Skip to content
Search to learn about InterSystems products and solutions, career opportunities, and more. Results include content from our developer community, product documentation and education websites in addition to InterSystems.com.

Maximize Today’s Downtime to Train ML Models for Tomorrow

computer programmer working with machine learning data on multiple screens

For some, especially first responders, the current crisis continues to look like an overwhelming challenge every day. But, for many businesses the crisis looks like… downtime. The maxim “never let a good crisis go to waste” instantly springs to mind here (Rahm Emanuel said this, although it’s often attributed to Winston Churchill).

And while it’s a natural reaction to hunker down during difficult economic times, businesses shouldn’t be letting this crisis go to waste. Instead, you can use this valuable downtime to better position your company for the future.

As we look ahead to the coming weeks and months, whatever the ‘new normal’ looks like, Machine Learning (ML) and, more broadly, Artificial Intelligence (AI) will be part of it.

Today, ML and AI have a wide variety of use cases from automating administrative tasks, to driving faster and smarter ways to approach new business development and delivering unparalleled data quality for sales and marketing teams.

Continued advancements in ML and AI have huge potential in many domains. The key is to surface low-risk, high reward business solutions to ensure your organization continues to thrive, while also weathering the effects of an economic downturn.

Why is now the right time?

ML and AI continue to be ‘high promise’ technologies that have potential but take time and focus to work through, with Gartner’s CIO survey reporting that only 14% of enterprises have deployed AI, despite 91% having an intent to deploy it. This period of downtime provides that opportunity to focus and test some ML-based predictions – something which tends to give way during busier periods to projects that are urgent and immediate.

As I’m sure you know, there’s a talent shortage around data science and ML, which has slowed many organizations. However, the advent of autoML tools, such as IntegratedML, are enabling data scientists to become much more productive, and approaches that put autoML into the hands of less specialized developers are now available.

Most importantly, business involvement is more possible during a downturn – and this is a critical, often overlooked, factor for AI and ML success.

Training today’s data for tomorrow

As you look to seize this period of downtime, it’s important to recognize that training AI systems is a bit different to training people.

ML generates models based on the data you feed it, and these models can be applied on an ongoing basis, although they often work best if the new data has the same characteristics as the training data. As such, if you train on today’s business data during a period of downtime, tomorrow’s data during a recovery period will likely be different – making the model less accurate. If trained on yesterday’s ‘normal’, tomorrow’s new ‘normal’ may not look the same.

You should train on the data available, explore potential applications and identify a couple of “low-risk, high reward” areas. Don’t just rush these into production blindly, though. Expect to retrain or adjust your models and monitor models in production for the “drift” found when the current conditions deviate from the training conditions.

Make the most of this opportunity

Now is also a great time to encourage experimentation and facilitate organizational learning. ML and AI projects rely deeply on cross-functional collaboration; our new work-at-home operations are actually well suited to this, since they remove the differences in geographical distance that keep many groups in silos. But focus on real value creation, not just learning for its own sake.

Identify the right use cases by involving multidisciplinary teams, then quickly evaluate the potential business value, and create a proof of concept (POC). Next, leverage the POC to identify lessons, requirements, and any potential roadblocks that may arise when scaling up the POC into a fully-fledged ML project.

During a downturn, smart investing is as much about where you spend your time as where you spend your money. Fortunately, ML and AI can be applied with very little capital investment – it’s a matter of exploring the new technology and new potential areas with a positive, experimental mindset, even if your instinct is to hunker down and go dormant.

If you use this time to find high payoff, low-risk opportunities for ML and AI, you’ll come out of it with a competitive advantage. Take advantage of the crisis - train your people, but especially train your machines.

Listen to Jeff Fried's podcast episode, "A Deep Dive on Data Operationalization.

Read more blog posts on Data Excellence.

Other Post You Might Like

As digitalization increases in financial services, firms are dealing with a much higher volume of data in a variety of formats, from structured to unstructured.
Combining both internal and external data improves performance in trading, while gaining greater insights into reducing risk and regulatory compliance.
Silos, complexity, and governance requirements are making it difficult for financial firms to democratize data. Here’s how new data technologies could help.
InterSystems IRIS 2022.1 will allow application developers to leverage data more easily and powerfully than ever, with improvements in several key areas.
Director of Product Management - Data Platforms, System Development
Real-time access to a holistic, integrated view of data is among the top technology challenges facing the financial services industry, according to our report.
Global Head of Financial Services
Analytics anywhere, insights everywhere
Global Head of Product and Industry Marketing
Over the past few years, InterSystems IRIS has become one of the top data management platforms on the market. How will we improve it moving forward?
Director of Product Management - Data Platforms, System Development