The Future of AI Pricing: From Human-in-the-Loop to Fully Automated Decisions with Alex Halkin

In this episode of Pricing Heroes, we speak with Alex Halkin, Founder and CEO of Competera, returning to the show two years after his first appearance in 2024. In that conversation, Alex predicted a major shift toward AI-driven, contextual pricing in retail. Today, he revisits those predictions, assesses how the space has evolved, and shares what's coming next — from autonomous pricing decisions to four-dimensional optimisation models and algorithmic trading-style speed.

Key Topics:

• How AI adoption in retail pricing has accelerated since 2024, and why CIOs are now advocating for it rather than resisting it
• Why contextual AI pricing is not for every retailer, and how to assess if it's right for your business
• How advanced AI pricing algorithms use model routing to orchestrate multiple AI models to achieve holistic pricing outcomes
• Whether AI is necessary for velocity ratio index (VRI) and lifecycle sell-through rate optimisation for seasonal inventory
• Why some overstock problems are demand problems, not pricing problems
• How macroeconomic volatility is making pricing resemble algorithmic trading

Connect with Alex Halkin on LinkedIn.

----------

Get your free copy of Get Ready for the Future Of Pricing with our A-Z Guide.

For more information about AI pricing solutions, check out our Corporate sponsor Competera.ai.

Transcript

Aaron: Hello, and welcome to Pricing Heroes, a podcast sponsored by Competera. This is a series of interviews with the best in class retail pricing experts driving bottom line metrics for major retail brands and the industry as a whole.

Today's guest is Alex Hawkin, founder and CEO of Competera. Alex is a returning guest on the show. He first joined us back in 2024 for an episode titled The Future of Retail Pricing Powered by AI. In that conversation, we explored the early shift towards AI driven pricing and the emergence of contextual multi-factor pricing models.

Today we're bringing Alex back to revisit that conversation, taking a look at how those predictions have held up, and assessing how the space has evolved since. Alex, welcome to Pricing Heroes.

Alex: Thank you Aaron.

Aaron: So, Alex, when you were last on the podcast in 2024, you said that AI pricing was just beginning to take shape.

Looking back now, has AI transformed retail pricing as much as you had predicted, or has adoption been slower than you expected?

Alex: Yeah. Actually this is so cool. Time really flies. It's two years now, right? Yeah, Aaron. Absolutely. A hundred percent. Right now I feel like there is huge demand for this. Of course, it's like a tailwind of global AI adoption and this drives the trust for the machine learning algorithms that can really make the decision.

And what I see even now is that acceleration is so hard that retailers are asking to really not even have people in the loop. The computer is designed to really give you the prediction, give you the optimization results. And asking you to review and confirm is now a plan to request.

Can we just go, and then it'll stop if something goes wrong. We don't need to — we have really tiny touch on this kind of optimization scenarios and say yes or no, but even then, customers don't even want this. The trust is tremendous. This is what I'm super happy about, and this is finally here.

Aaron: So what would you say has changed the most and what hasn't changed at all?

Alex: I feel like the first change happens on the IT side. I feel like all the CIOs right now, they are advocating for AI. Back in 2024, when we spoke with you, they were super against. They were very — I will say — conservative about getting access. Now everyone is thinking about how to get access with less destruction. This is a completely changed mindset.

Aaron: So at the time you also emphasised that it wasn't the algorithms and compute that were the biggest constraints for AI adoption. It was actually data quality. Is that still the case or have retailers started to close that gap?

Alex: Yeah, they're starting to close the gap — definitely still a big topic — but I see less and less of that sensitivity now. And some of the... probably, yeah, this is the 2024 to 2026 comparison: most of the data is already stored. Even what, for me, is impressive — mid-size companies already have a very good data foundation.

They remember everything. They at least stopped deleting the data and deleting the historical stuff, because everyone understands that they need context and the context is hidden in your historical data sets. And of course, if you ask me — graded from zero to ten — it's around maybe six to seven, progressing to seven to eight. In 2024 it was what? Three to five? This is a tremendous shift.

Aaron: Do you think it's the emergence of LLMs that's brought to the fore the importance of collecting data so that you can actually train models and begin to use these within your organisations?

Alex: I feel just in general, yeah. If you speak 2005 to 2024, actually it was huge — the conference that Google hosted, the Next conference — that was... yeah, I remember this. Everyone is laughing at all conferences about AI... and I feel yeah. This is a side topic of that. It was actually the push for data adoption. It has, in my perspective, been like two big pushes. It was companies like Snowflake — that was the first one that started growing. What Snowflake does technically is allow you to move your legacy data sets to the cloud, and use this data more accessibly. Then another wave with Databricks appeared, that gives you the ability to speak with the data and transform the data quicker amongst all the different infrastructure.

