The choice of approach for each decision product must be established from the start and will depend on the pattern most suited to enabling its users. Even with automation, humans are never completely out of the picture, instead moving to a “managerial” role where they curate the possible outcomes and assess performance. In fact, AI should be thought of as a co-worker, and we believe partnering with AI is the best way to apply it. The ultimate situation would be to get to a position where people and machines are learning from and enabling each other. 

Dynamic pricing is a good example of this. To be competitive in the world of e-commerce, pricing needs to be dynamic. Brands must be able to respond to changes in demand, perception and competitors in real-time, updating their pricing by the hour, by the minute, or even by the customer. Traditional manual and rule-based pricing systems can’t handle this level of speed and can lead to 5-10% in gross margin being left on the table. 

Automating this process by using reinforcement learning (RL) to optimize the price in response to rapid fluctuations in demand can drive up profits. While this automation looks to transform the cognitive load away from manual pricing updates to the more important strategic work of what to sell and when, it’s critical to ensure this is actually the case for pricing teams. If they don’t understand or, more importantly, trust the pricing recommendations then they will continuously reject or adjust them, creating an alternative manual, low-value workload. Ensuring the tooling is designed as a decision product, optimized for use by the pricing team and not just a metric, will ensure successful uptake and increased margin. 

The decision journey

Along every business process, there are numerous decisions that need to be made. While decision products use analytics and AI to optimize each of these decisions, they only consider them individually, in insolation. This can lead to competing priorities as each product tries to optimize within its own scope, not considering the wider implications. 

Returning to the dynamic pricing example, the algorithm will use feedback from purchases to optimize pricing. If the customer didn’t buy the product at the recommended price then it’s not an optimum price. The customer, however, might have been influenced by other aspects of that process. For example, a marketing product could be running a campaign for an alternative product at a discount - creating competition that unnecessarily drives increases in marketing spend and decreases in margin.  

To prevent this, systems thinking can be used to consider the process holistically, creating a journey of decision-making along it to deliver end-to-end value optimization. This will involve striking a delicate balance between striving for the global optimum and local optima. At points along the stream, it will be beneficial to optimize just for the decision in scope, at others, the overall goal will take precedence. 

We call this a decision journey -  a framework for integrated decision-making across an entire process. Each journey is made up of four constituent parts: 

  • Process: the sequence of activities that deliver value to a stakeholder. It can be external (e.g. a customer journey) or internal (e.g. a value stream). 
  • Decision points: the distinct decisions that inform how the process is executed or improved upon. 
  • Decision products: the collection of intertwined models and tooling improving the decision-making process.
  • Data products: the products that surface the data required to make decisions.