, I think for anybody who's had that flying bug, it never leaves you. helicopters are my passion.
Brent Peterson (01:02.013)
Wow, that is a great hobby and you don't hear a lot of people that aren't professional helicopter pilots. okay, well that's perfect.
Rochelle Thielen (01:08.8)
I happen to be married to one. So that's a one career per household choice. So years ago, I don't regret it, but I went the other direction. But at home, I have a husband who flies for UCLA Children's Hospital.
Brent Peterson (01:17.064)
Yeah.
Brent Peterson (01:26.325)
that's so great. Good. Well, we won't get into flying and things like that. Let's talk about before we do before we get into data and big data and AI, I do want to tell you a joke. All you have to do is give me a rating one through five. So here we go. I just read a book entitled How to Survive Falling Down a Staircase. It's a step by step guide published last fall.
Rochelle Thielen (01:30.743)
Yes.
Rochelle Thielen (01:42.35)
Let's do it.
Rochelle Thielen (01:53.518)
Okay, rating. I already just have to tell you that the higher the dad joke, the higher the score. So I think that one's right up there at a four out of five for me. It was a growner, but it had some sophistication too. So we're going to go with.
Brent Peterson (02:09.363)
Yeah, thank you. had, there's a couple of little points in there that were, that had, that, you know, somebody could get confused about. So I had to make sure I got that all in.
Rochelle Thielen (02:18.03)
Wonderful. Thanks for kicking it off with a great growner. I'm going to bring that home today.
Brent Peterson (02:25.255)
Okay, good. All right, so let's talk about, let's do, in the green room, we talked about the data and how people are using it and the amount of data that's out there and the amount of data that's being just scraped. Tell us a little bit about your role and how are you using that and how are you, how you see your clients using it.
Rochelle Thielen (02:42.602)
Yeah, absolutely. think it's really interesting because whether we're going to conferences, reading publications, listening to podcasts like this, there's so much talk right now around AI.
things like this. But what isn't, I think, focused on as much, because frankly, it can be overwhelming and intimidating, is the amount of data and where all this data is being pulled in order to make these solutions what they are. And when we're talking about quantity, I mean, we're fortunate here that we happen to work with enterprise-specific users that need large, large quantities, the top retailers in the US and frankly, worldwide.
When I say large quantity, this is in certain cases, 100 million data points per day. So on a typical range, it's usually probably 25 million plus per day. So when you're thinking about scraping data, which is just like this term that people are uncomfortable around sometimes, it's something that in this AI space, people have had to get very comfortable and learn about very quickly in order to do what they're aiming to do.
Brent Peterson (03:55.443)
I know that when people think of AI, they still think of generative AI, they think of chat GPT, but really the magic in AI is how it uses that big data and how it finds patterns within that big data. Tell us a little bit about how companies are using it, both their own data, but also their competitors' data.
Rochelle Thielen (04:13.868)
Yeah, a great, you know, usually a great way to look at this stuff is examples. So we hear about a lot about some things like, let's just take, for example, dynamic pricing that has become so competitive. In order to do things like that, you not only need to be aware of your own catalog of SKUs or products, but you have to be basically instantaneously going out there and looking at all of your comps.
as well because these things are so fast to change. So in an example like this, you need to in an omni-channel fashion without a human being involved to be able to react and adjust and play within the rules you've set. So you can just imagine the type of systems that behind the scenes are able to both play within their own rules, look at their own
items and react to different things happening on the web all around you. And there's examples of this in chatbots. I already mentioned dynamic pricing, customer journeys. It's just the way everything is headed right now.
Brent Peterson (05:24.915)
How important is the human involvement in that process? And I do think about in the past, there has been scrapers that would find pricing on somebody's site, then automatically update your own site based on some rules. I guess the warning would be if you are allowing AI to do everything, theoretically, you could start having prices that are 1 % above margin or something like that.
tell us the importance of having a human that's involved in that.
