SignalFlare.ai’s Lukianoff on Data Analytics, AI

Flare dot ai. Signal Flare is a decision intelligence platform for restaurants and other industries that translates data into decision frameworks using AI and machine learning models. The company was the twenty twenty four Snowflake Startup Challenge winner. Welcome back to the podcast.

a lot since then. So and this is you know, that was a few a few companies ago, and you know the last one I uh, I sold and exited. And then you know, I took some time off, you know, around twenty twenty, and turned out that some other people took some time off around them too, And as I started to analyze data, because people were still coming to

me to say, you know, can you analyze this? And so forth, and a lot of the old methods, you know, when demand patterns just completely change like they did during that time, and even trade areas completely changed as people changed working from home and you know, and so you know, I had to come at a lot of these problems from a completely different in a completely different way. You know.

So instead of using a lot of these linear models that look at trends over time, say, okay, well, how do I change both the statistical approaches to more of

a machine learning approach but also the data sets. Right, how can I instead of just starting with the point of sale data, how do I pull data from mobile you know, mobile digital platforms, right, Uh, you know, cell phone devices to understand you know where people are shopping when they're shopping there, and you know who they are right going into every restaurant and every retailer, and you know, and then you know, bringing in local economic data and uh,

credit card panels, just masses of external data you that tells us more things about restaurant consumers before I start layering and the point of sale data. Uh. And what I found was two things, right, One is that it builds an infinitely more robust model. And in two, instead of always looking in the rear view mirror, you can

start to layer in things that are forward looking. So for instance, when you know when gas prices started to spike, right before you know the inflation like really went nuts. We were able to take that information because we know, you know, which restaurants are more dependent on people who drive more. We can get all of that from from the data, and we know and we can calculate what people's disposable income is, not just uh you know, not just you know what the areas, uh you know income

levels are. Uh. So when we can factor that in and we know that some of these things are leading indicators, uh, then we can use that data to say, you know what these restaurants are the ones they are going to be having trouble in you know, three to five months. So what kind of strategies to you employee? Right? Do you do you do discounting? Right? Or do you send different messaging? But when you start to use it in that framework, then you know, then it becomes uh, you know,

planning tools, not just you know, backward looking analysis. So so so long, long story short. I did that for a while and really moved it out with with some chains more on a consulting basis, and about two years ago I said, you know what this is. This this is a platform, right, this is this is a much different thing than what I've done in the past in terms of being able to make it, you know, scalable and accessible to like really any size restaurant. Yeah, it's really cool.

of sale data and so forth. But now we know what's actually going on, you know, in the surrounding community, right we know, you know how people are spending and what the trends are, and you know what segments they're spending and how much they're spending and all of that. So uh and we're putting uh, you know, an AI interface over it that also leverages all of the major major models. So you know, by doing that, we're able to cross section our machine learning models, you know, with

the broader market understanding. And you know, you know, before I was a in restaurant data and tech. I was. I was restaurant operator. So the idea that we can you know, go from you know, early days working only with the very largest mega chains to now being able to provide similar things all the way down to small chains and independence and that that excites me, right, not just for you know, for the technology, but for what we were able to do for the industry.

the starting point. But the way that I was building it, you know, in order to build a really robust pricing model. You need to not just understand, you know, what people are paying for the price, and you know, it's not just about competitive benchmarking. That's that's a part of it, but you really need to understand what the fundamental demand is around each restaurant so that you're not just saying, you know, okay, how is this trending or how do

I compare against them? But how do you layer everything from you know, sentiment to macroeconomic, microeconomic, you know, competitor performance data. How do you take all of those things? And once you bring all those things in, then the kinds of questions that you can answer go way beyond pricing, right, it goes to everything about you know, how do I optimize my market? Right? How do I look into new markets?

You know, how do I start planning budget planning? And how do I do scenario analysis, you know, to to be able to react or to prepare for contingencies you know, in the competitive market or in the macroeconomic landscape. I think, you know, what I've been doing my whole career really has been what I call that last mile of turning

the data into the decision. But in the past, you know that had to really happen by bringing insights and visuals and summary data to a really small our analysts, and all of that is is changing now because you know, when you can bring that kind of summary data in and have it read by an l M or or or an analyst agent, it need to be trained, right. It's still like having you know, an intern, you need

to you need to bring along and teach. But it's moving so quickly that I'm just every day I get more surprised and impressed and excited about what those you know, what those possibilities are. Yeah, that's cool.

