Impact Pricing Podcast

#422: How Machine Learning and AI Can Help You Make Efficient Pricing Decisions with Alex Galkin

Alex Galkin is the Co-Founder and CEO of Competera, a pricing optimization SaaS company converting technologies into ready-to-use products for enterprise retailers worldwide. Alex has been a mentor at his startups for several years, and he’s a Ukrainian who left four days before the war started.

In this episode, Alex helps us understand the pricing and ML work that they do at Competera as he talks about the three models they always put into use.

Why you have to check out today’s podcast:

  • Understand why you should treat pricing as a process
  • Find out how Competera helps their clients in making pricing decisions
  • Discover three models that Competera puts into use as MLs look for reasonable factors to make a price adjustment

Pricing is a process, and you need to continuously improve as any other process in your organization.

Alex Galkin

Topics Covered:

01:04 – How Alex got into pricing

03:07 – The work that Competera does in relation to pricing, and how they do it

08:28 – Using ML-driven price recommendations: Competera creating a ‘gray box’

12:35 – Talking about the portfolio-wide pricing/portfolio optimization

14:12 – An ML example of the decoy effect

17:20 – Alex explains how their smart product segmentation works; a product bringing more people in the store

23:09 – Alex’s pricing advice

23:49 – Connect with Alex

 

Key Takeaways: 

“ML is not a human. He’s not trying to play with the price; he’s changing only if he sees the reasonable factor to make this price adjustment.” – Alex Galkin

 

People / Resources Mentioned:

Connect with Alex Galkin:

Connect with Mark Stiving:   

 

Full Interview Transcript

(Note: This transcript was created with an AI transcription service. Please forgive any transcription or grammatical errors. We probably sounded better in real life.)

Alex Galkin

Pricing is a process, and you need to continuously improve as any other process in your organization.

[Intro]

Mark Stiving

Welcome to Impact Pricing, the podcast where we discuss pricing, value, and the cross relationship between them. I’m Mark Stiving, and our guest today is Alex Galkin. Here are three things you’d want to know about Alex before we start: He is the Co-Founder and CEO of Competera; he’s been a mentor at his startups for several years; and he’s Ukrainian who left four days before the war started. Thank goodness. Welcome, Alex.

Alex Galkin

Thank you, Mark. Thank you for having me here today.

Mark Stiving

Alright, and thanks for being out. I’m sure everybody wants to know that story, but we’re going to talk about pricing instead. So Alex, how did you get into pricing?

Alex Galkin

This was a really interesting story. I used to work in Deloitte where we’re doing a lot of the consultancy, building Excels for multiple brands and retailers around the globe. And this was built on top of econometrics. Econometrics is a part of math statistics. Technically, it was just a launch Excel formula. But my original DNA was on the engineering side. My diploma in my university was about electronic nodes. I used to work with neural networks back in 2002 and 2003 where there’s only first libraries available for the public use. And my engineering background really struggle every week and days where we continue working at Excels. And honestly, one day, I [got] back to my boss and asked him like, “Look bro, I have enough vision how we can make this really work on steroids really quick, fast and repeatable, not using these excels anymore again. And of course, very powerful for the customers who can use it daily.” And honestly, he told me, “Alex, you’re just killing our business because we’re designed to charge for hours, not for subscriptions.” And this is how Competera’s story started.

Mark Stiving

Oh my God, isn’t that a fabulous answer? Because if you think about Deloitte, they really are; they’re trying to sell consulting hours. And so, if you build a solution to a problem that people don’t need consulting hours, that does kind of ruin their business.

Alex Galkin

Exactly, yeah. I don’t know.

Mark Stiving

I was going to say on the other hand, having a software solution is much more scalable than selling hours.

Alex Galkin

After the years… by the way, Deloitte was my customer for four years. After that, they realized this point and start. Actually, right now, the world has changed since like 2005, right? And they start combining the products and consultancy. The consultants should lead the direction, etc., and some boring work, but we can use the ML and AI to help us to calculate. That should be done by machines; this is obvious right now.

Mark Stiving

Yeah, nice. So, what is it that Competera does?

Alex Galkin

Technically, we actually realize the vision of that exact story that way, with a simple mission like help retailers to find optimal price and keep as long as possible. This is what we are doing, like still the same vision.

