The topic of analytics, especially in connection with big data, is increasingly the focus of public attention. We define it as the application of methods of data analysis, data mining and machine learning in order to obtain future-oriented information relevant to data-driven planning.
In a second step, the data is the basis for mathematical models in order to derive the best possible decisions. The interface between supplier and customer usually requires special attention.
Pricing and Revenue Management
Revenue Management has its origins in US passenger aviation of the 1970s and 1980s. Today, it is also used in numerous other industries ranging from car rental and hotels to the marketing of advertising time and manufacturing companies. In most cases, the aim is to maximise the turnover from the sales process of different products, taking into account stochastically fluctuating demand and scarce capacities. Our focus here is on capturing flexible production structures and taking risk aversion into account.
Consideration of complex (customer) behavior
Due to the increasing change from a seller's market to a buyer's market, companies in the service sector in particular are increasingly confronted with new challenges regarding their pricing strategies. While in the past mostly short-term thinking, so-called "myopic" customers were considered, this assumption no longer seems appropriate in many sectors today. Given temporally varying prices, "strategic" customers seek to maximize their intertemporal value by cleverly choosing the time of purchase. Similarly, customers can respond to an offer by entering into negotiations or adapt their future behavior (learn). To address this, methods from microeconometrics can be used. Discrete choice models achieve a more precise description of customer choice through a disaggregated approach based on the consideration of individual customers or decision-makers. They are therefore often superior to conventional price sales functions.
Development of solution methods for dynamic programs
In practice, decisions often have to be made in stochastic environments at different times. For example, in the management of energy storage systems, it is always necessary to decide whether to recharge them or sell electricity on the market, while prices and - for most renewable energies - production are random. Sales processes are also characterized by stochastic customer behavior and repeated price changes. The outlined correlations can be exactly mapped by stochastic dynamic optimization models, but these can only be solved exactly for small problem sizes. For practical applications, highly sophisticated approximation and decomposition techniques of approximative dynamic programming (ADP) are usually required
In order to arrive at solvable models, the later solution method must usually be taken into account when mapping the problem in an optimization model. This often leads to a trade-off between an exact mapping on the one hand and an efficiently solvable - e.g. linear - model on the other. Here, simulation-based optimization (SBO) offers an interesting alternative, since it makes practically no demands on the models to be solved. Simulation-based optimization (SBO) offers an interesting alternative, since it makes practically no demands on the models to be solved. By combining simulation and optimization techniques, many problems can already be solved "out-of-the-box" with current software. Experience has shown that efficiency can be significantly increased by clever adaptations.