In the field of marketing, the framework known as marketing four P’s mix is often used, and price is one of the critical elements of marketing. Personalization is also one of effective marketing strategies, and that in "promotion" is well executed with widespread use of smartphones and social media. However, personalization in "price" has not yet been much practiced, though dynamic pricing has been gained its attention. Therefore, four objectives are set forth in this research. The first is to demonstrate that machine learning, which has not been much utilized until now as it is regarded as "black box", can be applicable to a price research of marketing science. The second is to simulate the personalization of dynamic pricing based on a machine learning platform with scanner panel data from actual market. It is then to compare the results with that of statistical models to further understand dynamics of pricing technologies. Finally, fundamental of price personalization is developed to prove as a new pricing strategy in marketing. Utilizing scanner panel data from actual retail store data in Tokyo, the model for the optimal prices at individual level has been demonstrated as a viable platform. Though statistical modeling is often used in marketing analytics, machine learning is proved to be applicable framework for pricing simulation in marketing. Unlike outcomes from a statistical model, the optimal prices from machine learning seems more realistic and display flexible price patterns reflected to the selling environment. Simulation results indicated 5.65% revenue increase by individually setting the optimal prices suggested from the model.

In an advanced digital society, retailers have been personalizing their product proposals and advertisements in order to capture consumer needs at the right timings. However, they have not yet conducted personalization in price, which is also an important element in marketing. The purpose of the research is to design a personalization framework of the optimum pricing. We developed a joint model of store visit and product purchase for all customers in a nested logit model with hierarchical Bayesian framework, then the individual optimum prices are dynamically back calculated with results obtained from the nested logit model. With scanner panel data of retail stores during 2019/1/2-12/31 in Tokyo, the optimum price for a ketchup product was estimated for 569 customers simultaneously using the Markov Chain Monte Carlo (MCMC) Metropolis-Hastings algorithm. We also demonstrated a general dynamic pricing model and price density on a daily basis as a bottom-up approach of combining the optimum prices from all customers. Personal pricing simulation results indicated to increase approximately 8.6% revenue than that from the actual pricing. The personal optimum price prediction proposed in this research is an example of resource-based process and introduced as one of most efficient utilization of Big Data in marketing.