HOW DO YOU KNOW IF A PROJECT IS SUITABLE FOR AI?
In the last years, a lot of new technologies are coming up: Blockchain, Virtual Reality/Augmented Reality (VR/AR), IoT, Quantum computing, and of course Artificial Intelligence (AI) and Machine Learning (ML).
Each of these are changing our lives in different ways and are impacting every industry.
In this post we will focus on AI/ML.
With Recommendation Engines, Chatbots, Predictive Maintenance, Fraud Detection and other interesting and useful implementations almost ready to use, companies are thinking about how to enhance their processes. But how do you know if a project is suitable for AI/ML?
When is it right to use Artificial Intelligence?
A common misconception about AI/ML is that it works like a magic tool that allows people to automate everything and drive the exponential growth of the company.
This concepts lead to the wrong assumption that AI can extend and improve any enterprise process, leaving apart the true and deeper principles about how to implement AI correctly and if it’s right answer to the business strategy.
Just to recap, Machine Learning, a subset of AI, “is based on algorithms that can learn from data without relying on rules-based programming”.
McKinsey& Co. It can be divided into three main categories:
• Supervised Learning – algorithms that learn patterns from past labeled data used to predict outcomes on new data (i.e. predict if the email is spam or not);
• Unsupervised Learning – algorithms that learn hidden structures on unlabeled data (i.e. clustering of customers);
• Reinforcement Learning – algorithms that interact with the environment through actions and produce rewards to improve their behaviors.
The main questions that most companies are facing are related to the type of business problem they want to solve with AI/ML and the definition of that problem based on one of the three main ML categories above.
After delving into these questions, here are some guidelines. AI/ML is useful when your problem:
• Scales-up fast – when this problem can be applied to other similar contexts (i.e. a predictive maintenance model applicable to identical / similar machines within a factory environment)
• Needs to be adapted in real-time – if data change rapidly, rule statements are not the most flexible option (i.e. ML models for customer segmentation in marketing, that reflect changes in customers’ behaviour)
• Handles very complex rules – not solvable with simple ruling statements (i.e. ML models identifying anomalies on credit cards transactions)
• Does not require perfect accuracy – an ML model will never reach 100% accuracy, so it is not the right choice if your process cannot stand even a very little uncertainty (which means that the human intervention is needed to confirm and take relevant actions and decisions)
Moreover, AI/ML problems need a good quantity of data to be effective. These data should be:
• Relevant – useful for the problem that we need to solve
• Representative – representative of the population and unbiased
• Enough – the algorithms require a lot of data to learn patterns and generalize better on new data
• Secure – protect the privacy of the users and stored securely
Our suggestion is very simple!
Don’t force your problem into being 100% solvable with AI/ML: the risk of wasting time and resources is behind the corner.
Nevertheless, if the business goal meets these criteria, you’re on the right track to empower your process with automation and Artificial Intelligence.