Over the last few years, our consultants have built several models and applications that facilitate predicting, simulating, analyzing and maintaining prices for various businesses.
It can be difficult to estimate the right market price, especially when selling configurable or complex products. This is because the right market price is not the same as a price list built into the ERP or CRM system; it is the highest price each customer is willing to pay for the offered product.
An intelligent app can help the sales person’s work by giving answers to these questions, for example:
- At what price have similar products or assemblies previously been sold?
- What is the right price level, considering all the features of the product?
- What is the probability that the product will be sold at a certain price?
- How fast will the product be sold at a particular price?
Identifying comparable configurations
In one of the solutions we built, a car seller enters the information of their car into a system which searches data from similar, previously sold cars based on the degree of equivalence of car properties. The app helps to evaluate a suitable selling price for a new car which has come up for sale. The seller can quickly see if the planned price is too low or too high. In addition, the selected variables can be changed from the app’s sliders to simulate combinations of different properties. This allows the users to investigate the products’ historical pricing with different configurations and equivalence degrees.
Features that should be considered include not only the various technical characteristics of the car, but also the usage history, the age of the car, and the expected car selling time. When searching for the right price level and optimizing revenue, it is also good to know what the car price would be in a couple of months if it isn’t sold quickly.
This kind of use case suits well to support the pricing of various configurable products when a product is created on the fly during the sales situation. Whether the product is combined from different chemicals, metals, fiber materials, veneer layers, machine parts or combinations thereof, or even a package of services, there are many similar applications. Let’s think about, for example, the pricing of various steel profiles or custom plywood boards, or a service concept created from different real estate services. By selecting the properties of a product or service from the app’s menu, it is possible to quickly check if a completely similar entity has been sold to someone before and at what price.
Predicting price based on product features
When the offer history is added into the model, it is possible to examine the realized sales and also lost offers, and how these lost offers have been priced compared to the actual sales. By adding customer data to the model, factors related to the current customer segment or market situation can be taken into consideration.
Based on this kind of data, the system learns to find the right market price or price range for pricing new products or configurations, as long as the new products are at least somewhat similar to ones sold previously. Based on the historical data, the model learns how each feature of a product, or combination of features, affects the price. In car sales, it depends on the brand how the kilometers affect the price as the car age increases. In elevator sales, the type of building and the market area have an impact on how some additional features affect the price. Artificial Intelligence learns all these principles by itself, based on the data.
Optimize sales margin, sales time or demand of substitute products
Based on only sales data, it is possible to support pricing by finding a suitable price level to maximize the likelihood of realizing a deal. By adding also some cost and profitability data, we can get to an actual profitability improvement, where it’s also possible to predict an optimal price level to optimize the total margin. With the same kind of model, it is possible, for example, in the wholesale trade to quickly apply for substitutes of the product and estimate whether we could get a better sales margin by offering another product that meets the customer’s needs.
Data and experience
At least some historical sales data is needed to build the model. At a minimum, the required amount of data is some thousands of sales transactions. The more data is used, the better the accuracy of the model will be. As a rule of thumb, the required amount of transactions is a square of the number of dimensions in the model. If we want to add profitability optimizing into the model, the cost data must be available with a sufficient accuracy.
Industry-specific differences and the data quality are immediately considered in the initial design phase. Identifying data gaps is as important as utilizing the data itself. If the required data, for example offers, are not collected in an analytical form, it is important to start collecting the data as soon as possible – almost without exception, some competitors have a head start in this already.
If you would like to hear more about the above-mentioned frameworks or have other questions about data collection, do not hesitate to contact us!
Bilot Finland: Ashwin Kumar, firstname.lastname@example.org, +358 50 564 5501
Bilot Sweden: Mathias Hjelt, email@example.com, +46 70 625 3461
Bilot Poland: Mariusz Papiernik, firstname.lastname@example.org, +48 690 540 522