General valuation framework for artificial intelligence models

Artificial Intelligence (AI) is a fast developing area that is applied to many daily problems, replacing the tried and true heuristics used by society for a long time. This paper provides a framework to estimate the value added by different steps and components used to create and apply AI models. Such estimations are useful to decide if deploying AI models makes economical sense, reported to the specific problem, data availability, existing domain expertise, and deployment costs. This framework leads to analysis of interesting subtle concepts like the difference between knowing the future and being able to predict the future, even if the prediction would be perfect. Such concepts are useful for benchmarking the efficiency of the model. As an example of applying the framework, real data from a bank was used to determine ways to decrease the operating costs by convincing new customers to opt for electronic statements. The results show that AI models are useful every time when heuristics can be successfully applied, but the improvement brought by AI may not always be spectacular, even in the case of complete information and perfect model. So estimating the real value of adding AI is important because it allows comparison between savings and deployment costs, as well as comparison between models.