and so forth. But you really the standard in the bar for data, you know, cleanliness and context and and usability is higher, not lower, right, because what an ll M will do, right, what these you know, you know, the chat GPTs and the clauds and all these they will give you an answer that sounds wonderful, right, even if there's no data to support it. They they you know, what's amazing about them, right, is that they can get creative, right,

as creative as a person. So you know, the data might not be there right, or it might interpret it completely wrong, but the pros sounds wonderful, right. And if you've got you know, crappy data that's not structured, and you know, in fact, really what they're best at is reading text documents. You know, if all of your data is in a data warehouse, you need another layer over that to say, okay, well how do I summarize this

and how do I chunk it out? And how do I make sure that it's in usable nuggets, so that then the machine can can can interpret it. You know, people are using you know, these for calculations, but they're not good at calculating. You know. Maybe it's a unpopular you know fact, right, I don't know popular opinion. It's a fact. They can invoke tools, right that can help to do a calculation. But when it comes down to it, right, I think you know the stats that I put out there.

You know, look, you know eighty percent of of AI projects are still failing, and seventy five percent of those are failing because they didn't get the data, right.

which is fascinating. But yeah, I mean hallucinations is just you know, a fancy way of you know, of saying you know, it just it's just b s's right, their language models, right, they read you know everything that's ever been written, and then they figure out how to you know, you ask a question and then it's you know, searching this repository to try and say it in the way

or the style that you specified. So you know, you know, and you know an example of like you know, bad mathroom an l M. Right, it's like you know, we we we're constantly doing you know, tests and validations, but you you don't know what the answer is that you're going to get because it's all probabilistic, right. What you know, what we do in analytics is deterministic. Right. You need a number, right that you've got some level of certainty and there's this and there's a clear calculation to it.

You know. L l M s are all you know, Hey, if you ask this question, then the probability of this, this, this and that right ur X and y and that's what all these sort of computational nodes are doing as it's doing the calculation. So, uh, you know, so if you telling it, you know, an l l M without any more instructions, you know, hey, tell me what the year of a year sales are for you know, X, Y and Z company. You know twenty you know this

year versus last year. The most likely answer you're going to get out of that is going to be something that says, oh, well, you're down about you know, fifty percent, you know versus last year. Why because we're about halfway through the year, right, So it's comparing this year, you know, the last year without any more you know, clarity than that.

So as you're either figuring out how to prompt it and ask the questions correctly, or you're trying to figure out how to do the engineering in the background, you know, the things that the ways that we talk about things and the common vernacular that we use, you know in this industry, as you know, it's like, okay, well, what's a check average? You know, as fifteen different operators, what you know what that means to them? Right? How are

you calculating that check average? It's all right, Well, you kind of need a single standard, right, or a very clear definition of all of these things in order for them to work and to learn.

So it's addictive very quickly, right because people or like I can do a better job with it. You know, it's one of the best productivity tools you know, invented probably since you know, since Lotus one, two three, right precursor to to Excel for those of who don't know. So, if you don't have a policy for enterprise level, which typically right the the g p T, the l M will say that in their enterprise versions, you know, none

of your data is used to train their models. There's another way of using it, which is, if you use the A p I S in your own environment, then you know, there's no backward propagation, right because it's it's being calculated in yours and you're using you know, you're getting the benefit of it, but there's no way that it can actually backward propagate, which is, you know, which

is the better the better process. But if you have no policy, right and you're not, you know, and then if people are taking your company data and just using it on a free version or a personal version, all of that information is being you know, you're you're you're donating to the you know, to the broader knowledge base of the world right with whatever you put in there.

So so yes, you need to be looking at enterprise grade tools to be able to use this, but don't fool yourself that, you know, if you're not using it, then uh, you know that then you're safe. You know, if you don't have a policy in place, then you're probably in more security danger than than if you have one and you're actually teaching people the right way to use it.

what we're seeing, right. We're seeing that it's you know, it's you know, it's a lot of the fast casuals and so forth that are doing a lot better, especially if the cuisine that they're serving up is not yet you know, at a saturation point and in a lot

of their markets. Right, I think what the consumer is telling us is that, you know, they are making choices, but their choices are being guided not by what's the least expensive thing that I can get, but what actually represents the value, Right, what's going to what's going to give me better service, what's going to be the taste profile that I want, because if I'm going to dine out, you know, maybe less often, I want to make sure that those occasions that I am dining out, you know,

I'm actually getting what I want.

in a market, in a slower restaurant spending market. I mean, I don't think I've never seen it, you know, since I've been in the business.

