Archimedes allegedly said “give me a lever strong enough and a fulcrum on which to place it, and I shall move the world”, many people think that with enough data and good compute power to process it, they can move the world. Sure. But just because you think you can, should you?
RoI and Action
AI costs the planet quite a bit, for example one chatgpt query consumes 2 to 5 litres of water due to cooling requirements of servers (1). Apart from that, it usually directly costs your organisation in terms of resoruces and effort. Even if you are running a non-profit, resources are limited, therefore you need the best bang for your buck.
I have argued earlier (in part I of this series) that any analytics activity should be tied to KPIs, whatever your organisation’s KPIs may be. If the resources spend do not bring you closer to achieving your KPIs, then there is a problem somewhere.
A second point to bear in mind is that the outcome should be actionable; analysis should not be a purely vanity or academic exercise.
Let’s start with this second point first.
Actionable
From my personal experience, most Analytical/ML/AI projects fail due to a very bad start; the business question to be solved is not defined properly, and even if that is done, often metrics to define success or failure are not agreed.
Motherhood statements
It is simply not worth spending efforts on analysis/AI/ML to solve motherhood statements, big, “feel-good” platitude with little substance or concrete meaning.
Motherhood statements are difficult to disagree with, because of the “feel-good” characteristic, but their lack of concreteness is what make it unwise to use Analytics/AI resources at them since it would be like throwing resources down a black hole.
A few examples:
We are committed to Innovation/Increase Innovation in the organisation
Sounds great, but what is innovation? Is it launching new products? Is it creating new profitable lines of business?
We aim to provide the best customer experience
What is best? And in what area? How do you measure customer experience?
We are committed to customer satisfaction
Again, sounds great, but how do you measure customer satisfaction? How does that tie to your KPIs, especially financially?
Analytics/ML/AI is usually designed to optimise something, and that something is usually declared explicitly. That’s the opposite of a motherhood statement. To solve this, define metrics clearly.
RoI
The other area is that the problem must be something worth solving for the organisation. This will increase chances of executive sponsorship and a clearer route to implementation. In commercial organisations, the things that are optimized are often related to the following:
· Revenue
· Cost
· Efficiency
· Risk
· User Experience
The key is that these must be defined using metrics and tracked, and do note that sometimes more that one of these will have to be balanced against each other, for example a financial institution may want to increase the volume of loans you issue, but this has to be balanced against the increased risks.
Conclusion
Data is powerful, and it can do a lot of good for anyone. However, it is important to focus resources in areas where the impact can be felt in line with the organisations’ objectives. It is crucial to start the analysis/AI project by setting the right goals, with measurable and achievable targets; only then will the benefits flow.
References :