Mark Stiving

Okay, and how do you do that? What data are you collecting? And besides price, what kinds of decisions are you helping people make?

Alex Galkin

Actually, I don’t try to build like army knife to solve any problem in retail. We really dedicate it for pricing, and we, of course, integrate a lot of internal data like sales transactions, store locations, stock information, etc., and combining with the external data set. What’s actually really cool on our site is like a seasonality, local events; for example, knowing when is the day of the mayor day in a small city in United States. And combining all those data, we’re also crawling the data for, like, over the last eight years. We crawled all the product-related information, like price, stock, availability. And if you can imagine, it’s like a huge data lake where all this data is sitting, and when retail comes to us, we give them those insights and understanding of up to 20 factors, how they impact their sales historically, and using those factors in the future, help them to optimize prices on daily basis in this case.

Mark Stiving

Yes. I’m still trying to get my arms around this. You are collecting your own data and selling information and decision process to retailers. Is that correct?

Alex Galkin

Yeah. You can imagine computers like Excel on steroids.

Mark Stiving

Yeah, but that still doesn’t answer my question, right? So, are you scraping Best Buy’s website to find out these are the prices of all of Best Buy’s equipment and then we go out and we scrape other companies’ websites to see what the same prices are for those products, and then we’re looking at who has how much in stock. Is that what’s going on?

Alex Galkin

Yeah, competitive data is just a part of the data source. In total is it’s 20 factors. We’re taking into the account where you should go, and the competitor is just one of them. Yeah, this is a good example. We track prices on the Best Buy, Walmart, Amazon, Google Shopping even, but all those data set is just one factor, like how competitors impact your prices. But we have also another 19, I can say, but actually is more than 19. And this is all the factors we’re combining together to find the optimal price position.

Mark Stiving

Got it.

Alex Galkin

Hope this is clear.

Mark Stiving

Yeah, it’s more clear. And so, you’ve got your own huge database with 20+ different factors that says here’s information that we can gather. And now, how do you tell? Let’s just pick a customer. Are you telling Best Buy how much to charge for this specific product?

Alex Galkin

Exactly. The best way how you can adopt machine learning is not ChatGPT, but for some cases, it’s really fun to use. But we’re using ML. It’s a sub difference, for your audience, between AI and machine learning. But we don’t want to be too deep here. Machine learning is, what I mean here, this algorithm learned historically from decision what you made and sales what you made historically in the context, for example, what price of Best Buy for that day and see how this impact your sales. Like we call the sales impacting factors, it’s like 20 of them. For example, you have some promotion in the store. Also, this store’s selling somewhere door to door to your direct competitor or some in the middle of nowhere. All those information, algorithm consume historically. We upload two years history of this data from the retailer and then match with this huge data lake what we already know and base it on all those relations, including the relations between different items. For example, in a grocery case, different flavors of the yogurts impacting sales of each other, and algorithm learning those relations; we call it cross product impact or cannibalization. And taking this into account, generating you price recommendation. But then, the manager needs to decide if he wants to apply it or decline. And we try, of course, to explain why those prices should be changed. Because ML is not a human. He’s not trying to play with the price; he’s changing only if he sees the reasonable factor to make this price adjustment.

Mark Stiving

Yes. One of the weird things about AI or ML is, it is like a black box; I don’t know. There’s no algorithm that you wrote that said “If this happens, then go do this.” It’s really “We’ve watched the world; we’ve said this is the way the world tends to behave. So now, this is what we think is going to happen when you make this action.” And so, if I were running stores, what I would certainly ask is “Let’s do a test. Let’s use your suggestions in this store and let’s use an expert in this other store over here and see how well they perform.” Now, I assume that you’ve done this in the past.

Alex Galkin

Yes, Mark.

Mark Stiving

What kind of results have we seen?

