This post was first published in 2021.
Machine Learning and Artificial Intelligence (ML/AI) prove to be similar to all other technology tools and techniques. Implementing technology is a challenge, but making sure technology works for the end-user and provides value is much more difficult.
In the last three months, we have come across three excellent articles that reinforce this point. Some of the content is challenging, but when you read through it, you understand that the focus should be on the end-user, not the technology:
Expanding AI's Impact with Organizational Learning - MIT Sloan Management Review, October 19, 2020
The Secret to AI is People - Harvard Business Review, August 24, 2020
Data Alone is Not Enough - Andreessen Horowitz (known as "a 16z"), October 22, 2020
An End-User Centric Approach is Needed
Our own experience with Financial Services clients has demonstrated this challenge. Applying ML/AI can happen relatively quickly, but turning that into process improvement and ROI has been a rocky road. We are building a product for a large financial data company that requires extracting high-value content and data from credit agreements that are part of SEC filings. Before harnessing the full power of an ML/AI solution, we took steps to deliver immediate value to our end-users, the financial analysts. In so doing, we put the proper data governance in place to ensure that its quality and quantity will more easily become part of a robust ML/AI solution in the next twelve to eighteen months.
For another fintech client in the college financial planning space, we did some initial work to automatically process students' offer letters. In this case, the client quickly realized that even though the technology exists and can be impressive, the end-user (families figuring out payment) is not impressed by the wow factor unless it makes their life easier.
Like most other technologies, ML/AI is not a silver bullet, so it is imperative to stay focused on the customer's needs. Challenges around implementing proper data governance and structures can mean several months of work before realizing the full benefits of an ML/AI solution, so it is essential to have clear short-term goals that demonstrate value along the way.
Deep Learning vs. Training Model
ML/AI can mean so many things. We divide the space into two segments for simplicity.
An example of Deep Learning is Natural Language Processing. Another example is data-based personalization. Deep Learning is complex and very difficult. Fortunately, many companies have spent billions of dollars developing it and have created APIs so that organizations can use it for pennies. Deep Learning can successfully be applied when it meets end-user needs. For example, figuring out what entities and values are in a paragraph text can be done successfully. Another successful application would be to figure out related content based on an article an end-user is reading.
ML/AI where you have to create a custom model and train it is not yet as simple, but there are tools on the horizon. AWS is continually improving its SageMaker platform, and C3's AI Ex Machina is still in Beta, but both will soon lower the barrier to access these powerful techniques. When training ML/AI models, organizations need to be aware of what it will require and the challenges ahead.
As always, the focus has to be on the end-user. It is crucial to only use ML/AI when it can help, and not to force ML/AI on your organization because it is the shiny new thing.
Like what you read?