And it was happening — those companies were really growing those years. And now it's getting to the point that you finally can use this data. And I'm happy to see that retailers are asking, "Hey guys, do you even have an MCP server right now working on your app? Can you deploy on Snowflake or Databricks?" Amazing questions, because before it was like, "Hey, can you upload your CSV file to this FTP bucket?" That's most of the questions. Now we speak differently — now they're all about database connections, streamlining the data, which kind of connectors we have with the real data. That is a tremendous shift. Yeah. They're still deploying of course — I can't tell for everyone — and we also usually work with retailers who already have some level of maturity. But yeah, in general, I feel like you're right. This noise AI created was a push for CIOs. I feel like if you're a CTO or CIO of some enterprise organisation, mid-size or otherwise, you know that you are not in the business — you are out of your job — if you're not adopting AI. Even last year. And this is what I see: honestly, a lot of leadership changes, a lot of shifts. What I see from retailers is that the younger generation is coming in to run the infrastructures.

Aaron: So in addition to data readiness, you also wanted to draw a very important distinction between elasticity based pricing and what you called contextual AI pricing — the latter being much more advanced and actually engaging complex AI models to provide pricing recommendations. Are more retailers adopting these advanced multi-factor approaches today?

Alex: Yeah, of course we're advocating for this. I just want to say the small caveat that contextual AI is not for everyone. You need to think about your organisation readiness and your business readiness and really where the need for contextual price optimisation is. For example, if you sell luxury goods on airplanes, you don't need much of that contextual AI — or you're limited by your business, in that you have no access to transactional data by nature, so you don't really need it. Why I'm saying that: because with this growing demand, of course, a lot of people are asking and creating a lot of noise and everyone thinks this is for them, but it's not.

But in general, yes. And we'll touch on this maybe deeper in the next questions. But yeah, in general, contextual AI is finally here — but it's not for everyone. And that's why we need to look at your organisation more precisely and say, "Hey, do I believe this is applicable for my business?" We have plenty of resources that you can read on this, and of course, even better, speak with one of our experts who can give you some kind of understanding of whether this helps you or not.

Aaron: So let's just say that some executives are listening and they're wondering if this is for them — contextual AI pricing, that is. What are some immediate markers that they might be able to think through in their head: "Oh, a tool like Competera is a tool for our organisation"?

Alex: Great question, Aaron. Thank you. I feel like the first thing — if you are a C-level executive or owner of the stores — ask yourself: the implication overall is zero point five basis points for your margin, and then you need to pay for this around also 0.5 of this margin. If this really matters for you — and again, why I give you this number is so you can understand: if you're a $10 million business, probably the optimisation of 0.5% of your margin is not something big that you need to really focus on right now. But if this number sounds sizable to you, this is a good question to start looking at — because this is about tiny price optimisation points where you have your product at $3.95 and now you know with high confidence that $4.10 or $3.70 is also okay. This is where the optimisation beauty comes in — small price tuning, not dramatic changes.

And second: do you feel like your customers are really price sensitive, and you have that kind of audience with very sensitive customers and you always have this kind of problem with price image — or you feel like you are underpriced, or customers think you are expensive but you're not. If those are the questions in your head, this is exactly for you. If everything that I've told you right now did not resonate with you, then contextual AI pricing is not something that really brings to your bottom line. Let's say it that way.

Aaron: You mentioned revenue and margins there. So pricing has traditionally been optimised around one objective — usually revenue or margin — but in reality, retailers have to balance multiple competing priorities: inventory efficiency, sell-through speed, price perception, et cetera. Do you think AI can help pricing leaders manage those trade-offs in a more structured way?

Alex: Actually, this is the beauty of it. Yes. And it's also... we always tell Kim about this — AI sees things very holistically. What we've built is actually a combination of pricing campaigns using machine learning to tailor or drive one specific factor. And it's super new — again, just for the podcast — we've just deployed this at the beginning of this year. We call it model routing. It means that the computer, under the hood, orchestrates the models — we connect different AI for different purposes, as you mentioned. For example, we have the assortment that needs to be focused on sales rate adoption and there will be a model there. And by the way, there's very simple logic behind that — to see how long the lifecycle of each item is and whether we need to speed up or slow down, et cetera. And it's like this is a combination of AIs working together to achieve the whole holistic effect.

And actually machine learning by default also runs where you give it one metric to improve and another one to protect. Right now it's absolutely clear. We are coming up soon with a new concept — call it four-dimensional optimisation. It means that even without the human in the loop — because, by the way, this is what I mentioned before, the market is asking: they don't even want to go that deep and say, "Hey, this assortment, I need this strategy; this assortment, I need that strategy." This is where the computer stands — the computer can generate and help the price manager to model "what-ifs." You see the different scenarios, then you need to combine those different pricing campaigns together to get the holistic effect. But now, because of this adoption of tools and engines, the customer has become very aggressive. They say, "I need the magic button. Just give me the tool where I can say, 'Give me the optimal price today.' And that's it. I don't really want to care about anything else." And this is what is going on right now.

