What types of artificial intelligence are applicable to pricing?

Let’s look at a simple example to illustrate both types of artificial intelligence. Suppose someone wants to convert 20 degrees Celsius to Fahrenheit.

In the first case, this person recalls the formula learned in school: multiply by 1.8 and then add 32. Applying the formula, they get that 20 °C equals 68 °F. This case represents the use of pre-existing knowledge, typical of expert systems.

Now imagine that this person does not know the formula. Instead, they have a table with historical data showing pairs of Celsius and Fahrenheit temperatures. Observing the data, they notice a linear relationship between both units. They then decide to use a statistical technique—linear regression—to predict the Fahrenheit equivalent of 20 °C.

Using a spreadsheet, they compute the best-fitting line and again obtain 68 °F. This second case illustrates the use of machine learning, where knowledge is built from data.

What type of problem can be solved with AI?
Not all real-world problems are suitable for being solved with artificial intelligence. Therefore, before building an expert system or a machine learning model, we must answer a few key questions:

Is there human knowledge to solve the problem?
If the answer is yes, an expert system can be built. A good example is robots that assemble cars by following step-by-step instructions.

Is there no human knowledge, but historical data exists?
In this case, a machine learning model can be built. But it is necessary to analyze the type of data available.

Do the historical data include the problem’s outcomes?
If they do not, unsupervised classification techniques can be used—useful, for example, to segment customers based on demographic or psychographic characteristics.

If they do include outcomes, we must identify what type they are:

Categorical outcomes: supervised classification models are used. For example, training an algorithm with medical images to identify positive lung cancer cases.

Continuous outcomes: regression algorithms are used, as in the temperature-prediction example.

Applications in pricing
In summary, there are two types of artificial intelligence applicable to pricing:

• Expert systems, which use rules defined by human experts.

• Machine learning models, which learn patterns from historical data.

Both approaches have different applications depending on the type of decision to be automated or supported.

What pricing decisions can be made with each type?
That is the question we will answer in the next article, where we will explore the applications of regression models in price-decision making.