Article

AI and machine learning: It may not be as difficult as you think

Extensive automation solutions are now accessible to the middle market

Sep 07, 2022

Key takeaways

AI and ML solutions are now more available to the middle market as costs have come down and software has advanced

Icon illustration of a cell

These automation solutions can increase process efficiency and business insight

Prescription bottle icon illustration

It’s not about replacing people, but directing resources to more valuable tasks

#
Digital evolution Digital transformation Data analytics

Artificial intelligence and machine learning are not new concepts. In fact, many applications have been around for decades. Unfortunately, many AI and ML solutions have felt out of reach for many middle market organizations. However, those times are rapidly changing as technology matures and presents more realistic opportunities for all companies to take advantage of increased efficiency and insight through automation.

What it is and what it isn’t

Artificial intelligence and machine learning are concepts that automate manual processes within an organization, taking tasks that commonly require human intervention and instead getting support from machines. The two solutions are closely related, with AI replicating human decision-making in a variety of real-world environments and ML leveraging data to identify trends and patterns and make predictions.

Many companies have misconceptions about AI and ML. From an AI perspective, companies utilized decision trees in the past with very specific programmed outcomes—"if this, then that.” This concept is not AI. While people may think the computer is making decisions, it’s actually doing exactly what it was programmed to do. It may look complex, but the computer is just following the rule that the programmer provided for it.

AI is an evolution beyond these decision-support systems that are pre-defined or algorithmic. But we are also not creating a thinking being that will solve problems we have not provided direction for—at least not for middle market business applications. Rather, for both AI and ML, we are trying to solve a specific problem by analyzing historical data and outcomes to predict or suggest a likely action.

With enhanced automation in place, companies can direct human capital to more valuable functions that we can’t train a machine to do and remove the burden of repetitive tasks.
George Casey, RSM Principal

Ultimately, the solutions are not designed to replace humans and disperse farm jobs to machines. Instead, you can identify specific problems and divert their manual efforts to AI and ML. With enhanced automation in place, companies can direct human capital to more valuable functions that we can’t train a machine to do and remove the burden of repetitive tasks.

And another misconception that keeps many companies from capitalizing on the value of automation is that AI and ML are not as expensive, labor-intensive or intrusive as commonly thought.

Why now?

As the technologies have evolved, artificial intelligence and machine learning solutions have become more affordable and accessible in the middle market, making now an opportune time to evaluate how you can optimize key business processes. As mentioned earlier, the technology isn’t new. But the availability of scalable compute, memory and access to massive data sets in the cloud have drastically reduced the cost and effort to experiment and prototype with these technologies.

For example, organizing data was once a major undertaking, but now you have more access to big data solutions. In addition, you have more effective computing resources available with massive machine processing capabilities that can complete complex tasks in minutes.

Software advances represent the most significant evolution in automation. Microsoft Azure AutoML is a prime example of how powerful software solutions have become more attainable for middle market companies. After extensive research and development, AutoML can help you develop and deploy AI solutions with the click of a mouse instead of extensive coding.

These advances have led to the rise of the citizen data scientist. Initially introduced by Gartner, a citizen data scientist is defined by the organization as “a person who creates or generates models that leverage predictive or prescriptive analytics, but whose primary job function is outside of the field of statistics and analytics.” Whether internally or externally, as automation capabilities become more commonplace in the middle market, your organization will need to take advantage of more extensive skills and solutions to guide business decision-making.

Sample use cases

Artificial intelligence and machine learning are valuable applications for a growing list of critical business processes. Many companies associate the technology with manufacturing, supply chain and inventory management. While AI and ML can certainly provide enhanced visibility and more efficiency to those processes, the solutions have now branched out into more key functions, including internal audit, abuse and fraud, and sales and marketing.

For example, some of the areas we have recently helped clients expand the use of AI and ML beyond traditional functions include:

  • Member/customer churn: Based on behaviors and demographics, we can predict the likelihood that a member/customer will renew or defect.  The follow-up exercise includes designing an experiment strategy to match retention techniques to customer/member segments.
  • Fraud detection: Using unsupervised/clustering learning techniques, we can identify outlier transactions in financial ledgers, payments, orders, etc., caused by fraud, abuse or errors.
  • Lead scoring: Based on behaviors and demographics, we can predict the likelihood that a sales opportunity will convert and also identify next best actions given the stage in the buyer journey and customer characteristics.
  • AI demand modeling: We can generate a time series forecast of demand not simply based on past sales but also based on millions of external economic, demographic and industry-relevant data sources.

Getting started

Initiating a journey toward successful AI or ML implementation should always start with defining a specific business problem. Perhaps your operating costs are excessive compared to peers, or your margin is eroding. Maybe your sales conversion rate is not where you want it to be. An effective automation program can help you make headway toward solving these and many other business challenges.

After determining the problem you want to solve, then you rely on your data. Your data has likely been collected for many years and contains truths about your business that can be converted to value. You can leverage technology to ask questions about your data to help identify patterns and relationships. From there, you can formulate a remediation plan.

A host of AI and ML tools and strategies can help unlock that value in many ways. Depending on your specific issues, solutions can fall into several categories, including:

  • Descriptive: What happened?
  • Diagnostic: Why did it happen?
  • Predictive: What will happen?
  • Prescriptive: What is the optimal business decision?

Potential obstacles

Many perceived challenges to adopting an AI or ML program center on data availability and resources. From a data perspective, you may not feel like you have enough clean data to draw effective insights. Chances are, you have enough data to work with—it does not take as much as many organizations assume. However, even if you are a new business or do not have a strong data foundation in place, solutions are available to create a base for enhanced decision-making within your organization.

In many cases, your business can use proxy data sets if necessary to substitute for your data. It may not be perfect, but it is representative of your industry that you can evaluate against performance. You can use estimation techniques until you have more internal data with a monitoring loop to continue making estimates better.

In addition, emerging data privacy standards in several states and multiple industries may dictate how much access AI or ML tools can have to data. These concerns can be alleviated with controls that gather the insights necessary from data but leave no way to map it back to an individual.

Not all middle market companies will have the internal resources necessary to establish an automation program, especially as the demand increases. While you can concentrate on building a team or developing skills internally, a trusted third-party advisor can step in and quickly develop a framework that capitalizes on your data and enables more advanced analysis of your business challenges and opportunities.

Of course, costs will always be a concern for any business. However, the investments necessary for automation have decreased significantly through the years, and the potential benefits—from increasing efficiency to driving more sales—can certainly outweigh the expense. 

Conclusion

Regardless of how extensive your automation platform is, a solution is evolving from something nice to have to a necessary element of middle market business strategy.

In many cases, just getting started is the most difficult part. All models get better over time and get smarter as they continue making predictions. By getting started, you’re creating a program that can yield significant improvements in the short term that become even more substantial over time through reinforcement learning.

RSM contributors

  • George Casey
    Principal

Related insights