Sanguk Han: We can absolutely get B2B price elasticity with win-only data with the caveat that for any given product or even product category, you will start to see increased or decreased consumer deals, maybe not volumes, but maybe total number of deals that you were able to accomplish at certain price points. And if you're accomplishing less deals in a product category relative to your other product categories.
When you increase the price, that's price elasticity.
Barrett Thompson: Hello everyone. My name is Barrett Thompson. I'm the general manager of commercial excellence at Zilliant. And I'll be your host for our podcast. I'm joined today by Sanguk Han, a senior data scientist out of our Zilliant UK office. Sanguk, welcome to B2B Reimagined.
Sanguk Han: Thanks, Barrett. Real glad to be here.
Barrett Thompson: Sanguk, before we get into our topic for the day, would you mind telling our listeners something interesting or surprising about you that they won't see on your LinkedIn profile?
Sanguk Han: Absolutely. I was born to a family of diplomats and it meant that growing up, I ended up growing up in [00:02:00] nine different countries before eventually settling down in London.
Barrett Thompson: That sounds great. What are some of the countries in which you’ve lived?
Sanguk Han: I'm Korean by ethnicity, but I was born in Denmark before we traveled to Libya. This was under Gaddafi's regime, actually, and my younger sister was born in Libya and then we went around the world to the United States where I picked up the accent that you're hearing today. And then to Korea, Singapore, El Salvador, and the list goes on.
Barrett Thompson: Yeah, that's truly an international tour will thank you for joining us.
Today, we're going to talk about an area in which you are a specialist, and that is price elasticity and Sanguk thank you for being on the leading edge here at Zilliant helping us refine our B2B price elasticity capabilities.
Sanguk Han: My pleasure.
Barrett Thompson: It's still curious to me and I'd love to get your perspective.
I continue to run into folks in the B2B pricing space. These are practitioners. [00:03:00] Who say that price elasticity is a, B2C only phenomena. That's their honest perspective. And, I would love to get your view on elasticity and B2B. Is it real? And how do we know it's real? What does it look like?
Sanguk Han: I think that's a great question. And I empathize a lot with the B2B practitioners that believe that B2C pricing is the only real price elasticity phenomena. Just because it's more observable in B2C, that's a certainty. So when we talk about price elasticity, just to help define it, let's start off by defining price elasticity.
We're talking about the demand response that consumers have when the price of a product is changed. And generally in B2C environments, this response is either immediate or very rapid. So we can easily identify that volume response by analyzing our own historic sales data. In B2B pricing, there's usually large lags because a lot of [00:04:00] supply chain logistics are involved when purchasing deals in the B2B sector.
You might sell your product to your customer through an annual contract and mid contract increase the price of some of the products that you're selling. That customer mid contract will have a supply chain that's centered around procuring from yourself, inputting that into a manufacturing process. They might find it very difficult to move at that moment in time. That does not mean that you haven't increased the probability that they will switch contracts once it ends next year. And so at least it's my personal opinion, Barrett, that B2B price elasticity is a definite thing. It's definitely harder to measure than B2C as well, but that does not mean it doesn't exist.
Barrett Thompson: And that makes total sense to me. You know, the switching costs that you described. Many of our customers are asking us to help them pass on cost changes, mid contract, where, and as they can, according to [00:05:00] the terms of that contract, but you're right. Eventually, right. The consequences of all price changes ultimately play out in the market. Don't they? And customers decide.
Sanguk Han: I have a case study on this from the UK in an industry I used to work in. So I used to work in the shipping container industry for logistics and in the UK, the business model is generally shipping containers come into the big ports and the last mile delivery is done through HGV drivers.
So these large lorries come up to the ports, pick up the shipping container and then drive them to their last delivery. Over a number of years in the late, let's say 2015 to 2017 HGV driver costs started increasing over time. Now these profits were first materialized in the HGV Holyor industry. But what ended up happening is the shipping players started getting smart and they started thinking, hang on a second. If the last leg delivery using drivers is getting expensive, can I start using smaller ports? [00:06:00] Actually, this is changing my economic equation of logistics. It's now getting cheaper for me to charter a smaller vessel. To move my containers to a port that's closer to my end destination so I can pay lower driver costs. And this caught a lot of logistics players by surprise, who thought they could just increase prices and you know what? They were looking around them. They thought their competition was other logistic HGV logistics suppliers. So when they thought, oh, they're increasing their prices too, we’re absolutely fine.
They were completely blindsided by the fact that the product that was replacing you, isn't just an HGV truck. It's the ability to do last leg delivery through any mode of transport.