Aaron: So you're saying that this four-dimensional price optimisation has some type of AI pricing agent that selects which algorithms should be leveraged and what weights are being considered for each price optimisation model?

Alex: Yeah, let me explain. The four-dimensional optimisation — there are two things I said. Right now, the four-dimensional optimisation will be a new type of algorithm that actually reviews all the possible variations of the future, including balancing between those metrics like revenue, margin, sales items and profit margin — four parameters. And this is not in production yet. Honestly, what we see is it's really hard to find the optimal amongst all the optimals. That's why the two-dimensional way — but with the combination where some part of the assortment is focusing on running your top line, some are protecting your margin, some are just putting items into the market — this is what the group of pricing campaigns does. What we've done this year, we call it model routing: the computer under the hood activates the different models that are most applicable for each case. Where we're going is actually one single, more universal model that can really understand all of this without needing very tailored models — but that's still under investigation and research. But in general, it's a very good kind of north star for where all this stuff is heading.

Aaron: Thanks for that clarification, Alex. When you're thinking about all of the trade-offs in the practice of pricing, one of the biggest challenges for retailers that they frequently face is inventory management, especially for seasonal businesses. There's this idea of moving from reactive discounting to a more proactive approach that we often hear about, where pricing is used to shape demand so inventory clears at the right time.

How does that actually work now and does AI improve that approach?

Alex: I'd probably be honest — this is not much improved to do with AI. You're right. This is a big question for a lot of seasonal businesses. I'll probably not uncover everything, but what we are doing in this space is actually — and it's probably also innovative — there is a part of AI there that helps to tailor the rules, but it actually comes down to: we calculate the velocity ratio index, VRI. And technically, using stock as a signal. Each item has a specific sales lifecycle. And where AI technically plays a role is in trying to triangulate and discover those signals. The signals mean that we have historical seasonality, all these factors — we know how items should be selling during a given week, let's say week 10 of the year, amongst all the further weeks. And each time at each optimisation point — every week or biweekly or mid-week — we're optimising the price. AI says, "Hey, we're too slow on sales — speed it up," which means increase the discount, or "slow down" because we have a risk of selling too many items too early. And slowing down means: stop decreasing the price, keep the existing discount. We call this the lifecycle sell-through rate optimisation algorithm. And again, because the holistic model can route those items to this very statistical algorithm immediately. But this is not AI — this is actually statistical optimisation, linear optimisation, let's say. But the beauty is that AI detects this and delegates to the right algorithm for the right products.

Aaron: So in this case, it's AI in the loop, not human in the loop.

Alex: Yeah. Yeah. And actually, you see that the whole future would be agents — one AI delegating something to another. And even if it works better, why do we need to spend... and actually I like it because it's more transparent. And the problem with all machine learning — we discussed this in a previous podcast as well — is transparency. It's really hard to open this black box and say, "Why did AI make that decision?" In this kind of approach, right now, with this lifecycle sell-through rate optimisation, we can clearly say the price is increased to keep up with the sales. And what the algorithm does — all the time, never tired — it calculates each day how many days are left for this item, what's the high or low probability that we'll end up with a bad outcome.

And this change — of course it's nothing to do with demand optimisation that the customer feels — is optimised completely to sell out your stock, guys. And if you're buying those items or you manufacture them and your manufacturer lines are busy and they produce all this stuff, maybe one of the good signals you can have is that you're discounting too early in the middle of the year — which means your product, perhaps no one needs it. And Aaron, we spoke pre-podcast about this — this is the biggest concern I see in general in retail and high seasonal retail businesses. There were the cool-back days, and they built all these big manufacturers that produce the items for them. They're trading with the collection — some collections are really great and they repeat them year after year. For example, New Balance — the 574 collection, right? Across the years, et cetera. And for these items you can use very nice demand, classical, contextual AI algorithms, where those sneakers will be priced a bit differently in New York versus California, with all the seasonality, et cetera.

But you have to be honest — you do have some items with no real demand, and then you force AI to give you some magic, and demand optimisation just can't find the optimal price position, because in their estimation it doesn't matter what price you set — you will never sell it because no one needs this item. And that's why you need to delegate to a different algorithm that says, "Hey, I have this amount of stock — please sell out this stock by the end of this day or end of this week, otherwise a lot of money will be locked there." And this is the kind of wise beauty of AI to handle that. And of course the retail business needs to be holistically re-optimised to start asking the question: why do we produce so much of this stuff? And you probably hear how much retailers try to — it's really hard to make tough decisions and say, "Hey guys, we need to decrease our manufacturer lines, we need to manufacture less of some products that no one needs," rather than pushing this to a pricing algorithm: "Please sell it out."

