Sterling Simulation consulting company was chosen to provide an agent-based marketing model for a pharmaceutical firm. The company owned two competing non-generic drugs on the same market. One drug was well established and tended to be the industry leader, and the other one was recently introduced.
There were several concerns about how to obtain a useful market share for the newer drug, while maintaining or increasing the market share for the company’s drugs as a whole. The company used different forms of promotion, including direct-to-consumer (DTC) marketing strategy, which usually concludes in advertising through TV, pint, and other mass and social media. The model was supposed to answer the question, “When should the company stop DTC marketing for the newer drug to maximize total prescription sales?” on DTC marketing investment. The provided answer, if followed, could save the pharmaceutical company tens of millions of dollars.
Traditionally, marketing analytics have decided on different spend scenarios by using a marketing-mix model to determine the impact of marketing spends. However, this approach does not provide articulated understanding of why the spend changes impact the results. Marketing-mix models can give you weights, but they cannot tell you why these weights exist. The company wanted a model that would be at least as good as a marketing-mix model, which meant the contractors had to find the percentage point of the market shares over time.
To obtain a better understanding of the mechanics behind marketing-mix models (for example, why does DTC marketing give diminishing returns v. sales rep visits?), one alternative is to use agent-based modeling (ABM). It provides the secondary benefit of removing assumptions from the model, which allows for a more complete understanding.
One thing to note about ABM approaches in general, and this model in particular, is that the data requirements are different from those in marketing-mix models and, in general, higher. In this case, the pharmaceutical company had the data and was able to take full advantage of the ABM approach. Agent-based models tend to be more open for assumptions and provide the insights, which are more profitable than plain answers.
Once the approach of ABM was selected, AnyLogic was chosen as the software platform for model building. These were a few of the reasons:
- The company was familiar with the software and its capabilities.
- AnyLogic allows for the greatest flexibility in modeling frameworks.
- AnyLogic has the best visualization possibilities for modeling.
The model framework differed substantially from traditional marketing-mix models. Specifically, the model considered the entire patient/doctor interaction in order to determine the impact of market spends. Additionally, the impacts of the new drug’s introduction to the market was integrated to obtain correct market share information.
The model consists of the following elements:
- sales representatives;
The latter two elements were not described in the case because they had little behavior in the model.
The patients in the model were all diagnosed with the specific illness that the drug market handles. The disease involved was not life threatening, so this drug category is elective. The behavior of the patient included:
- Meeting with their physician every three months.
- Determining whether they desired a specific drug (the result showed primary DTC market impact).
- Awareness of different drugs based on advertising and the ability to request it from doctors.
- Whether they filled their prescription (depended primarily on drug’s price).
- Whether they stayed on a drug (on average, the first month and second/third month loss was calculated to be around 40% and 20% or less, respectively).
The doctors in the model had different specializations related to the illness, and had differing numbers of patients, depending on specialization. The behavior of the doctor included:
- Handling patient appointments.
- Determining what drug to prescribe to a patient (theoretical preference was based on clinical drug performance, practical preference was based on patient reaction on drugs).
- Whether to provide samples or a script to new patients.
- Interacting with sales reps.
You will find the preassigned model of a patient’s behavior during and after an appointment, as well as the description of a patient’s lifecycle below.
Sales representatives were assigned to a pool of doctors. They visited their doctors at a different rate, based on the patient pool of each doctor and historical information. During the visit, the representative tried to change the doctor’s attitude towards a certain drug by adding samples to the doctor’s supply.
The drugs in the model were the two company drugs, another specific non-generic drug, and generics as a group.
The model did take a while to calibrate properly. This was because the data was more than a little sparse in areas of the desired information. The model was calibrated primarily to each drug’s (or drug family’s, in the case of the generics) market share in terms of both patients and prescriptions per month. Once calibrated, the model showed that the ideal time to stop DTC marketing for the new drug would have been six months before the current date. This was established by noticing that there was no difference in the calibrated metrics over the same period. Since this was unfeasible, the given answer was to stop DTC marketing in the near future.
Another interesting insight was connected with sales rep marketing. It was clarified, that over time, the doctors’ preferences overwhelmed patients’ preferences to the drugs. That is why the money invested in sales rep visits, in contrast to DTC marketing, did not show any diminishing returns and always influenced market share. This was expected, given that the availability of samples was directly tied to the visits, and had a broad impact on the willingness of patients to try the drug.
Concerning the budget issue, it was safe to say, that should the pharmaceutical company follow the results of the model and eliminate DTC marketing, they would save at least $10M a year.