Barrett Thompson: That's very interesting in describing B2B price elasticity. Are you saying that we should expect an individual customer's quantity on tomorrow's order might change if I change my price?
Sanguk Han: There's a big misconception on B2B [00:07:00] pricing. Is that. Increasing prices can result in decreasing volumes purchased by a single customer, that a single customer is choosing the quantity to purchase based on the price of the product. For the most part, this isn't the case. It's actually - based on the price that's offered to the market, your real change in volume is coming from how many customers decide to take you up on your offer at that price? Once the customer is taking you up on the offer, the actual quantity they buy is actually independent of the price that you offer, because that's based on their actual business needs. You know, if they're a restaurant and they're buying expensive wines from you, it's based on the size of the restaurant. They won't suddenly decide to buy half their wine supply from you and half their wine supply from another supplier. The actual volume response is the number of restaurants that are deciding to procure from you at that price point.
Barrett Thompson: That makes total sense. And I really appreciate you clarifying that for me. So we've established that [00:08:00] well, yes, it is in fact possible to calculate B2B price elasticity and beneficial too. Are there some challenges in doing that compared to say retail, something that makes it more difficult or something that B2B companies need to look for that might be different than experiences or econ 101 training that they had about the retail elasticity?
Sanguk Han: Absolutely. The two biggest complexities we find in measuring B2B price elasticity is, one, in the time lag. So this is the timing difference between when the pricing decision is made and when the customers actually respond. And secondly, it's in data sparsity, which is, there are significantly fewer transactions going on for B2B products than there are B2C products.
And we here at Zilliant have a number of methods that we use to overcome these challenges. So if I actually talk about the data sparsity one first, in B2C, it's quite [00:09:00] simple. If you're a retailer selling a bar of Kit-Kat and you're change that price on the bar of Kit-Kat, there are probably tens of thousands, if not a hundred thousand customers walking through that counter regularly, making their one-zero purchase decision on whether or not to buy a Kit-Kat and you end up with a demand curve that looks quite smooth because while each individual decision is one-zero, when you aggregate it up, you can start to see, okay, 10% fewer customers purchased it a nd this demand is, is immediate, but let's say you're doing B2B sales and you're selling a highly configured product. Let's say that you're selling construction materials and you're selling insulation, 24 millimeter, 35 millimeter for specific use cases. You might have only sold that product to three or four different customers in the past six to 12 months.
So how do you measure price elasticity when there's only been three or four customers that have actually even looked at the product? Well, here at Zilliant, what we do is we use a process of inference and something that we in data science called hierarchical [00:10:00] learning where we say, all right, maybe for that product, you've only seen three or four customer purchases, but across all installation products can we find something where there's actually this other more general insulation product that has at least 20 to 30 customers. And can we learn the price volume relationship there and extrapolate it to this niche product? Can we look at the observations of a niche product and determine statistically, whether it comes from a population distribution, that's similar to a broader product that you're selling.
And this level of statistical inference really helps to get rid of this data sparsity problem that you find in B2B transactions.
Barrett Thompson: So it sounds like we're triangulating from that, which is well known, in order to make our best guests or a reasonable data driven inference on that thing, which hasn't been observed very often in our market.
Sanguk Han: Yeah, you're absolutely getting it right. And the retail equivalent of this would be [00:11:00] what if you've been selling Kit-Kat for two years, but you've never sold Hershey's. How would you then infer the price elasticity on Hershey's? Well, if you truly believed that the end consumer that buys Hershey's is a similar one to the one that buys Kit-Kat you could take the Kit-Kat price elasticity. And use that as your best estimator for Hershey's price sensitivity until you built up enough Hershey's pricing data to then allocate its own elasticity model to measure it. That's exactly what we do in B2B, except instead of Kit-Kat and Hershey's it's 24 millimeter insulation and 35 millimeter insulation.
Barrett Thompson: Yeah, that's a very clever and really practical approach to things that are realities of B2B. Not all products have the same velocity, not all products are sold into all of your segments. And so you've got to have a way to deal with sparsity.
Sanguk Han: Absolutely. And here we go to velocity as well. So I mentioned that the two things that make B2B elasticity measurement a challenge is the time [00:12:00] lag in price volume response and the data sparsity. So let's say you can take care of data sparsity by doing hierarchical modeling. Well, it turns out when you do hierarchical modeling, you can also share information on time lags as well. You can also see for products that you've sold for considerably longer periods of time.
Let's say two to three years. What has been the price fall in response there? Because when you start measuring it over longer time periods, and you have to be careful to normalize the data for other economic events, but once you do that, you get a generalized price volume relationship that you can start to estimate shorter term product elasticities with, and you keep measuring to check whether this product that you've just released in the market, let's say three months ago is still following the general trend of the product that you've extrapolated from. But you can actually start sharing this information and tracking it and therefore enrich your data so that you can use longer-term historical [00:13:00] signals, as well as cross-sectional data from other product classes.