Alex Galkin

So cool that you asked me this because very often, actually, just imagine we start selling this in 2018 and most of the owners of the retailer stores say, “Alex is a black box. We’re not comfortable. We don’t trust any algorithm. We even sometimes don’t trust Excel who is straightforward,” and they’re absolutely right. But if you make the AI from black box into white box, it would be not AI anymore. It’s why we tell them to say like we’re trying to create a gray box. It means that each price recommendation really has the explanation. For example, we can show you which important factors model take into account. For example, lifecycle of this item or competitor pressure. We can show you what main factors drives to this price suggestion. And you’re absolutely right; in most cases, we’re telling like “Hey guys, take our car, show your kids and family. If you don’t like it, you can give it back.” We give them test drives. And in average, for our last year, we give 6.8 gross profit lift for our accounts without losing any revenue. Who can tell you it can’t be more? Probably this is not true. That’s an average. Of course, we have smaller customers getting more like 10-15 percent, but the average is around 7% gross profit bottom line, except for grocery, a bit lower, like maybe four to five.

Mark Stiving

Yeah, but even in grocery, because their margins are so low anyway, that’s probably a huge impact on profitability.

Alex Galkin

That’s huge.

Mark Stiving

Yeah, exactly.

Alex Galkin

Absolutely huge. I just had a call early this week and there’s just a customer to call me like, “Alex, we already built the stores in any possible city and we don’t have any space anymore to grow just by building the new stores. And we need to right now find a new way.” And this is a great way to do it, right?

Mark Stiving

Right, because you’re going to improve in store revenue; certainly in store profitability, if not revenue.

Alex Galkin

Yup. I can tell you it’s very extensive, this question. Very often, the retail needs everything, like give me more revenue, give me gross profit. And when they tune the algorithm, we never sell you like a fish. We give you the fish and stick and say, it’s a fine effect from this AB test. Retail wants to increase their revenue, and algorithm is not a big name. We call this pricing follower; from the pricing structure that is like someone is a follower. And if you’re a follower, you can’t actually drive the market because you’re a follower. Just guess what the algo’s doing that way. Algo, of course, they’re comparing how their internal pricing tool like just follow competitor fight against the ML-driven price recommendations, and guess what happens? ML-driven price recommendations also suggest to go to buy competitors because it’s the most obvious way if you want to improve the revenue. And the customer was struggling and rising the questions, “Alex, why are we buying this software if I can do the same in Excel?” And for me, it’s a beautiful answer. If AI is capable to recognize that, for you, it’s the best strategy to go the competitor; it’s a great algo. In this case, algo sees the sum opportunities to uplift some prices on the less sensitive items that give them gross profit. It was flat on revenue, it was back-to-back zero effect on revenue, but was positive effect on gross profit. And customer was not happy. He’s like, “I’m buying you to increase my revenue. You’re not increasing revenue; you’re increasing gross profit.” It sometimes happens like that.

Mark Stiving

Yeah. When I think about pricing in a retail setting, I often think about how we have to manage categories of products. And so, you brought up yogurt so maybe, I’ve got multiple flavors of yogurts and multiple styles of yogurts and multiple sizes of yogurts, and yogurt is different from ketchup. And so, it’s almost like I need a category analytics as opposed to store-wide analytics. Do you program that in or does the machine learning just learn that?

Alex Galkin

Cool. Nice question. I call this a portfolio-wide pricing; it’s another very great output of this because computers are capable to catch those relations, cross product groups and relation, and we call this portfolio. We create a kind of optimization groups. For example, if you’re selling TVs, when you’re selling HDMI cables, no one come to your store from a big consumer electronic store to buy HDMI cables. Or that happens maybe zero point something person fails, it’s just a guy come to HDMI cable because most often you go Amazon, right? This is what AI and ML are capable to grab, and we call this portfolio optimization when we know how actually sometimes very not direct correlations. For example, from the algorithm perspective, consumption of toothbrush is a similar vs. a toilet paper. And this, we call it cross product dependency or impact. This is a short answer, of course. This is why this is much, more more advanced from the human brain. Honestly, as a consultant, we also do this exercise in defining those groups, what can be priced together less sensitive, and it was very manually, honestly done maybe once a year. Right now, it can be done daily. Like all those relations is already there and we trained on each repricing cycle because you’re adding new items and you discontinue some items continuously.

Mark Stiving

Yeah, I love the HDMI cable example. Have you ever seen a cross demand that was unusual and it was like, “Oh, now that makes sense.”