Rochelle Thielen (05:54.894)
Yeah, very important. again, just to reference what's going on today, even in best practice cases, humans are still controlling what the AI is doing. So an example of that would be, take a case of 100 million data points being pulled. There is in every case that I've seen, humans that are going through the data at some quantity. Call it maybe 15,000, 30,000 data
just to check that the data that they're using to train these models is what they think it is and is accurate. So that's one example is it's that whole trust but verify approach towards AI. And I think second, where just human involvement is so necessary in AI to begin with is really in an oversight nature. So AI is programmed, trained, and taught to do things through patterns, like you mentioned.
There are humans that are adjusting constantly how they're reacting and what patterns they're trained to behave and choices that they're given. So I would not say we're at a free for all stage where in any way AI is just making these decisions without being constantly monitored and told what to do.
Brent Peterson (07:17.535)
How important is it to make sure that, well, I guess two parts of this. Number one, I'm assuming everybody's on a private model for LLM. And then how important is it to make sure that you're using something that's more updated in terms of where you're getting the model from and when the last time it was used?
Rochelle Thielen (07:43.404)
Yeah, so really when it comes down to it, speed, scalability, accuracy, reliability are the cornerstones of making sure that your AI is functioning the way you expect it to. If you, at this point, are more than 24 hours old on data, to do the things we're talking about with pricing, with really getting to a level that's practically psychic level, understanding of what customers are expecting to see.
when they visit a site, you're pretty far behind. I use 24 hours as kind of like the limit, but really it should be so much quicker up to real time is really if you're in the top tier playing in this space, that's where you need to be.
Brent Peterson (08:30.035)
Do customers still expect some, we call it hallucinations in terms of what it comes back? And I have seen some LLMs just do math wrong. And then they come back and say, you're right. did, you know, four plus four is not nine. It's eight. Thank you for pointing that out.
Rochelle Thielen (08:48.398)
Customers do not expect this, by the way, especially when they're buying in retail. I think there's a sense of fatigue as well. When you're a consumer, mean, I can just put myself in this position I'm sure you can too, Brent, is like, if you go to a site looking for something and you're presented with a hallucination here where I'm looking for, say,
sunscreen for my next vacation and I start getting all sorts of products meant for my husband and skincare or something like this. People have fatigue and they'll just leave and you lose them that quickly. do consumers expect this? Absolutely not. They don't tolerate it. They don't, you know, in many cases understand or care what goes behind the scenes in order to make those experiences happen. Nor should they. If we're putting these type of products out,
It all comes back to, again, that data source that we're using and the rules we're building. And if it isn't up yet for prime time, it shouldn't be there.
Brent Peterson (09:50.239)
Do you see the use of some of those models, especially in personalization, a hindrance if the data is incorrect? Like it can be worse and it would be better.
Rochelle Thielen (09:56.706)
Yes.
Rochelle Thielen (10:02.99)
100%. The example that I just gave is a great one. The other piece in the marketing side of retail too, is marketers are having to think now conversationally, right? So when they're looking for things like SEO and keywords, when we're translating that over to the personalization of those systems, we have to now think about, I used a sunscreen example earlier, is like in order to find consumers like that, we have to be looking for terms like,
going on vacation or thing, you know, what things do I need to bring with me to Hawaii or things like that nature? And if we're not, you know, reacting to the way consumers personalize their own conversations and the way that they're buying things, we're going the wrong direction, right? We're better off sticking with original keywords before we get into that type of technology and educating ourselves so we don't end up having a reverse effect.
Brent Peterson (11:02.687)
You started off talking a little bit about chat bots, and I know that it is a great use of AI to use a chat bot to kind of prefront a customer coming to your site. How important is transparency in saying I'm a chat bot as opposed to, I'm real agent from a customer point of view? And I'll just say from my experience, I've now experienced chat bots on LinkedIn.
Rochelle Thielen (11:22.86)
Yeah.
Brent Peterson (11:31.027)
where the person insists they're a real person. And I've said, well, I don't think you can give me that much information in a matter of 15 seconds. And then I go on to say, which is the biggest Hawaiian island? And they tell me.
