If you are reading this, chances are that you are responsible for technology and operations for a Retail or Supply Chain business. Perhaps you are also considering an AI-driven tech project, wondering if it would make economic sense, and where it might falter.
While the ROI equations are fundamentally the same, the factors that drive the costs and the benefits of AI are markedly different and unique. So, we look at the classical costs and benefits factors, but we also have learned that the nature of AI technologies is such that its success and risk factors are not clearly understood yet. So, what are these?
We find there are five distinctly unique factors that determine the economics of an AI project.?
- Cost of hyper scalars - Much of the practical implementations in the world of AI is happening because enterprises, product developers, and third parties are able to leverage out-of-the-box capabilities of cloud-based hyper scalars like AWS, Google, Microsoft’s Azure, Open.AI, or the likes of Prophet from Facebook. It is very attractive for developers to start using hyper scalars at a near-zero cost, and the reach is provided to the potential users. However, as the usage increases the free tiers are consumed fast and escalating costs start hitting the projects. Therefore, businesses should evaluate AI projects for their long-term investments, strategic architecture, and infrastructure, and target the big picture instead of solving point problems.
- Multi-phase investments - All things big and complex are perhaps addressed with an iterative approach. What adds to the complexity of the AI projects is their probabilistic outcome. Therefore, as the projects proceed, it is absolutely critical to evaluate the desired versus actual outcomes, realign goals, and accordingly invest in the next phase.
- Facing disruptions - The financial model underlying the AI investment should be designed to withstand disruptions. Because, if you are building an AI system, you are not only a potential source of disruption, but you constantly face the risk of facing one. The world of AI tech is inherently so due to the unprecedented scale at which it is driving automation, the new ways of doing things, and the speed with which it is changing. Therefore, from an economic perspective, one should seek to consolidate and capitalize on the business gains with each phase. For example, in a 4-phase project at a CPG whose objective is to increase the margins earned per SKU, each phase should address a distinct product line whose supplier planning, and demand planning can be optimized, and therefore monetized, within that phase.
- Establish the required accuracy level - AI technologies, ML models in particular, inherently have probabilistic outcomes. In other words, with a predictive model, there is a confidence level attached to the predictions it makes. Therefore, depending upon the business need it should be established how much of “inaccuracy” is acceptable to the business. For example, perhaps in the world of patient care, a failure rate of 0.1% may be deemed unacceptable, but in the world of inventory management, a 5% error rate (i.e. correct replenishment prediction 95% of the time) may be accepted as great. This is totally antithetical to the way computerized systems have been perceived so far.
- Mitigate data obsolescence - Machine Learning models are parameterized based on the data they are trained with. So, not only would the biases of the training data creep into the model, but as the ecosystem changes, the stimuli to the ML models are also changed. Therefore, without investing in re-training the ML models, a healthy ROI will turn into an error-prone loss-making venture.
