In an economy where every sale counts, North-West University (NWU) researchers have been looking at ways to help retailers improve the decision-making process in their marketing campaigns.
Chanel Bisset and her supervisor Prof Fanie Terblanche from the Faculty of Engineering in the research field of engineering analytics, recently conducted a study on a stochastic programming approach for marketing campaign optimisation.
According to Chanel, the main focus of this study was linear programming under uncertainty, also referred to as stochastic programming.
“Stochastic programming originated during the 1950s and was later extended into prominent application areas, including energy, production planning and finance,” she explains.
“One research opportunity identified is that retailers require predictive models that can accurately describe customer behaviour and predict future sales to maximise profitability.”
For the study, the two researchers formulated a two-stage model by combining two base models identified from literature.
“A two-stage stochastic programming model consists of two decisions made at two different periods,” Chanel says. “The first decision is made without any knowledge of the future outcome and is mainly based on judgment and experience. The second decision is influenced by the random outcome's realisation which is based on a probability distribution.”
She adds that retailers need to make these decisions concurrently and need predictive models that can assist them with this decision-making process.
The model's first-stage decision includes whether a customer should be targeted for a promotion campaign in a specific period given the marketing cost. The second-stage decision is based on whether the retailer should promote the product in-store.
“If the customer is targeted, there is still uncertainty if it will positively influence the probability that a customer will react to the promotion campaign. Retailers are constrained to a budget and can only target and promote specific customers and products during the promotion campaign,” says Chanel.
“This model will assist retailers by only targeting the most valuable customers, and also clearly predicts if a product should be promoted in-store as part of the second-stage decision.
“A deterministic version of this model is also formulated where all the probabilities and additional data are known. The results of the two models are compared by using simulated data,” concludes Chanel.
Chanel Bisset.
