Optimal planning with AI
How a leading US retailer used AI to improve marketing spend.
Every year, a top American retailer sees some $14-15 billion in marketing-driven sales, which means decisions on how to allocate marketing dollars—and specifically, media spend—aren’t taken lightly.
But using historical data to decide where to spend among the dozens of channels available—from traditional TV to Tik Tok—isn’t easy. The data is often stale by the time it’s available to analyze, and the number of new channels and platforms grows all the time.
With so much money at stake and the difficulty in getting quick answers, increased speed and agility were at the top of the retailer’s wish list, and the company issued a challenge to Accenture: To get more specific, actionable insights faster.
Accenture partnered with the retailer to design an AI-powered solution that would enable faster and better data collection and more precise modeling to optimize media spend. The first task was speeding up the existing data flow process, then aggregating and processing all the data from media channels, sales and spend that fed the measurement model. By customizing AIP+, Accenture’s pre-integrated AI services and capabilities, to do the data aggregation, we helped cut the existing process by 80% using automation to accelerate processing and validation.
With data flow addressed, the team looked next to alter the underlying model that produced the measurement. Previously, these models were hypothesis-driven, i.e., people would painstakingly hypothesize every possible interdependency between different channels. New machine learning was introduced to the process, helping to proactively identify those interdependencies between channels that potentially drive sales. With the new monthly cadence, the team could refresh the models every month, iterating from the previous month’s model instead of starting from scratch. By hosting deep-dive training sessions for employees on the modeling methodology, the team offered them transparency that earned buy-in and trust in the solution.
The number of marketing channels included in the modeling was increased nearly 40%, allowing them to thinly slice the data (for example, by breaking out a catch all “social media” channel into each social media platform).
The results were significant.
The solution shortened the lag between the measurement period and performance insights from five months to five weeks, opening up a 10 and a half month planning runway for the same period the following year. Also, going from one annual measurement (where performance was expressed as an average) to monthly measurements meant that insights were more nuanced, so the team could see how one channel or another might vary in performance throughout the year.
Even more concretely, the team estimates that $300 million in media buying opportunities and value creation was unlocked by implementing the new tool. This meant the team could spend the same amount on media and generate an additional $300 million in sales.