Trade Promotion strategies continue to evolve from retailer-focused promotions to targeted shopper-focused promotional approaches. Promotional plan optimization is often impacted by numerous factors, some controlled by the manufacturer such as price, display type, and advertising type; and other factors that are out of the manufacturer’s control such as seasonality, market conditions, and competition. These factors, combined with other determinants such as data availability, account and product hierarchy, promotional types and frequency, result in the exponential growth of data volumes and complexity.
The common yet not so common
The complexity & size of data bring in a lot of other challenges which can affect the stability of any trade promotion optimisation platform. The most common challenge is the availability of sell-out data which is typically the POS data and needs to be sourced from direct retailers or syndicated agencies like Nielsen, IRI. While this path is the most accurate source for any predictive model development, it may not be economically viable for manufacturers (especially the tier2/3 manufacturers) due to the cost incurred in buying the data. This leads to the opportunity of using the sell-in data or shipment data which typically the manufacturer would store in their in-house enterprise systems. With few assumptions and model training comprehended by limited POS dataset a reasonably accurate model development can still be achieved for a sound promotion planning.
How much is sufficient
While the quality of sellout data is key to building a sound model, the history is also important. There are a lot of reasons why sell out history or for that matter promotions/event history is important. With rich data history, lot of variability in the data can be explained but again this needs the data to be backed up by causals and promotion history. Another key reason for historical data is the requirement for modeling seasonality, seasonality can help to time your promotions appropriately and also can act as a proxy for competitor price. However, a minimum requirement for including seasonality into the models is atleast 2 years of history (capturing 2 seasons). For new products, which would have limited history there are techniques available to model proxy equivalents with necessary model adjustments and assumptions.
Lost in transition
With the advent of machine learning techniques and advanced statistical models, data imputation and massaging has become manageable. But, building accurate models needs more care and rich data history and can go beyond imputation techniques. It may look trivial but managing product transition or SKU transition is a common data issue which many a times gets overlooked. With the transition, history of the product is mapped to the new product code and hence this can lead to multiple issues. There would be limited data history associated with the new product code and also promotion history cannot be mapped to the right product code. It is imperative to maintain the SKU transition history to ensure the correct mapping and also ensure the data is apportioned to the new code when the transition happens. This sounds trivial but is a vital cog in ensuring robust models.
While history and managing product transition is important, the biggest impediment to model accuracy is still those outliers. If the data foundation is built on outliers, no learning framework implementation can help in delivering the necessary business outcomes. Data outliers can be natural phenomenons like stock out periods, black swan period or an actual outlier, this can hamper the model development and render disastrous accuracy. There have been significant developments made in this area to create outlier handling algorithms based on exception handling but nothing to replace a strong data discovery backed up with the knowledge of business or domain experts.
Trade promotion optimization platforms can offer usual insights, but for that to occur it is imperative that data has to be managed effectively. For any Trade promotion optimization solution implementation and to ensure strong user adoption, a strong data foundation is quintessential.
Wipro Promax TPO has predictive planning capability – providing users the ability to manage data effectively with hypothesis based modeling techniques enabling to review systematically generated business outcomes & deliver the best combination of promotional inputs. Our data science and professional services teams are led by experienced domain experts to help you manage data & create predictive models to give your business competitive advantage through best practices in trade promotion optimization.
For more information, visit our website https://promax.wipro.com
For queries, write to us at WPAS-Promax@wipro.com
Author : Saju Ramachandran, Lead Data Scientist, Wipro Promax