Alex Galkin

I can show you again another example. For example, retailers also, they’re very KPI driven and sometimes the motivation of the people and they have their own private labels. And what does it mean, private label? For a wider audience, some items what really looks like a popular item but actually made it by this particular retailer. It could be napkins or it’s same ketchup, for example. But people love to continue buying some well-known brands, right? And you can’t replace even if you want and you can’t put it cheaper. And what happens from this not obvious relations that we saw with the case, I can’t say the name of the customer, of course, but it was their private label. And historically it is expensive; this was TVs actually. This is like quite TV, and their private label was very low, cheap, like $100 and the last TVs, like very cheap, and they’ve never sold. There’s always 10, five left on the shelf and they can’t sell them. But KPI for managers always say “You need to turn all those items”. And guess what machine learning started doing? They started increasing the price of this private label. And because she saw it– anyway, he will not sell this TV because it’s probably something else. It’s not a price. Because they also recognize that it’s not only not pricing factors, it’s also the factors. And this model, using the shadowing techniques, like increasing the price of the TV was never sold and increasing the probability that we’ll never sold again, but they drive the sales of the items with staying very close to this item. And it’s a really nice example. And then, of course, the customers telling us “Guys, what is it doing? You see increasing price. I still have a lot of stock of these items.” They say, “Guys, look on the portfolio level, you winning. On the portfolio level, in this A group, you have a much better result and this model just decided to give up one item just to drive the overall sales or other items, because they’re not biased.”

Mark Stiving

So, what you just described was an ML example of the decoy effect. And so, as we make the decision easier for a customer, so we take a bad product, we make the price really close to a good product, a customer looks at those and says “Oh, this is an easy decision. I want the good product because it’s almost the same price.” And so, people make decisions when it’s easier to make decisions.

Alex Galkin

Yeah.

Mark Stiving

And what’s cool is that you found that with machine learning without knowing, “Oh, this is the decoy effect” or “this is the thing that we’re trying to get accomplished.” So that’s pretty neat; pretty interesting.

Another hard question—in the world of retail, one of the big decisions is what products bring my customers to my store, and then what products do I sell once they’re in the store. When we think about the products that bring someone to the store, usually, we’ll advertise those or their prices that people remember on a regular milk at the grocery store, something like that. How do you guys think about that and what do you do?

Alex Galkin

That’s great question, Mark. Thank you. First of all, we call that smart product segmentation. Like the KVI traffic generators, loan tail, tail items. We have a few methodologies. It’s actually a bit different algo work for that; honestly have nothing similar with the pricing algo. But this algo really technically just mark those products and helping another network to use this as extra features to the parameters. Like, this is a KVI item, this is a loan tail, this is a tail item to be more picky and more confident on the decision what another model made on the pricing. Honestly, the three different embedded models works together. First layer model learn just factors with impacting the sales and just learn the impact historical, this day, deltas, et cetera. Another model really working on those labeling of the items on their roles. We’re actually not touching, like we don’t answer your question what items you need to bring to the store? You will not touch assortment planning; it’s like different space and different software. But what we can say is what actually products cannibalize the sales and you have very bad margin that there’s no sense for having those items. Because sometimes, Mark, you manage very right some items even not bringing your attention, but they cannibalize sales of the good marginal items and you have very bad relations with the vendor or cost structure for those items and if you will cut them off from the assortment you will get better result on the portfolio even don’t having these items. This is something what we can show you. But in the third layer model, just using those learnings and try to yield as highest as possible result on target function what the manager select. I won’t increase revenue, hold my margin or don’t hold nothing, increase my sales items, whatever. This is like robots learning from robots in this particular case. We’re able to mark the role of the item; we call it smart segmentation—a smart segment—and using this as factors. Sometimes, honestly, customers really rely on that. Because for example, if you buy Coca Cola, a Coca Cola manager selling like “This is a best seller. I don’t know this tea. This is the bestseller.” And they were right from their angle of you. But you’re a retailer, you have a shelf and your customers need to consider several pricing tiers, different brands, you want to see the selection. And in this case, even different products, they’re telling you—-sorry, where are you based right now?

Mark Stiving

I live in Reno, Nevada.

Alex Galkin

Nevada. I am in Manhattan. And we probably, like in the grocery store, for me, KVIs for you and me would be different. Even people, like country-wise, this is definitely Coca Cola #1, but maybe local-wise, it’s a different KVI. And this is what algorithm is capable to pick up and tell you, like “In this is store or this is location or this cluster, this is a different KVIs”, and we call it smart segments of the items.

