Who fits the bill: ML or rules?
Amit Asawa
technology-driven enterprise problem solver on weekdays | social impact innovator on weekends
Companies are using software to automate more decisions and to augment certain human decisions, but many of them are inclined to choose AI, ML in particular, over traditional approaches. As a result, in the majority of IT portfolios, ML solutions are getting ubiquitous even to the places where simple techniques like business rules would perform far better.
Cognitive computing has come a long way but still possesses a few considerable challenges. For example, its cost has come down significantly, the maintenance of cognitive computing systems is still an expensive affair. ML projects are also riskier than traditional implementations. Most importantly ML projects generate a significant amount of technical debt comparatively.
Organizations, therefore, should try first to solve problems for favorable cost-benefit-risk trade-offs rather than taking ML as the default choice. They must know when to employ traditional programming and where to deploy ML to solve their complex business problems. To do so, they should be asking well-crafted questions to themselves. One of those questions could be :
" To solve a particular business problem, what kind of truth would be acceptable or desirable ? Literal or statistical "
If the answer to this question is LITERAL TRUTH then BRMS (business rule management system ), or any other similar conventional technique like IF ELSE statements, would outperform ML without any question. Cognitive computing should be a platform of choice if STATISTICAL TRUTH would be an acceptable solution and literal truth does not exist as a solution to the problem.
Literal Truth :
Business problems influenced or determined by laws, rules, regulations, and business policies are effectively solved by finding the literal truth. Healthcare insurance companies, for instance, would not like to verify benefits based on statistical truth or ML-based solutions they would need to know insured persons' eligibility for sure. Similarly, HR(human resource) functions that are strongly governed by corporate policies like C&B (compensation and benefits) are not a strong candidate for ML-based deployments either. Also, the business logic to replenish certain medical supplies such as insulin and blood sugar test strips make more sense to be coded in rules rather than embedded in ML algorithms.
Statistical Truth :
Forecasting, categorizing, and recommending related business problems can not be solved with 100% accuracy. For them, statistical truth is usually an acceptable solution. Thereby, traditional approaches are more likely to fail here than ML-based solutions. BFSI (banking, financial services, and insurance) companies, for instance, would not want business rules to flag fraudulent transactions. Similarly, online retailers would have a miserable time if they try to run their recommendation engines based on IF and ELSE statements.
In other words, business rules excel in solving business problems that can be solved with certainty (100 % accuracy), whereas machine learning is a proven technology to manage uncertainty (probability). Thus, the question, " To solve a particular business problem, what kind of truth would be acceptable or desirable? Literal or statistical ", can also be rephrased like :
" Can a business problem be turned into a prediction challenge ? "
If the answer is YES then adopt machine learning else explore other alternatives if not business rules.
In short, the misconception that probabilistic technologies (ML or deep learning) has the potential to outperform deterministic technologies (rules or traditional computer programming) in nearly all scenarios is simply not true. Enterprises must watch for inflated expectations and fear of missing out on AI value to avoid hasty and poorly thought-out decisions when it comes to AI/ML selection over traditional approaches. For better decision making, they should strive to know which particular technology fits the bill for their purpose - asking questions perhaps is the best way to do so.
Find out more perspectives on the power of asking questions in the entire AI adoption journey :
Send me your thoughts, questions, observations, and predictions by hitting "Add a comment" to this post. Differing perspectives are always welcome do not hesitate in posting them as well.
Thank you for reading my post. I regularly write about the newest digital technologies and digital transformation initiatives. To read my future posts simply click 'Follow'. Feel free to connect with me as well.
Also, you might like to read more about cutting-edge digital technologies in Artificial Intelligence: The Star of the Digital Galaxy: A study of Digital Disruption, Innovation, and Economic Transformation. It's packed with real-life examples and intended to serve as a primer to simplify and explain the concepts, implementations, and implications of the AI-powered digital galaxy.
About Amit Asawa
Amit Asawa is a strategic business & digital technologies practitioner and advisor. He helps enterprises improve their business performance by seeking, evaluating, and implementing technological advancements. He has extensive experience in IT implementations, digital optimizations, transformations, and modernizations including systems integration, business process re-engineering, and organizational change management.
Note: This is the author's personal opinion. This content has not been read or approved by a current or former employer before it is posted, and does not represent their positions, strategies, or opinions.