Today’s AI/ML technologies are geared toward non-technical business users to make it easier to solve everyday problems. Identifying the problems to tackle first is a critical decision with several key variables at play. You may already have some ideas of what you want to tackle first. If not, there are a few common areas that are “target rich” with opportunities.

• Risk and fraud: Consider recent audits where controls identified potential areas of risk or fraud such as high-volume transactions (payables, receivables, expenses, procurement, etc.). Using pattern recognition and other AI/ML techniques allows for complex pattern recognition and policy violations which would typically not be noticeable by humans. This is true where fraud occurs over multiple related transactions.
• Manual tasks: Robotic process automation (RPA) has been around for a while and is very effective for automating routine, manual tasks. Many of the RPA tools utilize advanced AI/ML to do more than just repeat a single process, such as data entry, and allows them to use reasoning similar to human logic to make value-added decisions.
• Forecasting: AI/ML is great at reading vast amounts of not only historical data but also combing external data to detect correlations that may have impacts on future results. Increasingly, all business leaders are asked to predict the future. This is an area where AI/ML can help.

Highly trained technicians such as statisticians and developers are still needed in some capacity, but many of today’s standard office tools such as Excel and cloud ERP empower traditional functional business operators. Before you embark on your AI/ML journey you need to understand what you have to work with in your department and within the broader enterprise.

• Specialized AI/ML tools and infrastructure: Look to other departments at what they are doing with AI/ML and understand how they are accomplishing it. Meet with your IT group to learn about capabilities such as cloud computing, data management and governance capabilities, and specific AI/ML tools that are available to you.
• Techniques and knowledge: Again, identify those in the organization already doing data science and AI/ML. Chances are they are using best, or at least accepted, practices and standards for modeling and overall delivery. Many IT organizations are employing functionality for ModelOps – which is the controlled way of building, deploying, and managing AI/ML solutions.
• Cloud ERP and Office-native technologies: Excel is the most common, well-known data tool in the world. And now that Microsoft has rolled out Copilot, the common tools you are comfortable with now have highly advanced and powerful capabilities that can be fully used by simply using human-like conversational language. The same goes for modern cloud ERP systems — and it’s one of the key drivers for migrating to the cloud. For example, many standard AP modules come with AI/ML built in to automatically detect fraud — with no custom coding or expensive data scientist required.

The saying “garbage in, garbage out” will be amplified by AI/ML. Fully understanding the input data is critical. AI/ML can also expose your data to security risks. Be sure you work closely with your enterprise security team and understand how sensitive data is being used and what impacts AI/ML may have on its control and dissemination.

• Accuracy and completeness:  Review your input data to ensure it accurately represents a true picture of what is being measured. Data that is missing can sometimes be compensated for using traditional data science techniques.
• Unstructured data: AI/ML is exceptionally good at consuming large amounts of unstructured data such as contracts, articles, and reports. Take full advantage of this to summarize contract language and create user-friendly chatbots for employees to get answers quickly.

It does not matter what great discovery or value AI/ML has made; if not implemented carefully, it will be seen as a failure. For this reason, there are several things to think about.

• Control the chaos: There is a fine balance between allowing team members to explore on their own and controlling the efforts. Users should be encouraged to learn, experiment, and succeed (or fail) without management intervention. However, as these grass-roots efforts start to develop real solutions for use in daily processes there needs to be a control mechanism to manage the deployment and management of these solutions. Again, this is where IT and ModelOps can be a big help. Knowledge needs to be documented and shared.
• Manage expectations: Of everything I’ve written, this is probably the most important. Perception becomes reality all too often and otherwise successful programs are deemed failures due to one undesirable outcome. You will have undesirable outcomes in your AI/ML journey. This is guaranteed. Leadership must understand that, just like science, tools, techniques, and new discoveries happen almost daily. AI/ML is a technology that “learns” over time like a human. The model that doesn’t appear to work today can work better as it gets smarter. Expectations must be managed accordingly.