Rochelle Thielen (11:45.998)
I love it. I have the same reaction as you do is like, I'm going to break it. Let's just see. Let's just try to break this. Right. And I don't think we're the only ones out there that feel that way. you know, I don't obviously there's different opinions on this. My strong opinion is you need to be transparent. Right. When people I don't think when it comes down to it, people really will.
negatively look at interfacing with a bot as long as the bot is equal to or better than the human. The frustration with bots is this earlier technology and the pain that we've all gone through and feeling stuck and not being able to get what we perceive we could get better from a human. So if you're really holding yourself to that standard and only releasing products that are stepping up from what you could get before,
That's not a bad thing. That's not something to hide, right? It's like you said, if you're actually speaking with a bot that can answer those type of questions and is this nimble, what's wrong with that?
Brent Peterson (12:55.261)
Yeah, I suppose, but if you just start changing around and saying, I have a chili recipe? And they give it to you, right? There should be some kind of like filter that says, you know, we're not really talking about chili right now.
Rochelle Thielen (13:01.038)
Yeah.
Yeah.
Rochelle Thielen (13:09.354)
Yeah, I think it depends on if you're going to buy shoes and you're asking for a chili recipe and you're somebody who, you know, maybe is just looking for the conversation and the interface and there's nothing, no judgment, there's nothing wrong with that as well, you're probably not going to get that out of the bots because they're trained for a certain use case. But if you're looking for an efficient way to buy shoes and you find a bot that's incredibly knowledgeable about every kind of shoe you're after.
That's fun. It's like a lot of AI solutions. Let's test it and see if we can break it and how much it actually does now.
Brent Peterson (13:45.747)
Yeah. So a couple of years ago, well, two years ago and chat GPT is two years old now. I had predicted that there would all, there will be retailers that are using specific chatbots that know everything about a product that would know more than a salesperson could possibly know. And you could transact through a site with a, with a bot that knows all your product attributes. Are you seeing retailers already deploy this sort of feature now?
Rochelle Thielen (14:12.28)
Yeah. So you talked about scraping earlier and that's where scraping is really important because if I'm a retailer and I already obviously have access to my catalog internally, if I'm able to constantly be scraping, like in the example of shoes, sites that are talking about new technologies in shoes, new types of soles if I'm a runner or new types of performance or even articles that are out there.
I can now bring this into even like a chat bot, like we're talking to to educate consumers as they're going through this process. That's the type of value that even humans to a certain level can't deliver. So I think that's where going outside and looking to see what information can we pull that actually supports our consumers, educates them, makes them feel more empowered as a consumer is a perfect use and why we should be excited.
this technology and how fast it's evolving.
Brent Peterson (15:14.115)
We're recording the day after Black or Cyber Monday. What sort of trends did you see and do you have enough data right now to kind of see how have things changed over last year in terms of how people are using the data and how people are using AI in the data?
Rochelle Thielen (15:30.732)
Yeah, quantity is one. So we are compiling all of the data because we usually do come out and do some reporting on some of the events that occur in this period. But the quantity of data is exponentially higher. People are using things in new ways, right? So an example of that would be like holiday hours. So if you're
a retailer that uses a service to do deliveries, for instance, and you work with stores that over the holidays, like we know, have very much fluctuating hours that aren't normal. One of our partners saved literally millions of dollars by pulling those hours real time on when places were open and closed.
because they're not paying to have a third party go out, pick up merchandise that they can't pick up because the store is closed. And also an unfulfilled customer who just ordered something and now they can't get it. So that's one kind of like side interesting case that we're seeing in just a reaction to how dynamic this period is. I think the other case is a lot about dynamic pricing. That's at a whole new level to what it used to be.
And like we were just talking about time, the real time nature of the quantity happening in just a constant flow to react and better position retailers to compete during this time and in this space. Those would be the two things like right off the bat without even digging very deep that we can tell just from conversations and requests coming in from the partners that we work with.
Brent Peterson (17:12.307)
I know a big part of your business is API access and delivering data through the APIs. Are you seeing an overwhelming amount of calls being made this year compared to last year?
