Trade promotions are a massive expense for many consumer goods companies that often gets considered part of the “table stakes” to remain competitive in the market. Neal Analytics is challenging that assumption using machine learning to provide account and promotion managers with the knowledge of what’s working and what isn’t in their promotion calendars. Our trade promotion management and optimization solution clearly identifies which historical promotions have provided a positive ROI, leveraging that data to recommend future promotions that are the best fit for each product/market/customer, or other defining factors.
The Neal Analytics TPO solution is a rapidly deployable set of models and dashboards which:
Create a sales baseline to determine what sales would have been if a promotion wasn’t run:
Visualization to illustrate the uplift of a promotion on sales volume over the Sell Out period (Representative Data)
Calculate historical promotion ROI, volume uplift, redemption rate, and other KPIs using cost factors and sell-through data:
Illustrative calculation to demonstrate how we determine Promotion Profit from the Sales Baseline.
Sample KPIs. (Representative Data) Additional KPI development is available depending on your business needs.
Build optimal promotions given a set of desired characteristics:
Example interface for calling the Machine Learning API to predict the performance of a potential promotion. Final interfaces are developed to our clients’ specifications.
Our Approach to Modeling Baseline Historical Sales
For the baseline model, we need to isolate the effect of promotions on sales independent of other factors. To accomplish this, we use an advanced time series forecasting algorithm with additional regression components developed by Facebook and the open source community known as “Prophet” which we custom tune for your unique business factors.
By training this algorithm on historical periods where promotions didn’t occur, we can impute data to accurately predict what sales likely would have been had a promotion not been run. This will help your business understand exactly what value add promotions bring to your bottom line, separating promotion lift from seasonality, product assortment, market growth, and so on.
Example Datasets and Features
- Number of Sellers
- Wait Time
- Length of Customer Interaction
- Ad Spend
- Retailer Marketing
- Retailer Promotion Events
- Presence of Key Products or Product Groups
- New Products
- Product Lifecycle
- Length of Interaction
- Competition Sales
- Promotion Events
- SKU Overlap
Trade Promotion Management using Post Event Analysis
With promotions increasingly viewed as a cost of entry it is critical to inform management of the true performance of promotion activities and which ones are growing sales and market share. Many organizations do have some intelligence here, but a promotion with 100% redemption rate may be unprofitable and result in significant forward buying, suppressing future demand. Our visualizations provide an easily interpretable view of historical promotion performance as well as a variety of additional capabilities to analyze by Product Group, Customers, Geographies, etc.
ROI vs Spending Map visualization in Power BI. A variety of additional reports and dashboards come with the solution to contextualize the ML outputs and provide post event analysis of promotions.
Building Promotion Calendars with Machine Learning Recommendations
Our Trade Promotion Optimization solution extends your capabilities beyond typical TPM solutions by using the performance data generated by this first stage to feed a second machine learning algorithm to recommend the optimal promotions subject to the conditions and restrictions the promotion planners and key account managers require. This can be used to plan for an entire season in a calendar application custom developed to your business specifications.
Representation of a calendar application used for planning multiple future promotions. This optional application can be deployed as a web or desktop app and is custom developed for your specific use case.