Taking a Data-Driven Approach to Provisioning Alternates

Mihir Kittur, Co-Founder and Chief Commercial Officer, Ugam,

“The feedback that we got from our work on alternates was really positive. In fact, that was the most positive feedback that we’d gotten on almost all the network changes we made; and fortunately, as a result, our sales started to turn around. We started growing year over year again, and our customers were very pleased,” noted Ken Sauers, VP of Pricing and Business Integration at Essendant in his presentation at B2B Online 2018.
These are words of a happy distributor, but about a year ago, Sauers was telling a very different story. Essendant, a multibillion-dollar distributor of workplace essentials, had consolidated platforms and migrated their Jan-San customers to an advanced portal offering superior customer experience. However, this move wasn’t communicated well with their customers creating a poor experience. Post-migration, they rationalized their range and did not provision alternates for 80 percent of their items. They were losing sales and their NPS (Net Promoter Score) dropped to -20.

In reality, we weren’t meeting our customers’ needs. They had a hard time finding what items they wanted to purchase and when they couldn’t find an item, we weren’t giving them much information about what alternate items they could purchase instead,” noted Sauers.
To solve this problem, they partnered with Ugam to deploy a data-driven approach that would improve the ratio of alternates to each item. This process included five steps:
1) Workshops to define data dictionaries: We held workshops with merchandizing and category leaders to define data dictionaries (that is, understanding their data and how the data relates to each other). We leveraged their tacit knowledge and derived right terms for each attribute for each product. We differentiated defining and describing attributes to help refine customer queries further.
2) Obtained, reviewed and blended data: The “data cleaning” process was a crucial step to allow for success in subsequent steps. To make the data easily accessible and complete for the algorithm to work efficiently, we organized, cleaned and blended their data.
3) Established criteria and developed algorithms: To arrive at item matches at scale, we established criteria and developed algorithms using the clean data.
4) Reviewed and refined results - 31 out of their 400 categories were reviewed initially and refined to set the process right and scale out to the rest of their product categories. To achieve accuracy, we complimented the machine learning algorithm with human expert validation. With several iterations, our algorithm improved to provide a match score with the closest attributes of the matched product.
5) Provisioned the “right” alternates - Finally, we made sure that all the alternates for the items were active, well stocked and nationally available to the public. This in turn helped in improving the customer experience.
After finishing the work, Essendant conducted a net promoter survey of their customer base. “Our net promoter score improved dramatically. We had gone from scores in the –20 a year ago to consistently getting scores of +30 at present, so our customers were really pleased with the changes we made,” said Sauers.
As a follow-up to this project, Essendant is now using the data collected on alternate products to identify and offer “preferred alternates”. This will further strengthen their customer and supplier relationships. 
To hear the full story, watch our keynote from B2B Online 2018.


Sorry, your search did not match any relevant results.