Rochelle Thielen (17:24.854)
Yes, in short answer, yes. So we actually, as a company, made the decision almost two years ago now to completely rebuild our backend for scalability because we saw these quantities increasing at such a rapid pace. And because of that, people requiring this level have been attracted to working with us. And that has just kind of built on itself.
So the scale is beyond, think, I would even have expected two years ago. And it's just the nature of what's required to teach the models, the breadth and the depth that's required now for what retailers are expecting it to be able to do.
Brent Peterson (18:15.593)
How do you advise a merchant when you say, going to, like, I think as a merchant, often will say, yes, of course I want all that data, but then it becomes, there's just so much and sometimes it's hard to know where to start. you, how do you advise somebody to get going, to start using that data and making it work for them?
Rochelle Thielen (18:34.67)
This is a great question. So both of us met at Shop Talk a couple of months ago and I was floored by the size of retailers I was meeting with that had that question of just like, this is on our roadmap for 2025. We know we need to get a handle around Omni Channel. We know we need to be monitoring our brand reacting, not just reacting, but leading in this way, but we don't know how to do it. How do we do this? And I think the important thing
is to realize you don't have to go huge to start. You don't have to pay tens, hundreds of thousands of dollars to build these teams. You can actually, even in a sandbox environment, we're actually releasing very soon, be able to visualize. So even if you're not an engineer, basically, and have to dig through code, you can visualize and do a search for, say, Apple AirPods across the web.
And it will give you a very simple screenshot of whatever you're selling, what it's how it's described, what the prices are, who's selling it, even in certain cases by location inventory. And you can export that even through a CSV, right? Where you can view things in Excel or the thing that breaks obviously is just in quantity and scale of what things like Excel can handle. But if you're a small to middle retailer, and like I said, even at a large retailer,
just trying to create an ROI and a use case to get the investment to start building a team around this. These are the type of projects I love working on because there's a lot of just feel good about giving someone the minimal amount of education required to start feeling like they're in control and they can do this and they can, you know, baby step into it and don't have to be sold some huge clunky really is what happens with it level tool that they don't need at this point.
Brent Peterson (20:33.951)
I know one of the things that AI is meant to do or is promised to do is democratizing how people use the data and the people that how they access it. How have you seen smaller merchants being able to now leverage some of those tools that in the past, I mean, let's face it, AI has been around for a long time. Machine learning has been around, but it's really been at the enterprise level. Do you see it coming down to the mom and pops that are able to now access some of these?
to help them grow their businesses.
Rochelle Thielen (21:06.688)
Absolutely. One of the best examples isn't fraud. So fraud tools have come so far. We work with a lot of fraud prevention providers who will partner with mom and pops to, for instance, a great example that I just learned about a couple of weeks ago is if you're selling and you've got hot product.
Some of these partners now can actually monitor if what's been purchased by a consumer is already for sale on sites like eBay as an unauthorized seller at an incorrect price and basically holding this inventory like you know, when you're a mom and a pop, you only get so many, right? So in a worst case scenario for these retailers, if those, you know, we use shoes before, so let's just say those shoes don't sell, that now becomes a return.
that goes after this hot drop period for whatever items they're selling or the holiday periods like we're in right now and causes this chain of effects. So if I'm using a solution, an AI solution that within a matter of a minute or whatever it is for a sale to be completed during the process is able to go out there and search for some of these things and find those patterns and stop that sale from happening, that level of protection for mom and pops is irreplaceable.
And that's all the way to the most sophisticated I've seen right now, but it's regularly out there that these great companies are partnering to protect brands and get unauthorized sellers who drive down prices or, you know, just kind of like create experiences that don't encourage consumers to purchase as much of something as they should. So 100 % the stuff today is out there. It's trusted. It's, it's used in fraud.
And I think it's, I'm not thinking, I know it's having a major impact, especially, you know, for small and mom and pops that don't want to be fronting those type of costs, nor should they have to.
