Background
CPG and retail companies rely on data to guide sales team efforts and optimize revenue. Given the pandemic’s recent impact on bars and restaurants, this has become particularly important for the alcoholic beverage sector.
When a product sells well in a particular market, it should also sell well in an underserved similar market. Sales teams can make a big impact if they’re able to identify these opportunities and then focus on expanding distribution or placing more opportune brands and products on the shelf in existing accounts. The key to identifying these opportunities is in creating the like-set of accounts or markets to drive the comparison.
Machine learning can help create these like-sets. Unsupervised learning, such as clustering algorithms, are naturally suited to creating like-sets of accounts or market areas. Supervised learning algorithms can also apply custom segmentations that may already be aligned to particular sales strategies to help organize sales coverage across the entire account universe.
Cosine similarity and other comparative calculations can quantify the difference of a particular account from the segment average or top performers. If we can flag an account or a particular brand that is underperforming compared to its segment, passing that information to the corresponding sales representative is the ammo needed to spur a discussion on missing out. FOMO is an effective sales lever.
Challenge
Developing and implementing an end-to-end ML-driven pipeline to segment accounts, identify opportunities, and create a targeted list of opportunities for nontechnical sales personnel has many technical and practical challenges.
Identifying viable data sources for active accounts, current sales, market demographics, and recent sales team activities has always been hit-or-miss. Combining these data sources to effectively drive the segmentation algorithm requires complex data structuring and feature engineering. For larger clients, data volumes require specialized architectures.
Identifying actual opportunities may lead to many spurious results, requiring market knowledge to filter and target effectively. Rolling out recommendations to sales teams requires automated distribution pipelines or effective CRM platforms.
Solution
Keyrus uses the latest technologies to help our clients realize these segmentation solutions.
For large data volumes, we use cloud-based ETL and ML services (Glue, Data Factory, Sagemaker, Databricks) on AWS and Azure.
For more manageable data volumes, or for organizations who don’t have cloud-based skill sets yet, we kickstart our solutions using end-to-end data science platforms such as Alteryx and Dataiku, which provide visual ETL and ML programming capabilities.
Our segmentation results are carefully integrated into downstream reporting and CRM platforms to help sales teams consume the recommendations and opportunities.
Results
Our solutions have impacted our clients in several ways. At a large spirits and wine importer, we’ve saved several days per month of account tracking and targeting work that the sales team was previously doing manually.
At a spirits manufacturer, we immediately identified over $650K in possible sales at underserved accounts across the broader US region. The team worked to increase distribution in these accounts by 30%.
At a beer importer, segmentation allowed the company to identify effective sales activities that were working well in one market, but not fully implemented in other like-markets. The company reallocated several million dollars into more effective trade marketing activities.
Segmenting accounts or markets is an effective way to create comparative sets for sales teams to better place products in existing accounts, or increase distribution in underserved areas. While the actual segmentation is only a piece of the end-to-end project, the latest cloud and data science platforms make it far more accessible for any organization to quickly realize these solutions.