Dynamic pricing in industries
Our solution for automatic pricing using the example of retail, industry, cargo transportation, fitness clubs
In the modern world, the market situation changes so quickly that market players do not always have time to optimally adapt to new conditions and it is necessary to leave the adjustment to algorithms
Pricing is one of the key factors influencing sales results in any business. A high price can send buyers to competitors and affect the sales of not only its own product, but also related ones. A low price may reduce margins without increasing revenue, and may not offset the cost of the product or service. The market situation can change very quickly, for example, the balance of supply and demand, prices from competitors, costs. Accordingly, it is necessary to be able to quickly calculate and set optimal prices.
Despite this, in most companies pricing is based on "manual drive". Commercial managers use Excel to analyze a small set of factors and often set a price based on their own opinion. The price may be updated once a week, month or quarter, while the market is updated every day and sometimes several times a day.
Modern methods of machine learning/artificial intelligence make it possible to analyze many factors in real time and look for optimal points. These factors and approaches are completely different in every business.
Data Studio has developed approaches to calculating dynamic prices in different industries and implements its solutions in companies of various profiles.
For example, in retail it is necessary to calculate the elasticity of demand depending on the prices of competitors and on the own price of the product, take into account the impact of sales on related products, find analogue products for exclusive products and identify products based on the prices of which customers base their purchases and even when choosing a store. All this must be done for several thousand different products about once or twice a day. In industry, the story is a little different, where you need to predict the price of a product several weeks in advance, since this is the time it takes to deliver the product to a given geography. In transport, it is necessary to very accurately calculate the cost of transportation, which often consists of empty delivery runs to the point of departure and runs from the destination point. These logistics costs are not paid by the client. Also in all stories it is necessary to analyze the client's sensitivity to price. It is impossible to calculate all this in Excel. In the modern market, prices can only be determined using algorithms and powerful computing resources. But the effects of such "smart" pricing amount to billions of rubles a year.
Case examples
1. Retail Price forecast based on segmentation of goods by price elasticity and competitors' price offer

2. Railway: Product for transportation Calculation of prices for the transportation of goods, taking into account the cost of providing services by competitors, micro-segmentation and price elasticity of customers
3. Industry: Forecasting the optimal price for products (polymers) based on micro-segmentation of customers, market parity (spot prices of competitors), price elasticity of customers
4. Fitness clubs (World-Class): Recommendation system for products and services based on customer consumption and customer profile

1. An example of pricing in retail
A feature of pricing in retail is that prices need to be determined for a large number of products (called SKU = stock keeping unit). Further, not all of these products are the same in the eyes of buyers. Some goods are key in the consumption of buyers and buyers compare prices for these goods from different sellers, preferring the seller whose price for this product is lower. These products are called Key Value Indicators. Other goods are rather related and the buyer, having purchased the main product, is ready to pay a higher price for the related product. These products are often called Longtail. Accordingly, it is necessary to be able to distinguish between KVI and Longtail products, since the pricing for them is different. In addition, the demand for a certain product, including that controlled by price, affects the influx of visitors to the store or to the seller's website, which in turn affects the purchases of other goods, and these dependencies also need to be able to evaluate. Also, sellers often have unique products that are difficult to compare directly with competitors and, as a result, set a price for them. And, finally, in real (not online) stores it is not always possible to change prices frequently, since you need to physically change the price tags. This is a time-consuming task. For this reason, optimal prices must also be predicted.

All of these calculations must be entrusted to algorithms, which at the end will calculate the optimal price for each SKU and explain why this price is currently optimal. All this information about prices and the factors on which they depend is available in the commercial manager's workplace (a special program for managing prices and assortment). There, the commercial manager, if desired, can agree or disagree with the recommended price and see the sales forecast, subject to acceptance of the automatic price or the price chosen by the manager himself.

The important thing is that the manager switches from the mass calculation mode to the mode of targeted adjustments with the ability to model: the recommended price is this - sales are this, but what if you increase/lower the price, what will happen to sales? Algorithms make it possible to calculate and see this.
2. An example of pricing in cargo transportation
Let's look at what we need to know to calculate the tariff for transporting cargo by rail from station A to station B. This example also works well for taxi services, road freight and air transport.
So, the carrier company needs to transport cargo from station A to station B on the client's order. To do this, you first need to deliver an empty (empty) car to station A and pay for it, since the client only pays for the transportation of his cargo. After the cargo has been delivered to station B, the empty car again needs to be picked up and delivered to the next customer and again paid for. Moreover, we do not know in advance where this next client is, since at the time of concluding a transaction with the first client, we do not know who will be next after him. All these costs for empty mileage are part of the cost of transportation, and it must be included in the tariff.
Further, all the time that the car is occupied for a specific client, this car is not available to other clients, which is logical. There is a loss of profit that must be included in the tariff. Why do we call this lost profit, and not just marginality, since this transportation is actually our service? The fact is that during the time when the car is unavailable for other orders, not only transportation, but also loading/unloading and, mentioned above, empty mileage are included, only not in terms of kilometers, but in terms of days. Accordingly, it is necessary to estimate the time the car is fully occupied (called "turnover") and include it as a cost in the tariff.
So far, we have only considered cost factors for the service (without fixed costs, of course). How can we determine the rate? We can use the cost-plus method, when we take the cost and add a fixed marginality. However, is it wise to do this? There are several reasons why it is not wise. Here, at least, these. The carrier company may have advantages over its competitors that it would like to reflect in the price. For example, empty runs from different stations, that is, the cost of removing a car after delivery of cargo, may differ. Since the cost price is different, then different prices can be set. Further, at the moment in a given location there may be a shortage or surplus of cars (balance of supply and demand), I would also like to be able to take this into account. Finally, different customers may have different sensitivity to transportation costs (for example, if transportation is a large or small part of the customer's value chain). If the advantage in empty mileage can still be taken into account relatively easily, then to take into account the balance of supply and demand, elasticity curves need to be calculated, and to take into account the specifics of the client, segmentation needs to be done.

