How Engineering Meets Marketing: The Drivetrain Approach
The Drivetrain Approach was introduced by Jeremy Howard, Margit Zwemer and Mike Loukides in 2012, suggesting a four-step framework for building data products. This approach has drawn loads of attention in the data science community and has been widely discussed among data scientists, due to its effectiveness using data to maximise revenue in various business sectors. So, let us dive into what this approach is and how it could be used.
To who are unfamiliar with the four-step Drivetrain framework, it starts with objectives, meaning what goal is to achieve. In a business sense, the objective could be to make more sales, to grow a stronger customer base, and eventually to make more profits. The second one is levers that we can control to achieve the goal, including tools, methods or even a third-party partner. Then, it comes to data, which we can collect. Speaking of the data collection process, Howard had once made a bold recommendation to some insurance firm directors that they should randomly change the insurance price for several months and record the behaviour feedback from clients. Based on the data that they collected, those insurance companies could adjust their insurance policies accordingly to suit their customers better. And finally, the last one is models, the predictive machine learning architecture that not only makes predictions but also makes actionable recommendations. This part has a modeller (models that statistically construct causal relationships based on the data), a simulator (a model that produces the outputs by giving a wide range of inputs to test the relationships) and an optimiser (a system that recommends the optimal outcomes after the simulation process).
It could be daunting to grasp all those abstract concepts, so let us see it in an example. Assuming that we are running a burger restaurant and bar and trying to improve the customer experience through learning their feedback, the first step is to specify the objective, which is obviously to provide diners better food, drink and dining experience, so we will put pricing aside for now. Since the goal has been defined, we need to know what the levers we can control. In terms of food and drinks, they could be the ingredients, food quality, meal options, beverage selections and chef cooking skills; when it comes to the dining experience, what we can improve could be the atmosphere, room temperature, customer service, furniture and cleanliness. Then, here comes the data, which could mainly be gathered through either feedback by word of mouth or online comments. Meanwhile, we could put additional data into account as well, like current affairs or something outside the box. If there were an occurring national sports event, we should consider how the factor of having a big-screen television broadcasting the sports match live increases the number of customer coming in to dine. The final piece is the models. We would build the modellers that link the relationship between the satisfaction level of diners and all the factors that increase their satisfaction, following by the simulation that shows the outcomes when the inputs of all the factors have been adjusted respectively. Once we have found the pattern from the simulator, an optimiser would be built to make recommendations, such as promoting spicy meals during the summertime as people tend to eat peppers when it is hot.
In a nutshell, it seems like how engineers are learning marketing tricks to improve business models, or marketers using engineering knowledge to do the same thing. Certainly, I believe that using the Drivetrain Approach in the business world is just a tip of the iceberg, and there are plenties of fields and areas to explore with such an approach.