You know, you had you know, there was you know, there were a few businesses that were as good as you knows as having a drive through right in an underserved market, right. And what ended up happening is that unit account started to grow, but it was all limited service.

So the ratio of limited service to full service restaurants is completely different than it was back in twenty eighteen, twenty nineteen, big overbuilding of you know, of limited service restaurants, not not as much, you know, catching up with the full service. So you know, and yeah, and the way that the trade areas you know change, right. You know, people sort of you know move to more you know, more rural where they could because their workplaces would do it.

Maybe they had to move a little bit closer, but you know, not so much into city centers. So all of those factors together, right, are showing these traditional metrics of same store sales as opposed to total industry growth right or total industry usage. They are two very different looks at the industry, right, because the the saturation of markets, especially when you look at different cuisine types and uh

service styles. Because population growth is also, you know, is not near what it was five years ago either, So rest the restaurant build build out is outpacing population growth by about you know, about five x.

of time, all of those just just vanished. Right. That's starting to come back, But they sort of retrained customers, right, They started, you know, started to find things that were you know, oh well, I can get better food you know all the time you know, at these different types of uh, you know, restaurants, you know, also limited service you know. But I but I like it better, right, So it really I think it created something that you know,

I mean, we'll see if people revert. But qs R every day is getting more and more like off premise grocery you know, or you know, and even the convenience stores are getting you know, getting better, you know at you know, really upping their game. So those options, right, you know, competing there where it used to be the safe the safe bet you know, you know, recession resistant, it's not really that that's not really the case anymore now that they seem to be more vulnerable, you know.

those are happening from uh international migration. Right. What we haven't seen yet in this year, which I'm i hear a lot anecdotally, is how much has the you know, tightened immigration policies affected border towns and so forth right. It is how much you know what's going on with the population base, and is that population based less likely to you know, go out and use restaurants and so

forth right. So I think that even before that we saw that, yes, you know, the restaurant growth, even after taking into account closures and bankruptcies, is still growing a lot faster than population. I think we're going to see, you know, in retrospect that community that you're in makes a big difference in terms of, you know, how saturated you are and how hard you've got to work for that, you know, that share of stomach.

end of last year, we've seen that full service price increases are are are now ahead of you know cp I for for limited service. So it'll be interesting to see, you know, how that happens. And I think. Look, it's it's consumer driven really because you know, when we compare, when we compare cp I going back you know, even three years, uh two, check averages you know from you know,

credit card sampling data and so forth. You know, we're seeing that, you know, check averages in the industry are not growing at nearly the rate of of of cp I. So what's that telling us, Right, that's telling us that restaurants are increasing their prices. But h but consumers are managing their spend, you know, to not go too far outside of what they're you know, what they were, what

they planned, right. Uh So that that's that's that I think is important because it means that restaurants that think that they can just you know, do a one time you know, three percent, three three and a half four percent increase, they're probably only you know, after your customer you know, manages their checked, they're probably only getting you know,

maybe two percent of that, right. I mean, the real ratios of just you know, across the board types of increases, you know, are about fifty percent flow through, right, Whereas if you instead said, you know what if we went lower you know, in a more measured way, you know, and hit the one and a half two percent. You know, when you're doing it that way, you're staying more within

what the consumers willing to take. You can measure it, you can maybe even do more, but you're more likely to get in the nineties, you know, than you know, and that ends up being much healthier for for the brand. Right. I think a lot of the larger companies understand that, you know, and you know, they may a lot of them even like may say, oh, we're not changing our prices, but you know what they're franchise systems, So yes, they

are increasing the prices. But the better guidance they can have in terms of you know, being moderate, right and making sure that you're measuring what the consumer is willing to spend along with it, that's how you make sure that you're growing and growing in a healthy way.

Cracker Barrel. They're also starting to lean into strategic pricing right now.

look at our check average. It's like, okay, well that that's that's not normal times in the world. Right. So it's actually on the back end of those types of inflation bubbles that it gets it becomes much harder. But costs, haven't you know, come back down to earth, especially you know in our industry where it's the cost of labor, right, this is becoming even more scarce. How do you how do you make sure that you are threading that needle right because it is you know, if you go too high,

you're going to get punished. By the consumer. But if you go too low, we're a thin margin business, right, You've got to you know, you are going to have to pass on some thing to the consumer just to make the economics work.