Brent Peterson (23:11.933)
What's your predictions for next year? We're going to get through the holidays now and we're going to get a whole bunch of new data to kind of analyze. What do you see happening now coming up next year? And I'm assuming you'll probably be at shop talk again. What do you think is going to be the big thing there?
Rochelle Thielen (23:28.802)
I don't think things are gonna change. I think they're like a lot of technology gonna work their way down. So as I mentioned before, the big, big retailers are leading in the way because they have to. They have the resources, they have the sophistication to build these omni-channel, both monitoring as well as lead the way with dynamic pricing with all these technologies. But the learnings from that are cycling down. So the conversations that we're having now,
don't require as much of this kind of leading edge, bleeding edge, we're gonna go out there and fail and figure it out mentality. Those things are being learned to a point where they can be employed now by large and mid. And even like we just gave an example of some of the smaller retailers being able to use that. So just from the conversations from the last shop talk, what I'm anticipating now going into 2025 is that a lot of these things have already been allocated for in budgets.
So we're going to see people moving through 2025 starting to release these next level and higher expectation AI tools to support that whole range of demands coming from the business.
Brent Peterson (24:43.199)
What do you think is the big growth area? Do you see its data or actual visuals where people will be scraping images to try to get more data off of images? Or do you think video where people are actually scraping and getting information out of video? What's the next big area that we're going to dive into?
Rochelle Thielen (25:00.526)
Well, it's omni-channel, right? So right now, pretty, not pretty, it's very expected that you're gonna have the basics and I would include images in that. Where we're headed in 2025, again, you can just tell from top down where things are today is into that more like TikTok and those like sort of what's on the ends, the outer ends of pulling in data.
But that's more difficult to do and also even more dynamic than what we've been historically focusing on to build these tools. So at least from our perspective, we're going to be looking at scraping farther out into those areas and being able to provide that data as part of that bigger solution because it's critical. It's not a nice to have like it was, I would say, in the
maybe 24 months, it's now an expectation to stay. This is a, I think one of the fun things of being in this space and also stressful ones is there's no choice, right? You have to be out there in the land of the unknown in order to keep up. You can't rest for even a minute because things really do change that fast. So those kind of headwinds are what we lean into and we try to get ahead of that. Cause that's what
expect from us too. But it is nerve-racking. It will give you many sleepless nights because there isn't anyone to look at and be like, we'll just do that. We have to be the ones to figure it out. We have a lot of adrenaline junkies on our product and engineer.
Brent Peterson (26:44.467)
That's awesome. So last question, what is your most popular product that you're seeing your customers use right now?
Rochelle Thielen (26:57.326)
Two things, both Rainforest, which is pulling all of our Amazon information for retailers, and then also our SERP WOW product, which is going out and pulling everything from Google Shopping and essentially everything that's available throughout multiple search engines. Which basically, answer that question, is the products that pull almost anything and everything you'd want to have from the internet.
that's where everyone's looking.
Brent Peterson (27:31.007)
That was awesome. Rochelle, as we close out the podcast, I give everybody a chance to do a shameless plug about anything they'd like. What would you like to plug today?
Rochelle Thielen (27:39.982)
Ooh, I would love to plug all the hard work that our engineering team has done over the last two years to really set us up to scale to the level that the largest enterprise organizations expect. So if you're looking to pull data at that up to 100 million, 200 million level, we would love, as I said earlier, for you to try to break what we've built. That's what we're here for.
And also just anyone who is out there like we are on this leading edge and open to figuring things out as we go. We're here for that. We love those conversations and that's what makes my days exciting. So that's gonna be what I'm gonna be plugging for sure.
Brent Peterson (28:27.527)
I like that challenge of trying to break what we've built and that's a great challenge for any developer, right? Rachelle Thielen, the CEO of Traject Data. Thank you so much for being here today.
Rochelle Thielen (28:29.516)
Yeah, let's see.
Rochelle Thielen (28:38.168)
Thank you so much, Brent.
Transcript source: Provided by creator in RSS feed: download file