Barrett Thompson: This is a very interesting and intriguing technique, if you will, you know, combining the near-term and the long-term signals. I'm curious to know, are you giving equal weight or equal signal intelligence to those, or how do you combine those together? What's your thought on that?
Sanguk Han: This is very commercially driven and it's an exciting discussion point when we actually configure our pricing software together with our clients.
Because in some industries it's so fast moving the actual underlying economic regimes are switching every three to six months that you don't even want to be using data that's a year old or right. Or if you are using data, that's a year old, you want to overweight your most recent transactions because you're certain that that's more important in that.
We actually have a method of overweighting that relationship in the most recent quarter, so that yes, you still use the year, but you make sure that it's weighted to more recent [00:14:00] events. Now in other industries that are slower moving where things might not have changed all that much. You can actually use equal weighting because you're just as interested in what happened at the market last year, as it did happen this year.
But this is something that a statistician cannot tell you. In my opinion, this is something that has to be business led and the model has to be commercially configured to be able to take this into consideration.
Barrett Thompson: I'm excited by this because when I think about the range of customers and range of industries that we serve, it might be all the way from stock HVHC products, which are sort of run rate year in and year out and contrast that with custom blended chemicals where the underlying commodities are being recosted, or the supplier costs are changing every day or every week.
And so I've had customers say to me in a sense they've they of set the stake in the ground that says, you know, no price point from a year ago has any information value on what I do today. And [00:15:00] sometimes I've thought, well, maybe it does, but we might have to be careful about it. Maybe the net price from a year ago means nothing, but maybe the markup over cost from a year ago, teaches us something about where an industry values your product above another.
Or maybe the quantity from last year compared to this year is not directly comparable, but perhaps I'm looking for seasonality signals across multiple years. So when I look in a relative way, I am extracting some useful signal. But not getting tricked or tripped up by the idea that the net price, the price per unit that I collected at some way distant time period, should somehow apply in today's market, especially when I have a very dynamic market.
Sanguk Han: I think the key here is when you have a dynamic market, it's much better to identify: “Is there a variable that I can actually normalize my time series data on so I can use most of it rather than to just cut out the data [00:16:00] completely.”. So for example, in certain European countries, we've seen a trend towards shopping at bigger and bigger retail or wholesale stores as a direct result of COVID. Convenience retail has gotten smaller, whereas large-scale wholesale grocery has gotten larger with fewer shops. Now, when you do price volume elasticy analysis and you just look at big shops and you do year over year analysis, it'll seem like, wait, hang on a second, no matter what, volume's going up, but hang on a second.
What you could do is, but for each product, what if we normalized the volume sold of that product by the footfall of consumers that entered the store. You would suddenly normalize out for the overall market shift into the bigger stores, because now you're comparing for every hundred people that entered the store, what percent actually bought the product at that price.
And suddenly you can use multiple years of data again, and this is so much more powerful than just stripping it [00:17:00] away and saying actually, you know what? We can only look at the last four to five months of data.
Barrett Thompson: I love that. That makes such sense. Let me ask also I will have customers from time to time say, my market is growing. I'm growing organically, maybe I'm growing inorganically. And so, I've got this year on year volume increase that really has nothing to do with price. So it's like a trend in the data or back to seasonality in the building supplies industry for example, there's often seasonality. In the food distribution industry, seasonality by product line, you know, fresh fruits and veggies at certain times of the year. Other things other times. Do we have the ability in our B2B calculation methodologies to account for trends, seasonality, or other factors that we might need to take out in order to get down to that essential price volume relationship?
Sanguk Han: Absolutely. And the way we do it is different in every industry, but just to take an example, a [00:18:00] recent example.
So we were just talking about footfall at grocery stores, but actually it sounds like a good idea to just divide all of the sales by footfall, and then you have a normalized variable, but actually when you dig into it, very few stores know their actual footfall because nobody really measures how many people entered without having purchased anything.
So we don't have footfall, but we need some proxy of overall market demand, which can also proxy for seasonality because when people are engaging with you in higher numbers, you would expect underlying products to also sell more. Whereas when overall engagement goes down, you would expect each individual product station to go down as well.
And here's where at least in retail, you look at tobacco sales. Why? Because tobacco sales are some of the most consistent measures of footfall you can imagine. It's very insensitive to changes in seasonality. It's very insensitive to what the weather is like, and it will help you normalize out things like industry growth [00:19:00] levels.
