AI and ML: Focus on the Customer, not the Technology!
As a Data Scientist, it can be easy to get carried away with the abundance of AI algorithms and the clever ways to solve problems. It’s no crime to be a tech fanatic, but it’s important to step back from the technology and understand the business value being delivered. One of the most crucial questions to ask when implementing an AI solution is “Will my client benefit from it?”
Yes, some algorithms will create an instant “wow factor” and build curiosity, but after the initial buzz there must be a clear roadmap to turn the vision into reality. Many solutions have fallen short in this long hurdle race due to various factors that were overlooked during the fascination of the idea.
More often than not, the problem has not been examined from a business perspective. Thus, it is important to understand the client scenario, requirements, and their capabilities before a solution is pitched.
Here are some “do’s and don’ts” to help you create client-centric solutions where technology is just an enabler:
- Create a model that the client cannot host: It is futile to create a complex, multi-layered neural network to predict customer churn if the client does not have the compute power to host it on-premises and has no desire to move to the cloud.
- Create something that is beyond the client’s analytical maturity: A model that performs “AutoML” may not be too appealing for a client who does most reporting work in Excel. Yes, they may be impressed by what you have created, but old habits are difficult to kill and an organization-wide cultural change will be required to move away from Excel and adopt more advanced techniques.
- Create a complex black box model for a simple problem: You might think that a client would be thrilled if they asked for a Ford Focus and received a Ferrari, but you would be surprised. More advanced models usually lose out on interpretability, and if a client cannot explain the decision-making process to their senior stakeholders, it may not go down well in board meetings.
- Choose a simple solution to solve a simple problem: Occam's razor states that the simplest solution is usually the best. If the problem is not too difficult, don’t overengineer the solution. Go with the simplest approach and the client will be delighted with the value it brings.
- Design a solution that the client is confident enough to use: Whether it is prioritizing accuracy or interpretability, ensure that the client is comfortable with the outcome that the model yields. If they want to see why a decision was made, use decision trees instead of SVM (Support vector machine).
- Always ask the client what they want!: When a business problem is being shaped, it is important to ask as many questions as possible. That way, you cannot fall into the trap of creating something that isn’t usable on day one.
James Adams is a Data Scientist in the UK&I Digital and Analytics Practice team at Atos.