It Takes More Than Luck for High-Impact Machine Learning Solutions: The Four-Leafed Clover of Success

Four things you need to define before putting machine learning solutions in place

It Takes More Than Luck for High-Impact Machine Learning Solutions: The Four-Leafed Clover of Success

It Takes More Than Luck for High-Impact Machine Learning Solutions: The Four-Leafed Clover of Success

Juan Gorricho

Juan Gorricho

Juan F. Gorricho is currently group executive, data & analytics, global product innovation at TSYS.

More Info

There are two types of people in the world: those who use machine learning for the right reasons and those who don't. Those in the latter camp depend on hype and luck to take them over the goal line. But those who have it figured out do a few things differently — four of them, specifically.

As I have said before, there is no doubt that data and analytics are here to stay. Companies are reinventing themselves around data, creating new products and services based on data, and reengineering their current products based on data. This includes, of course, the large and confusing space of artificial intelligence and machine learning-based solutions.

However, when it comes to machine learning, some companies continue to overemphasize the wrong things at the expense of the bigger whole.

Four critical elements

In my experience, the most critical element is to think holistically and remember why you’re here in the first place: to add business value and solve real business problems. Anything that deviates from this, or that misses any of the critical components, runs the risk of wasting time and money.

As I see it, there are four critical components you need to fully define before implementing successful machine learning-based solutions. Furthermore, what will make a difference is the extent to which the four components are seen as an interconnected system that is in balance.

The Four-Leafed Clover of Success

What is the business problem?

While it's pretty cool to apply machine learning for the fun of it, corporate deployments focused on creating value should have a business problem to tackle. Ask yourself, "Just what business problem do we need to solve? How critical is it? And how is it aligned with the organization's strategic objectives?"

This has two seemingly obvious implications. First, the impact of the solution will be easily measurable since it is directly growing revenue or reducing cost. Next, by being linked to a business problem, it is then possible to identify the business leaders who will be accountable for the solution's actual use. Without this last element, the exercise will merely be a theoretical one.

How will the solution be operationalized?

The next critical component involves clearly articulating how to operationalize the solution. Now I know that's quite the mouthful — what do I mean by operationalization? In a sense, it's integrating the solution into the 'real world' as part of a company's business processes. It's taking responsibility to embed the solution in the day-to-day business — bringing it to where the rubber hits the road.

How will the outcomes of the algorithms be put into practice — and in which business processes? What changes will need to be made to those business processes to incorporate the solutions? To what extent can they be automated? And, most importantly, who are the key business executives accountable for successfully making it happen?

Without this clearly defined, teams leading machine learning-based solutions will produce fancy, involved algorithms that are not linked or embedded in real business processes. This will definitely limit (if not completely hinder) the ability of the solution to deliver tangible value and maintain executive sponsorship.

How much time do we spend on the technical elements?

Next, we have the technical components of the solution. I like to say that this is usually the easy part, although it's usually where the ugly side of data quality and data management come into play. The reality is that data teams still need to spend an enormous amount of time preparing data to then be able to develop models. Seriously — it's way more than people think.

But the key is to always remember that the technical solution is a part in the larger puzzle and, therefore, the amount of time and resources spent on it needs to be balanced accordingly to ensure that progress is being made. In other words, perfect is the enemy of good. A simple, functional solution that is linked to a business problem with a clear path for business operationalization will have a significantly larger impact than a 'perfect' algorithm that doesn't solve a business problem and is not embedded in operational business processes.

How do we improve the system?

Finally comes the last element: a working feedback loop. After all, the process will only get better if there is a clear path to learn and adjust as the solution goes into production. It's important to define how success will be measured, technically and from a business perspective, since without those measurements it will be very hard to tell if things are headed in the right direction.

It’s also important that the solution is nimble and can be incrementally adjusted as feedback comes through, meaning that flexible solutions will be preferred to rigid ones. And don't forget model governance. Artificial intelligence and machine learning will allow models to self-correct and adjust, making parts of the feedback look autonomous. But that doesn't mean these models can just be left alone. Proper monitoring is needed to understand how they're changing and to assess the impact of the adjustments.

It is easier than it seems to have high-impact machine learning solutions in place. Organizations get stuck because they give disproportionate focus to one of the components outlined above at the expense of the whole system. Take a step back, think holistically, and start with something simple that delivers quantifiable value. Then, once the system is working, get on with the next improvement. No luck involved!

The statements and opinions of the writer do not necessarily reflect those of TSYS.

Other Articles by Juan

Juan Gorricho

Juan F. Gorricho is currently group executive, data & analytics, global product innovation at TSYS. In this role, Gorricho leads the development and implementation of TSYS data and analytics strategy aimed at enhancing existing, and developing new data products for TSYS clients.

Prior to this role, he was chief data and analytics officer for Partners Federal Credit Union, where he led the development and execution of Partners' data strategy. Gorricho has more than 15 years of experience in the data and analytics space and frequently speaks at data and analytics-related conferences.

Share this story via email or social networks