Building a Scalable Data Science Solution to Outperform Sales Execution in Traditional Trade

RB is a multinational consumer goods company with more than 40,000 employees operating in 60+ locations and a portfolio of leading brands such as Airborne, Air Wick, Clearasil and Lysol. RB serves the ‘traditional trade’ markets globally which are a complex network of more than 1.2 million small retailers, corner stores, open markets and street vendors. This makes it difficult to drive effective sales strategies in a competitive market due to limited range, disparate data, and high attrition. To overcome these business challenges, RB developed a solution that analyzes years of buying patterns and market specific data, ties them to sales strategy and generates a weekly sales order at the individual store level. Using the scale-out compute power on Azure Databricks enabled us to quickly deploy the solution across multiple markets where we are able to process orders for up to 50,000 stores per hour. In this session we will share our approach to building this solution.

展开查看详情

1.WIFI SSID:Spark+AISummit | Password: UnifiedDataAnalytics

2.A Scalable Data Science Solution for Sales Execution in Traditional Trade Markets Harish Kumar, RB #UnifiedDataAnalytics #SparkAISummit

3. To make a difference by giving people innovative solutions About RB for healthier and happier homes People Consumers Shareholders Communities Environment 40k+ 20m+ 132% 765m 61k Products sold daily Total Shareholder people informed tonnes of CO2e Return since 2012 through health and saved from the hygiene initiatives purchase and generation of renewable electricity #UnifiedDataAnalytics #SparkAISummit 3

4.Traditional Trade Markets Central & Eastern Europe Russia 86,565 Turkey 3,259 Japan Egypt 21,319 15,594 MENA Thailand 19,433 22,634 Malaysia 80,403 India 400,403 Indonesia Pakistan 80,443 141,487 Bangladesh Brazil Sri Lanka 71,539 100,000 52,151 Argentina South Africa TOTAL STORES 11,369 4,608 ~1 275 949 The volume of stores and complex logistics limits the range to drive central strategies 4

5. Instore execution challenges Customers Instore Good Margin Knowledge proposition 60+ Househ olds 20+ Reps each day No from different Inventory /Cashflo Mfrs w system Wholesaler A good proposition and in-store knowledge is key to compete for fair wallet share 5

6. 2 01 7/0 1 2 01 7/0 1 602 2 01 7/0 2 2 01 7/0 2 C B Oth er s 1,460 Mo rtei n Othe rs Mo rtei n Ae ro sol 2 01 7/0 3 2 01 7/0 3 2 01 7/0 4 2 01 7/0 4 629 2 01 7/0 5 2 01 7/0 5 312 C B Pas te 2 01 7/0 6 2 01 7/0 6 Mo rtei n C oi l 2 01 7/0 7 2 01 7/0 7 R ob in B lu e L iq ui d 312 2 01 7/0 8 2 01 7/0 8 716 2 01 7/0 9 2 01 7/0 9 211 2 01 7/1 0 2 01 7/1 0 128 H arp ic Ba thro o m 2 01 7/1 1 2 01 7/1 1 208 R ob in L iq u id B le ach Mo rtei n L ED C om pl ete 2 01 7/1 2 2 01 7/1 2 Assortment Pattern Buying Pattern Sample Store 2 01 8/0 1 2 01 8/0 1 190 Disparate Buying Patterns 2 01 8/0 2 2 01 8/0 2 167 H arp ic L iq ui d 2 01 8/0 3 2 01 8/0 3 750 Mo rtei n L ED R efi ll 98 2 01 8/0 4 2 01 8/0 4 2 01 8/0 5 2 01 8/0 5 174 2 01 8/0 6 2 01 8/0 6 1,351 2 01 8/0 7 2 01 8/0 7 3,486 2 01 8/0 8 2 01 8/0 8 1,422 2 01 8/0 9 2 01 8/0 9 2,141 Sporadic buying patterns result in bad inputs for traditional forecasting models 2 01 8/1 0 2 01 8/1 0 390 2 01 8/1 1 2 01 8/1 1 272 6 2 01 8/1 2 2 01 8/1 2 1,292

7.Solution Objectives Retain instore knowledge and experience Smart Order ü Drive Sales growth Ensure Sales strategy is built into orders ü Improve product range ü Reduce Attrition impact ü Give more Control to HQ Fragmented situation managed by AI Augmenting Reps to deal with executional challenges 7

8.Smart Order AI Engine Buying Pattern Revenue Smart Order Optimizer Seasonality Seasonality Promo Adhoc Incentive Initiatives Recommend Volume & Plans Optimize for Highest value Store Profiling Central & Clustering Margin Strategy Optimizer Recommend Assortment Phasing Sales Rep Hand Held Category Max SKU Optimize for Contribution Assortment Ranking Highest GM Objectives & Constraints Smart Order has become the medium for executing ALL field Sales actions 8

9.Solution Landscape PREP & MODEL & INGEST STORE INSIGHTS TRANSFORM SERVE On-premise Azure DWH Local Data Azure Blob ADLS Databricks SaaS Data DATA SCIENCE CONSUME Azure Data Factory “2x performance at ½ x price” 9

10. Process Flow Data Clustering Prep for Optimization Preparation Optimization and Results § Channel – Class § Channel-Class § Clusters of § Store SKU list is Data Preparation level stores are stores are optimized clustered further put into § Top Categories together based grids/sub- § SKU volume selection on volume, clustered based limits are assortment and on potential modified based price across on Trade § Data Cleansing categories § SKU ranking Offers/Seasonal and SKU ity volume limits are estimated § Store Volume per category limits and other constraints are § Assortment used to limits are optimize and estimated per maximize category revenue for each store 10

11.Solution outcome Pre Smart Order Post Smart Order Monthly targets and Sales Rep level Daily targets at Store level Channel level MSL Store level MSL Many KPIs to track: Only 2 KPIs: NPDs, Numeric & Merchandising Smart Order compliance for Value drives, MSL, MTD target and Range achievements, Phasing ü Drive IMS, Improve product range, Reduce Attrition impact, more Control to HQ 11

12.DON’T FORGET TO RATE AND REVIEW THE SESSIONS SEARCH SPARK + AI SUMMIT