Operationalising AI: whimsical dream to grounded reality

Let’s look at some usecases where AI application moves beyond one-off insights and can be operationalised – packaged into an app and used enterprise wide, getting embedded in the daily decisioning process.

Preparing for unpredictable order volumes

First, let’s look at how companies manage unpredictable order volumes. A major consumer packaged goods (CPG) customer of ours was facing the challenge of very spiky, unpredictable ordering behavior by a major ecommerce retailer. This was constantly resulting in elevated chargeback penalties due to stockouts, or excess stock, increasing inventory risk. The traditional statistical forecasting approaches it was using were failing to deliver the needed predictability to manage this risk.

However, the retailer in this context provides a wealth of information beyond traditional time series data, such as the product searches, customer page views and in-stock availability. While much of this data is not typically housed in an ERP system, which is the foundation for traditional planning systems, using AI and blending the data from traditional and non traditional sources, an app was built and deployed to the planners for the ecommerce channel.

This provided them with visibility into the next order quantity with probability factored in. The app also elevated the “cut” risk (order not filled in full), while making prescriptive recommendations such as stock transfer orders and production changes to ensure product availability.

Performing chargeback analysis and determining rootcauses

Our second CPG customer example looks at the struggle to meet ever shrinking retail customer delivery windows which causes increasing chargeback penalties from strategic customers. Upon close evaluation, much of the data the company had access to was not being effectively used. This included data on when the order was placed, manufacturing status, inventory status, when the order was changed, when it was picked and delivery information. 

With such an abundance of data, this turned out to be a great case for AI. Instead of manually piecing together information to understand a single failure or group of failures and ascertain the root cause, AI helps identify patterns in the data. This is the modern take on  the classic 5-Why approach.  AI goes through the first 3 or 4 Why’s automatically. It pinpoints the full sequence of events that led to a failure. Once it finds these patterns, it is easier to fix the root cause and predict future failures.  But, just running this analysis once is not enough. Hence an AI powered app that is refreshed with the latest data was deployed to make the process repeatable.

Embarking on your AI journey

While the power of AI can be transformative, it is important for organisations to filter the hype from reality, and keep in perspective that AI is the means to the end rather than the end in itself. Given the rapidly evolving nature of AI, senior leaders should consider engaging external partners with a proven track record in kickstarting initiatives.

Companies should start with a burning problem that can be solved by applying data intelligently. Trying to perfect the quality of data and AI models in the first go can be the enemy of speed to value. Instead, an agile approach of experimenting, learning, and adapting on the go, starting with the data already available (with all its limitations) is a practical approach. Quantify and communicate the benefits throughout the organisation so excitement can be built around AI for more use cases, and to stimulate additional ideas. By taking a pragmatic approach to deploying sustainable AI, organisations can unlock tremendous amounts of value that is simply not possible with traditional approaches.