It is necessary to use algorithms and machine learning methods. For each shipment, they can calculate empty mileage to the level of origin and destination station, taking into account the advantages in empty mileage over competitors. The algorithms will also calculate the turnover of the car and the optimum points on the elasticity curves to balance supply and demand. Algorithms can segment customers based on dozens of different parameters.
The results of algorithm calculations can be up to 50% more accurate than similar calculations on average. That is, modern methods of automatic ("dynamic") pricing allow you to take into account more factors, calculate and update tariffs more often, and also make it more accurate than traditional methods. This allows you to incrementally receive billions of rubles of effect.
The main task of automatic pricing is to remove from the commercial manager the task of calculating the optimal price based on many factors and give him the optimal recommended price, so that the commercial manager switches from the task of analyzing detailed data to the tasks of interaction with the client and development product category strategies. In addition, as a result of automation, the price should be calculated more accurately, take into account more factors and update many times more often than with a manual or semi-automated approach.
3. Example of price formation in industry
We have produced something and we need to determine the price for this product. Let's look at an industry whose consumers of products are legal entities, since the evaluation of products for B2C is close to the retail cases described above. So, let us produce products from which other products will then be produced. For example, polypropylene, which will later be used to make food containers, or car dashboards or bottle caps. How to determine the price of polypropylene. In fact, two factors are important here: the rating agency's quote and the client.
Machine learning methods take into account many factors when making a forecast. All this data allows you to make a more accurate forecast of quotes for weeks and a month in advance, which will allow companies to additionally earn billions of rubles.
Why is the client important? All clients are very different, since they all produce completely different products from our polypropylene, with different margins and in different volumes. Customers need to be segmented so as not to offer different prices to similar customers, and also to use the information that if some customers in the segment are willing to pay more, then it means that the rest are most likely willing to do so too. For this reason, segmentation is very important and is best done automatically by an algorithm.
In addition to segmentation, an even more important factor is rating agency citations. Agencies analyze sales markets, take into account the number of consumers and current demand, the quantity of supplied products (supply) and, based on this and their internal algorithms, set a recommended price, or quote. In principle, the quotes from rating agencies determine the price quite well. In addition, even if it were not so, everyone still focuses on it. Therefore, for pricing, we are unlikely to do better than a quote, except for segmentation.
However, we produce our polypropylene now and sell it a month later, since the product still needs to be delivered to different markets in different parts of the world. We are, of course, talking about spot sales, not regular sales. We want to send more goods to where the price will be higher. Also, if the goods have already been delivered, we want to understand whether to sell now or wait a little so that the price is higher. This requires forecasting.
Industry, how this problem is being solved now
Often using classical time series forecasting methods and human analysis. These methods are not bad, but they can be better.
We offer a solution based on machine learning methods, since the forecast takes into account many factors, including product prices in various markets, prices for primary processing products (semi-finished products obtained before polypropylene), closures/openings of factories, text information from the media. All this data allows you to make a more accurate forecast of quotes for weeks and a month in advance, which allows you to additionally earn billions of rubles.
4. Example of price formation in fitness clubs
In the fitness industry, pricing primarily influences two factors - the likelihood of contract renewal and the propensity to consume paid additional services during the contract. Moreover, the consumption of paid services also affects the probability of renewal. Accordingly, we need to choose a price not only based on the prices of competitors, the season, the price in a given location, but also the likelihood of customer churn and the estimated damage from his loss associated with the consumption of additional fitness services. The price for additional fitness services must be dynamically calculated based on the needs of a given client, identified on the basis of a recommendation service that compares this client with other clients, and also takes into account the time of additional training, the people with whom this person trains and many other factors. All this is very difficult to do manually, but is successfully done by an algorithm.

Economic effect
  • 1
    Building an automatic price calculation model greatly facilitates the work of the manager, because based on the already proposed price corridor, the manager frees up time to work with clients
  • 2
    Analytics based on real data helps to make decisions faster, as well as avoid errors in interpreting indirect data about clients and get rid of intermittent and distorted metrics obtained in surveys.
  • 3
    For large companies, the economic effect is measured in billions of rubles per year
We already have a ready-made solution
Time to adapt the solution to the business is two to four months
We integrated with your IT systems
We identify factors influencing prices
Building an automatic price calculation model
We integrate the model into the CM workplace
We have been developing cutting-edge solutions based on neural networks since 2016
The economic effect of cooperation with us is
tens of thousands of dollars a day

+7 (967) 215-75-05