So in this way, we usually have a process of working together with the client to understand. Well, what is the actual seasonality or growth trend that you want to normalize out for? Sometimes it's very simple; for construction for example, if you're worried about needing to normalize out for general construction activity, when you're selling, let's say plywood.
What if we just took a look at the total number of consumer transactions in a given month. And we normalize based on that, because even if consumers aren't buying specifically plywood, the total number of consumers transacting with you, or the total number of unique consumers transacting with you should be a good proxy for whether there's construction activity in the market.
So sometimes it's easy like that. You can use direct customer sales. Sometimes you've got to get really clever. It's always interesting. And it's a fun case by case basis that we apply.
Barrett Thompson: The construction example reminds me that one customer I've worked with, a manufacturer in the construction industry.
They have a department [00:20:00] that does their own sales, forecasting, economic forecasting looks at all sorts of statistics on what's happening in their industry. They'll look at the number of permits that are being pulled for new commercial buildings in a zip code. And so they may have unbeknownst to the folks over in the pricing team.
For example, they may have these, these signals, these normalizing signals at their fingertips, right? And the power of combining that over with the transaction data would be really huge. Another question that I often am asked is, do you need both wins and losses or don't I need win-loss data, both parts, in order to get a reliable elasticity signal.
What are your thoughts on that?
Sanguk Han: So we can absolutely get B2B price elasticity with win-only data with the caveat that for any given product or even product category, you will start to see increased or decreased consumer [00:21:00] deals, maybe not volumes, but maybe total number of deals that you were able to accomplish at certain price points.
And if you're accomplishing less deals in a product category relative to your other product categories when you increase the price, that's price elasticity. So you can still infer it because loss is the equivalent of just saying less wins, fewer wins. So just by knowing your win data and having some level of benchmark underlying volume and this could be your other portfolio, this could be, Hey, we price insulation and plasterboard separately. And do you know what, insulation is underperforming to plasterboard when we start increasing the price? Well, we know construction activity is still going on because there's other parts of the business where we're selling.
This one isn't doing so well. Well, that's the price volume response so we can absolutely calculate it without loss data.
Barrett Thompson: If I have the loss data, can you take advantage of it?
Sanguk Han: Even better. If we have the loss data, we don't even have to proxy [00:22:00] because statistics wouldn't be necessary if we had all the data. Now we're just into big data.
I don't even know if I would have a job because you wouldn't need someone to give you the price elasticity. You would just know the price elasticity. So when you have the win-loss data, you can get a very accurate measurement of, well, how is my win-loss percentage affected by my pricing.
That's the easiest way to do it in absence of that? Well, how's my win rate affected by my pricing. If that's too unstable, then how's my win rate in one department relative to another unrelated department influenced by my pricing strategy? So we can always start to create these proxies and at Zilliant we have the methodologies to do it, but of course, if you have the data, let's always use the data.
Barrett Thompson: And I've seen some really varied practices with respect to sort of traditional human led sales, where they might say, no, I don't know anything about my losses because my sales [00:23:00] reps are having phone conversations with customers. If the customer says, yes, they want to buy, then they log onto the system and begin a quote.
So literally there's no electronic trail, no bread crumbs because no one bothers to enter the quote first and then see if the customer will say yes, they asked the customer, will you buy this at that price? And if they say yes, then they go create the digital signature in the form of an order. I also have worked with customers who say I'm a manufacturer or a distributor, either one, I'm selling into a vertical, thinking of building construction here, commercial construction.
They'll say there's a new hospital being constructed in my town. And there are several general contractors that are trying to win the project. And so rolling up under them are all sorts of say electrical subcontractors, maybe as many as a dozen, and each of them is putting in a bill of materials that they want quoted, but there's only one building being [00:24:00] constructed in my town.
But to me, to my price team, all these trickle up to me, all the subcontractors are saying, what would you charge me for this product? It looks like 12 quotes, right? So let's say I price all 12 of them. And in fact, One of those subcontractors is going to win and they choose to use my product. I've captured 100% of the buildings being constructed in that town, but it looks like I've only captured 1/12th of the quotes.
So the losses are overstated because they really represent the same opportunity in the marketplace. Yes, this customer had loss data, but they told us those losses they're way over inflated. And so you have to be really careful if you were to get your arms around that. So I've seen some real variability there.
Something I find exciting is the emergence, in fact, explosion of the e-commerce and online channels and digital commerce behavior. So there I'm hearing questions like “I have very solid cart abandonment [00:25:00] statistics. I see when someone creates a cart, I see when they push submit and I get paid and I see when they don't. Hey Zilliant, hey data science team, can you somehow put that to work in my elasticity model?
