DS_market__retail_timeSeries

Requests from the project

• Hi Nicole, I hope you are settling in well in your new role. Martin has mentioned that you have already been able to access our databases. I’m happy to hear that! Since I joined DSMarket, I’ve been wanting to analyse in depth the current picture of the company. So far I’ve been looking at global sales trends, but I really would like to evaluate every angle of our activity. I’d like you to help me with that. It would really appreciate if you could start looking at the data from NY, Boston and Philly. My intuition says that we probably have some products that are not so popular anymore, and it’s likely that most popular products vary across cities, or even across stores (which might vary in prices as well). Our marketing actions will be exploiting those differences. We need to understand every single detail of the business! I trust you for that J. You and Paul should actually present your results to the executive board.

• Identify groups of products that behave in a similar way. Michelle was saying that with your magic it’s easy to identify groups of similar products, and such groups will be super useful to evaluate the performance of our different campaigns. How many groups do you think we should consider? 5? 10? 20? Also, do you think we could find a “solid” approach to identify how similar are stores to one another? Would store clustering also make sense here? Could you also do that?

• Predicting sales at a store-product level, and to obtain aggregated sales per department/store/city we add up the independent predictions. Would that be still a valid approach? Let’s start with 28 days predictions (4 weeks)

• The Operations Department is already considering the application of your advanced predictive models for one of their most critical use cases: stores replenishment. Albeit if needed, the stock for some products can be replenished on a daily basis, the supplies to provision stores are mostly distributed on a weekly basis (beginning of each week). You’re probably already visualizing the importance of sales predictions for that. Minimizing the remanent stock is every retailer’s desire, but that desire is even stronger for supermarket items. Could you please draft a proposal detailing your solution to apply the sales predictors to the stores supply use case? We would also need to specify any extensions that the models might require, as well as the productization details for the use case. Martin requested the deployment of an API for its execution. Is it something you would be comfortable with? Do not worry about the implementation for now. I would like you to design a pilot test to confidently demonstrate the improvements directly associated to the new approach (i.e. dollars!). The creation of a data driven culture should be one of our priorities at this stage, and it is important that everyone starts believing in the power of this type of algorithms. I’m thinking of a case-control study (maybe by products or stores? maybe both?).