Challenge
A North American retailer wanted to better understand how their revenues were being driven by various programs in both Sales/Operations and Marketing. They were looking to understand the difference in impact between New vs. Returning Customers to be able to customize their strategy.
They needed models that could evaluate a broad range of investments from those designed to improve customer experience, to those focused on improving fulfillment and performance marketing, as well as more traditional investments reinforcing the brand.
Given the advent of hyper-local marketing for retail, Marketing also wanted a very granular view on how programs and elements within channels were working at the channel and sub-channel level view. Yet, for Sales & Operations this meant modeling at store and department level in order to understand the impact of supply side factors and customer service improvement programs.
The client sought out a transparent analytics partner with a collaborative, on-site approach in an embedded, DIWM (Do-It-With-Me) fashion.
They needed quarterly model refreshes, thus a rapid speed of execution was crucial to meeting timelines and delivering results at pivotal planning times.
Solution
Data Management
Three years of daily data was collected on sales (split between new vs. returning customers), operational and marketing activities at the store and market level in order to get granular insights into campaign and messaging execution.
Sub-channel program and campaign data collection, although difficult to collect and scrub, was crucial to garnering the tactical insights needed to make fast decisions.
With campaign level data in many disparate forms and sources, it was critical to create a taxonomy of campaign dimensions and characteristics so that the data could be organized and aligned.
Modeling
A set of Hierarchical Bayesian models were built within the Marketscience Studio platform utilizing the daily, store and market level data, and accounting for external factors, to provide detailed channel, category, and message insights.
Store level is the natural point of most sales/operations programs, and market level is the natural level for most marketing programs – a hierarchical model is able to mix both levels of data.
Although the models were built with granular high frequency data, they also included components to understand long-term dynamics of customer and brand experience.
Results
The resulting granularity and consistency of insight from the models and across programs provided actionable optimization shifts via the Marketscience SimOpt.Studio optimization tool.
This allowed the business to implement spend changes quickly and efficiently across sales channels, media channels and sub-channels.
Some of the recommendations included:
- Minimum spend levels identified to continue attracting new customers which has a snowball effect for demand from retained customers.
- Baseline (incl brand health, seasonality etc.) and Discounts found as key non-media drivers, at varying levels between departments. With Women’s clothing and shoes attracting most demand, a carefully tailored brand and promo strategy was recommended for these lines of business.
- Spend reallocation from channels experiencing high diminishing returns such as Affilitates, App Installs and to a lower extent YouTube, onto highly performing and underspent channels such as Display and Paid Social.
- 12% increase in sales from reallocating the budget to more efficient digital channels.
What We Did
- BaseDynamics Hierarchical Bayesian Modeling
- Simulation and Optimization with SimOpt.Studio