Which Modeling for Which Circumstance?
In this blog I discuss further some of the differences between a statistical marketing mix model versus that of agent-based modeling. For major brands the stakes couldn’t be higher. Every share point is contested between rivals and winning can be achieved using better analytic techniques. Here are three additional, critical differences between statistical marketing mix modeling and the new agent-based modeling using ProRelevant’s MarketSim.
Single Brand Model vs. Category Model
Marketing Mix Models are typically built at the brand level. Unit sales at the brand level are used as the dependent variable, whereas price, media and other marketing activities are defined as the independent variables. The individual sales of the SKUs and competitive sales are ignored. MarketSim on the other hand models all SKUs in the category, including currently active, discontinued, newly launched, seasonal and promotional SKUs. Because it models the entire category, this allows a very robust ability to simulate future sales given specific actions of the brand or its competitors. It also easily allows the simulation and optimization of a new product launch.
ROI Measurement vs. Future Success Optimizer/Simulator
Whereas a statistical marketing mix model is typically built to simulate the past and project a short term future, an agent-based model is a full blown simulation, simulation all actions over the modeling time period. Because it is a simulation, it is a simple exercise to run the simulation into the future. In this way, the ROI of past marketing activities can be determined, marketing elasticities can be calculated across all 4Ps and then brand managers can test various hypotheses about the impact of upcoming marketing activities and potential variants. In this way brand managers can determine the risk of their programs and make trade-offs between risk, investment and even timing.
Data Periodicity Limitations
One of the major challenges a statistical Marketing Mix Model has is that marketing data doesn’t come in the same periodicity. It comes in varying periodicities. Digital data can be daily, media data can be weekly or monthly, sales data can be weekly or monthly and brand data can be quarterly or annually. Sponsorships and Events are big one time events and all of these varying periods wreak havoc on a statistical model. Time conversions need to be undertaken and they can lead to results that are often non-sensical or unable to compute.
MarketSim Agent-Based Modeling uses all media inputs. Insights from other brands and the category in general can be applied across all brands, such that the insights from the model can be strong across all types of inputs. In this way, one-time events as well as any mix of data granularity can be incorporated into the model.
Stay tuned. Our final blog post on this topic will be on its way in a few days. Please join us for our webinar on this topic on July 28th.