Mark Stiving

Yeah. If I think about not necessarily the algorithm but the impact of what I would expect to see, I would say that if you’ve got a product that brings people into the store and I lower the price, I’m going to sell more. So, it’s going to be very price sensitive and I could notice that in an algorithm. On the other hand, if I’m Best Buy and I lower the price of an HDMI cable, it’s not going to bring more people into the store and I’m probably not going to sell more. So, it’s probably not very price sensitive and that could be a good indicator of whether or not this is something that’s bringing people into the store. Is there something else that you could think of that we can measure and say “Hey, this is a good indicator of this product brings people into the store”?

Alex Galkin

Yeah, absolutely. Honestly, we just made it through ad hoc analysis and retailers would not even see this in size. They were not able to make the change in decision because this was an example with a banana. This a small like a grocery store, maybe 300-400 products only in the store. It was like London chain, can’t mention the name, and the people coming there, it’s very good quality, fresh, like sandwiches or something, and of course, bananas and cacao coffee is the main driver of the traffic. And we see like algorithm-wise that if you drop the price from banana from like £1 to 0.75, you can increase the traffic because people love coming just to buy banana, right? They will grab the sandwich or something else as well. And it really works from one side, but you know what happens? First of all, right to drop the price banana and show you that we will sell this banana cheaper than you buy it, like really hard decision, and really, sometimes really hard to push it forward internally, like algorithmically-wise, you know it, but there’s a lot of other business limitations that allows you to do so. This is one blocker what I have received from my experience. And the second one, honestly, it was another example with a fresh fish, and they create so much traffic, and they put this billboard to the stores that actually broke the store at all and people just coming and buying only this fresh fish and buying nothing else, and just they’re selling for nothing. It should be like a quiet balance. But if I can answer very simple on your answer, yeah, you can investigate this with our help or help of any other algorithm and they’re capable to highlight you those products and you can advertising, but you need to be a bit like accurate here because you can overdrive traffic and just kill the whole idea.

Mark Stiving

Nice. Alex, we’re going to have to start wrapping this up, but let me ask you the final question. What is one piece of pricing advice you’d give our listeners that you think could have a big impact on their business?

Alex Galkin

Treat the pricing as a process. This is something that’s continuously changing, and we are in the competitor, we also do pricing, but we do our pricing, and the pricing is a process. And you need to continuously improve, as any other process in your organization, and investing in pricing is always a great place to get extra gross profit for any business, not only for retailers.

Mark Stiving

Yeah, fabulous answer, because pricing is so powerful and so many companies don’t understand it, don’t use it. So if you create a process, you work it, you will definitely get better and make more profit. Excellent.

Alex, thanks so much for your time today. If anybody wants to contact you, how can they do that?

Alex Galkin

It’s a good question. They can use my email. I don’t know, like in a podcast. It’s [email protected]. That’s always open. Always happy to speak with the great pricing experts. And we actually lead the pricing community in LinkedIn. There’s a retail pricing community. Please join our community. There’s a lot of great pricing professionals there. We share a lot of information, including our books, a lot of free materials. We try to really invest to get the pricing stronger and make more and more pricing professionals in this world.

Mark Stiving

Awesome. Thanks Alex. And to our listeners, thank you for your time. If you enjoyed this, would you please leave us a rating and A review? You can get instructions by going to www.ratethispodcast.com/impactpricing. And finally, if you have any questions or comments about this podcast or pricing in general, feel free to email me, [email protected]

Now, go make an impact.

 

Tags: Accelerate Your Subscription Business, ask a pricing expert, pricing metrics, pricing strategy

Related Podcasts

EXCLUSIVE WEBINAR

Pricing Best Practices:
How Private Equity Can Drive Value Without Compromising Relationships

Don't miss out on this opportunity to enhance your pricing approach and drive increased value.

Our Speakers

Mark Stiving, Ph.D.

CEO at Impact Pricing

Alexis Underwood

Managing Director at Wynnchurch Capital, L.P.

Stephen Plume

Managing Director of
The Entrepreneurs' Fund