Sanguk Han: Yes. So e-commerce is incredibly rich with data. When you start to analyze these sales funnels to understand, okay, where did the traffic even originate? Was it from a pricing banner? Was it from a discount? Which ones are the most attractive?
And then after they land on your page, you check what was added to the cart, what was added together, what products are being purchased together. And then finally did the customer end up eventually purchasing the product? Instead of just coming up with a one zero price elasticity, you start doing market mix modeling around each of the different signals.
So when we make a different decision around where we promote does our inbound traffic increase, because that is a sensitivity. [00:26:00] Once the inbound traffic is on the page, the size of the basket. Cause you know, this based on the digital data. Is that influenceable by how we laid certain products out by moving certain products that are purchased together to always show up together?
And then finally, the final click through rate. Was there anything that we did, some of these are not price related. Actually the biggest one is the customer had to click too many times to get to the checkout page. This is massive. So, if you can remove one or two levels of that, what's the final click-through rate?
Each of these is important information to decompose so that you can start optimizing your pricing and marketing levers at each individual step, rather than just saying, oh yeah, we did something. We did the aggregate of all of these banner advertisements, these page layouts, these click-throughs, these prices. And the aggregate effect of that was a purchase or not a purchase. That's usually less powerful than analyzing each individual step of the transaction.
Barrett Thompson: I've often felt that some B2B businesses sort of [00:27:00] use price as a blunt instrument. You know, they use it to try to be the solution to problems that actually should be looked at a different way.
So if I have too many clicks, as you gave an example, I need to fix that problem directly and not hope that I can make it up by dropping price 2% and asking people to suffer through the click stream. Price is a powerful lever, but you need to think about that whole user experience.
Sanguk Han: Yeah. And this is also true in the non-digital space as well. So I did some work with a wine merchant a couple of years ago. So this is a B2B company that does wholesale distribution of wines. A lot of them high-end. And a couple of years ago, in order to try to save costs, they had moved one of their call centers from London to a more regional hub.
And they hired people who weren't familiar with wines into these call centers. [00:28:00] And what started to happen is these orders would come through their inbound orders were exactly the same. But do you know how complex high-end wines are? People will ask you, give me the Donia Tinto grand reserva 2005. And if you, if you miss that, or if you say, please come again, it will take forever to get these orders through.
And so their actual rate of losing the order after the inbound call increased. This had nothing to do with price. The person saw the price and was willing to purchase, but the salesperson that they were dealing with couldn't execute the final step. Even if they decrease the price, all that would have done is increase the funnel of inbound calls.
It wouldn't actually have yielded in the final sale. And so I could not agree more with you. We cannot lie to ourselves, that pricing is the only mechanism by which we communicate to our customers. There are so many ways we do that and pricing has to be done in conjunction with all the other levers.
Barrett Thompson: That's so true.
We need that systems view. Don't [00:29:00] we, right? Well, let me ask too, as a concluding topic, what can you tell us about how Zilliant’s approach for B2B elasticity is differentiated from others who provide price solutions in the B2B space.
Sanguk Han: Right? I think with Zilliant’s approach, we are very grounded in the commercial realities that our clients operate in. And we don't just trust the data. So as a data scientist, at Zilliant, when I'm working on, let's say, quote, unquote, improving our algorithms. Most of the time, I'm not working on making them more sophisticated or more complex or more accurate. I'm actually thinking about how to make it more robust. Does it reflect reality correctly?
How do we work together with our customers to make sure that these concerns that they have around, Hey, I'm not sure if you did price last, just over the last two years, you'd see anything because we were in a growing market everything's growing. So if you, if you normalize incorrectly, [00:30:00] it'll look like we increase the price and volume goes up.
And so much of our process is. Having a discussion, a commercial discussion to understand, are we correctly capturing market dynamics in our model? And I'd say that's the science. And in that sense, we're not data led at Zilliant we're science led and being science led means coming up with hypotheses, peer reviewing your hypotheses with different members of the team, and then analyzing the data to either confirm or reject a hypothesis. If you don't do that hypothesis step, you don't do the peer review step. And instead you go, we analyze the data. You know what price goes up, volume goes up. We're good.
You reach completely incorrect conclusions.
Barrett Thompson: I love the thoughtfulness and the transparency of that process that you described. And I've seen firsthand how it leads to outputs that the business can in fact understand embrace, and then capitalize on as they, as they take it forward to the market. I want to [00:31:00] thank you again for taking the time to chat with me today.
It's been a real pleasure and very informative conversation. Thanks for sharing your perspective with us.
Sanguk Han: Well, thanks Barrett. It was an absolute pleasure to be here.