and middle income customers that were still pretty strong. So but what we've seen really over the over the course of this year is that, you know, while the low income customers don't necessarily come back and drove, That's not what I'm suggesting, but the mix of customer you know, is has sort of stabilized at the lowest and the

highest income. But what we're really seeing is the middle falling out, right, So you know, it's you know, sort of this trickle across and you know, so I think that that's you know, it may be contrary to what

we've been thinking about. You know, you know, strapped customers still still an important consideration, but even in like you know, fast casual, what I was surprised to see is that, you know, the the mix of lower income customers and fast casual is actually growing because the middle has fallen

out so much. That has implications, right, which means that, okay, well, if the low income customers are now making up a bigger mix of your of your customer base, then that means that you're also likely to experience more uh, you know, more more price sensitivity.

you know, consumer survey based data and so forth. Right, you know you're seeing look, and a lot depends on what party you are, right, But but even there, right, there's a there's a universal agreement that you know, things are things are very uncertain, right, And you might say, oh, well that's good, right, or maybe that's the big differentiator by by party, Right, is it good or is it bad? That is uncertain Well, it depends what the outcome is

either way. It means that you know, there are customers and companies and so forth that are you know, don't don't love the uncertainty, right, So so they're going to

delay some decisions, you know. So to your point on this in the later half of the year, if there's tax cuts and so on and so forth, you know, could that help We'll eat, yes, But it depends on who believes that, you know, who believes that that's coming, right, because it's that it's for a lot of people, it's you know, how do I feel, you know, that dictates how much I'm going to spend, right.

doing very well. But then there's also you know, some that are that are not doing so well right because of you know, where where they are and how they compare to say, you know, the more affordable full service options. So you know, again I think like the we've rarely seen things so call it trade area dependent, right, it's you know that can really dictate you know, who's looking good and who's and who's.

And that's a huge leap, right, especially for marketing departments that you know, oftentimes you know came from you know consumer uh you know, uh consumer package goods background or something where you know, these psychographic studies that are have really been effective in you know, broad based advertising, but you cannot convert you know, those psychographics into you know, behavioral data, right that you collect within you know, within

those technologies that can do the one to one. The other thing is that you know, most of the ones that are being designed that are capable of one to one marketing in restaurant industry, it's all about the It's all about the CD right, the consumer or the customer

database our data platform. So and that means that in order to communicate certain things to people at that level, they already have to be in your database, right, an identified customer who you can talk to, which is great, right, it's an effective thing has been to wave through the industry. But the reality of that is that you can't you can't grow your customer reach, you know, through the database

of people who are already your customers. Right, So how do you take that same concept and apply it to people outside of your outside of your database, you know, and say, okay, well, I know that the people who like my brand, who are heavy consumers of this and so on and so forth, they live here, right, and here's how I can get to them, because it's close enough to my stores to be able to do that. That is not that is not a common, uh usage of the technology, So they sort of revert back to the

you know, to the mass market. Even in you know, even in d m A level you know, uh, digital spending, usually it's you know, you've got to buy the whole d m A right, if you want to get very

specific about your your your demo. But really the technology exists so that you know, you can say, look, I just want people in this neighborhood, you know, to be able to see this kind of ad because I know that you know, all of them are you know, within a distance, right, and that becomes much less expensive, uh to execute. But it's it's not commonly understood. I think across you know, across the marketing spectrum, and.

expand your reach, you know. And I think that you know, one of the things and we're starting to see it, right, is you know, what can I get for five bucks or what can I get for ten bucks? Right? Because really, if we look historically through the through the industry, people think about the discount as being so important, but really, if you look at the successful campaigns over the past twenty five years, it's really been about, you know, price certainty, right,

how much am I going to spend when I go out? Right?

Is it three dollars? Is it five dollars? You know, ten dollars even you know Ruth Chris had the Rus Chris classic, Right, It's like, okay, you know eighty five dollars, right, but you knew that you could go there and that's what you're going to spend done a per person basis, right, And I think that that helps people to make these these decisions, right to say this is how much money you know I'm going to spend today, right, or this how much money it is in per per head in

my family. Right. So so I think that you know, those I think we're going to see a lot more of that, you know, hammering certain price points that consumers see appealing.

I dynamically choose the right bundle, right and try? And you know, think of think about price for my customers, not as how much they're spending on each subcomponent of their order, but how do I know, how do I get them to spend more with me in that visit?

Not because I've forced them to you know, swallow a higher price on everything, but because I did a better job of using technology to you know, to upsell the things that they actually want to add to their order, right, so that you know, instead of spending five bucks with me, they're spending seven. And that's not because I just increased,

you know, the prices by by two bucks. Right, it's because you know they got an add on, right, or they got something that actually makes them enjoy the experience more.