Revealing the Secrets to AI-Based Profit Optimization
Last month, I had the pleasure of talking to Hamish Taylor who is the former Head of Brands at British Airways, CEO of Eurostar, and CEO of Sainsbury’s Bank. As you can see Hamish’s leadership background in business is as diverse as mine in machine learning and AI. We revealed the Untold Secrets to AI-Based Profit Optimization, and below I will highlight some excerpts from our discussions.
Introduction
Since Profit = Revenue – Cost, the first secret to profit optimization is the ability to manage both of these terms in this simple equation. The second secret to profit optimization is recognizing the fact that estimating the cost through opportunity cost is often far superior to simply using the material cost of production. Finally, the last secret to profit optimization is to use AI to maximize revenue through price optimization without segmentation.
Profit Optimization through Cost Estimation
Accurate cost estimation is vital for profit optimization because it prevents sellers from selling below the cost and realizing negative margins. While material costs are sufficient for some products, other products and services like hotels, rentals, airlines, railways, cruises, cargo and shipping require more nuanced approaches based on market demand. When supply is limited, opportunity cost provides a better cost estimate that is based on the current business environment.
Revenue Maximization via Pricing Optimization
Because pricing is the most direct driver of revenue (even more so than sales and marketing), the most effective way to maximize revenue is through pricing. However, traditional segmentation-based pricing has severe limitations in today’s highly volatile markets, where supply chains are being restructured, product complexities are increasing, and everyone is having access to big data. These challenges include:
- data sparsity
- lack of cross-segment knowledge sharing
- unable to support continuous attributes
- heuristic and rules-driven price recommendation
Using the big data (all the attributes of the product, the customer, and the transaction) as inputs, a deep neural network can be trained to predict the willingness to pay (WTP) directly without segmentation at all. This transforms the segmentation problem into a prediction problem and addresses the first 3 challenges.
To address the last challenge of segmentation, we modeled the logistic win rate from historical transactions. Having the win-rate curve allows us to get the expected revenue (or margin) as a function of price, which we can optimize to obtain the revenue-maximizing price.
The Importance of Explainable AI (XAI)
Although this novel 2-step approach overcomes all the challenges of the segmentation approach, it also created a problem by using a deep neural network, which is a “black box.” However, we overcome this problem using explainable AI (XAI) techniques to make the black box more transparent. This is crucial for trust, adoption, and ROI.
The Origins of the New Approach
Another secret we discussed is that this revolutionary approach is actually inspired by the practice of revenue management (RM) in the airline industry. In RM, it’s standard to use a 2-step approach to first forecast future demands, and then optimize the price under limited seat availability. By leveraging this knowledge and combining it with AI-powered predictions, businesses can develop robust pricing strategies that are resilient even in the face of global disruptions.
Finally, we provided a couple of examples to demonstrate the power and benefit of this new approach. Both POC candidates witnessed over 20% error (RMSE) reduction in price prediction (step 1) achieving a very significant business performance uplift. These success stories highlight the potential for AI to drive long-term competitive advantage and generate substantial profits for businesses. It’s no wonder why so many businesses consider this as their secret weapon and rarely speak about it publicly.
The Full Video Recording
If you are reading this, good for you. You are obviously interested. And if you are interested enough, here is the full video recording of our conversation with some Q&A. I hope you’ll enjoy watching it.