But again, pricing is where I am. I'm a pricing manager. I just need to do the best price. We'll try to help them and we are helping them just by these techniques. But it has nothing to do with demand in this case — just to be honest — and the optimisation here all focuses on sell-out sales rates, no demand factors. That's because the signal of huge overstock and a lot of money locked in your warehouses is a much stronger signal than any customer behaviour or customer incentives.

Aaron: Okay, that makes sense. So Alex, looking ahead, what do you see as the next major shift in pricing over the coming years and what will separate the leaders from everyone else?

Alex: Good question. First of all, I feel like the biggest shift we are seeing right now is this ability to really speak with data — some kind of interface where you can really run multiple experiments at the same time, even using a console style, without needing to even use the UI. Say, "Hey, I have these products. Can I model this and this?" I feel like the leaders — the cutting-edge guys you mentioned — they understand and can run a lot of experiments within a day, within hours. We have one brilliant account right now where they do six price review sessions a day. Even the UI isn't quick enough for running those kinds of experiments. The shift is towards a more console-style way of working with your pricing strategies and monitoring, et cetera. That is a huge evolution of working with data. Of course it requires a specific level of skill — but if you're capable of running your things in a terminal, this can be cutting edge. That's what the leaders look like.

But still — a lot of people with an open mind understand that pricing is experimentation, and the more experiments you can run across different functions and different strategies, the better your advantage. Especially in this world where retail is absolutely not stabilised — sanctions exist and then don't exist, they keep changing. Manufacturer lines and supplier channels are shifting. The cost of delivery right now is going to increase — a huge nightmare for retailers trying to translate all those factors. And I want to make the price manager able to plug in this new reality as fast as they can and run new modelling. Say, "Okay, our delivery price will probably increase by 5–6% by the end of this month, because it's already moving — gas prices going up." As soon as gas prices go up, everything goes up, because everything is built on movement. And your challenge as a price manager is to calculate the cost of doing nothing. The cost of no action. That's why we're experimenting on getting a more robust experience where they can run queries faster, gather more data and make faster decisions — because everything can shift in a matter of weeks.

Also — sorry for the long answer — it's a huge challenge from a modelling perspective as well. We're right now using temporal neural networks — we selected them because they are faster than deep learning neural networks. The problem with any neural network is they need time and a lot of observations to make a decision. And at a certain level, the human brain can do it faster. Or even in an e-commerce environment that shifts so rapidly — catching those changes algorithmically is a big challenge even for Competera. And right now we're experimenting with new types of models — CNNs and others — coming from the finance sector. We're looking at other types of models that are used for algorithmic trading right now. This is how hard it is, guys. Maybe one quote you can pick up for this podcast: pricing in retail right now is close to betting on the public markets, like algo trading. Very close in terms of speed of decisions required, the shifts that happen consistently, the transparency of what's going on. This is where pricing is going.

Aaron: So what I'm hearing is you need a Lovable slash Polymarket approach to pricing.

Alex: Correct? Yeah. This is crazy. Uncertainty, by the way. Yeah. And only the agile win. You can — guys, again — if someone listening feels like... some retailers don't even have a pricing team right now. Like a dedicated pricing team. They still have a category manager who does 10,000 different things and needs to decide on this. Guys, right now this is a competition of algorithms — it's not a question about whether this algorithm or that one, it's a question of which algorithms are more rapid to pick up all those demand changes and make the right move for you. And by the way, it was actually just recently in the news — Walmart actually opened up their pricing algorithms, et cetera. They've done a lot of great work there.

Aaron: That's right. I actually saw in the news today that they officially trademarked their pricing technology or something like that. Yeah. Cool. Natural language interfacing for pricing experimentation and selecting the right models — is that correct? Those are your predictions. All right. We'll have to get back together in 2028 to see if it comes true.

Alex: Yeah, I like it. Let's do it bi-annually. Yeah, sounds good. Catch up.

Aaron: Oh, alright, cool. So, Alex, last question for you. What books, podcasts, or resources have you been engaging with that you would recommend to our Pricing Heroes community?

Alex: You know, the most engaging community that I listen to is the Pricing Heroes podcast. I don't really catch up with anything else. And again — maybe we'll use this opportunity to say: guys, please, let's share more. We have this great Pricing Heroes pricing community, but it's really hard to get any news here because everyone is very protective of their tactics and their way of doing things — which is a great example of why it's hard. That's why I'm more right now focused on listening — listening about agent development, et cetera. I've gotten back into coding a bit and seeing how we can make things work, and we're trying to take in all the latest data coming through, et cetera. But yeah — listen to the Pricing Heroes podcast. We try to do the best here to get great minds to sit together and share their insights.

Aaron: Well, I appreciate that Alex, and thank you for being on the show and sharing your insights with us today.

Alex: Thank you, Aaron. Yeah, thank you.