Best practices for preparing your infrastructure for machine learning and AI
Machine learning and artificial intelligence are all the buzz, and implementation of ML and AI in organizations is increasing. Research firm Narrative Science and the US National Business Research Institute found that a majority of companies have already implemented AI, mostly for smaller projects, and implementation continues to accelerate.
There are strains around moving from research to implementation for ML and AI. Research can take place on dedicated hardware and software, with dedicated teams. The investment required might sound expensive - millions of dollars, to tens of millions of dollars, in larger organizations - but the amounts involved are generally a small fraction of an organization’s IT budget.
Research and development in ML and AI does not interfere much with existing operations. But, as organizations move toward implementation, things can get much tougher. Implementing ML and AI is somewhat pointless if the underlying IT infrastructure is siloed, slow, or already running at or beyond its overall capacity.
The best way forward is to make investments that serve today’s urgent needs while preparing you for the future. You can significantly improve your organization’s IT infrastructure, and the results you achieve from current IT operations, while also laying the groundwork for the potentially revolutionary benefits that you can achieve by implementing ML and AI.
Here are some changes - attitudinal and operational - that you need to undertake today to get the most out of ML and AI in the near future. Starting now will pay off today and for many years into the future.
The first change is attitudinal. The opportunities in ML and AI are indeed revolutionary, but each step of implementation is going to be evolutionary.
In fact, for most organizations, the most important step in moving to ML and AI is developing expertise in, and implementing, predictive analytics. The key value that you can derive from ML and AI is in improving the insights that your business derives from customer interactions, and in using those insights to improve each customer’s experience in dealing with you.
Predictive analytics is key to unlocking this value. And, while predictive analytics sounds less revolutionary (and less exciting) than ML and AI, there’s actually a chasm between traditional analytics - reporting and business intelligence, which most organizations are good at - and data science, which includes predictive analytics and its accelerants, ML and AI.
Standard reporting and descriptive analytics are table stakes on the way to ML and AI (figure courtesy Data Science Central)
The attitudinal change that you need to make is to understand and grasp this difference. If you have not yet perfected your reporting and business intelligence game, do that first. These are well-understood and even standard business practices today, so making them work well in your organization shouldn’t be too difficult. If it is, the changes recommended here will cut costs and improve performance for every stage of analytics implementation.
The key change to effective analytics today is to move from a siloed data approach to a streaming data approach. And the most effective technology for starting this change is the leading technology for real-time data pipelines and streaming data applications: Apache Kafka.
Think of Kafka first as a uniform pipeline for moving data from one part of your organization to another. Kafka has become so popular that databases and other messaging apps, both old and new, usually have Kafka connectors. By standardizing on Kafka as your organization’s data bus, you can steadily normalize data inputs and outputs throughout your organization.
Then, think of Kafka as a way to implement streaming data throughout your organization. You can put your operational users, such as the people who run and depend on transactions, extremely close to incoming data. You can then cost-effectively move ML and AI into these operations, improving organizational performance.
The bigger opportunity for ML and AI, though, is in predictive analytics - quickly deciding what business users and end users are likely to need, and proactively providing it to them. To get the most out of predictive analytics, you need a fast, capable database that supports high levels of concurrency - a truly modern database.
Such databases have been variously described as translytical (Forrester), HTAP (Gartner), HOAP (451 Research), converged infrastructure (various sources), and modern (various sources). NoSQL databases, though relatively new, don’t meet these requirements.
You should take two steps in preparing your infrastructure for ML and AI. The first is to assess your current competence across the areas of reporting, business intelligence, predictive analytics, and operationalizing ML and AI. Fix what’s broken and prepare yourself to move to the next level.
The next step is to consider implementing Kafka, or another standardized messaging bus, across your organization. Once you do that, look for databases and related technologies that will help you move toward streaming data across your organization, maximizing the value of your investments in ML and AI.