SolutionsOversight works across all of your systems and functions to identify hidden spend process breakdowns that can cost you millions of dollars. Find out how we can support you across the enterprise.See overviewPartners & Integrations
The world of corporate finance is rife with the potential risk for fraud, erroneous payments, and a whole slew of employee behaviors that add up over time if not corrected.
An artificial intelligence-powered auditing system can help CFOs get to the root cause of these problems and begin putting the issues they find into categories.
To do this, you need to use reason codes.
Usually when we throw out the term “reason codes,” executives get nervous. They think we’re about to hit them with some technical jargon they won’t understand.
The truth is: reason codes are critical to the success of an AI auditing system, but they’re easy to figure out. Furthermore, they save companies money, so while the term may sound complex at first, it’s worth it for CFOs to understand how they work.
In this article, we’ll look at what reason codes are, how to use them to optimize the performance of your AI system, and when it’s time to refine your codes.
What are reason codes?
Reason codes explain why data is being put into certain categories.
For example, employees often say that they weren’t aware that a particular action was a violation of company policy. That’s one common reason code.
Another common occurrence is that a manager approves a policy violation.
If you know why you’re putting things into categories (i.e. the reason codes), it’s easier to get everyone engaged with the automated auditing system to produce the best results.
To develop your reason codes, start by defining the metrics that you want measured once the system is deployed. If you know what data is essential and what you plan to do with the data, it becomes easier to tag things with the correct reason codes.
How can we effectively use reason codes?
Reasons code allow the AI system to aggregate the tags to discover root causes.
For example, if you see a plethora of “Employee did not understand travel policy” reason codes, new training is needed for the employees about the travel policy.
Reason codes let you see how many problems occur, so you can direct your corrective actions to the highest-impact areas and improve those processes.
Responses and subsequent behavior are also tracked, so you can see what corrective actions work and what interactions have no impact. You may need to change how you interact. Tracking responses also allows you to quantify the benefits.
For example, if you have problems with duplicate payments to a vendor who sends two copies of their invoices, you may ask them to only send one copy, and the system will track whether that conversation corrected the issue.
It’s important to wait for the right time to decide which reason code to apply. You don’t want to assign a reason code to everything the instant it is discovered.
Usually, you need to wait till you’ve interacted with the employees, managers, vendors, or customers and heard their replies. A sophisticated automated auditing system understands that and will ask you to assign a reason code at the appropriate time.
When should we refine our reason codes?
Once your categories begin to fill up with issues, what you may find is that one category is overflowing with issues. In this case, it may be smart to break that category down into smaller, more discrete categories of problems with more nuanced reason codes.
For example, your company policy may be that employees can’t stay in hotels that cost more than $200 a night. That works for small towns, but it doesn’t for New York City, and if all those New York trips go into a big, general “out-of-policy” bucket, you may never get a clear idea of why those out-of-policy expenses occur.
But if those trips go into the more discrete category “room rate not available at destination,” you can see that many of your employees violate the room-rate policy because it’s impossible to get a hotel in New York City for less than $200 a night.
By refining your reason codes in this way, you can more effectively diagnose the problem and track the resulting improvements.
One piece of advice here is don’t let the perfect be the enemy of the good.
Start with straightforward, simple reason codes and refine how you tag the data as you become more skilled with working with the AI system.
Just as you drive continuous improvement in your company processes, you continually improve how you use the automated auditing system. If you bought your AI system from a vendor, you should expect them to continuously improve the system.
Creating several different types of reason codes is useful.
Instead of only having a code for manager-approved policy violations, add codes for client entertainment or unavailability of corporate hotel rate. If you discover that a lot of times the lower rate wasn’t available, you need to talk to your hotel suppliers.
An AI system gives you nuanced judgment at scale. It learns that some of your valued customers won’t pay in 45 days. It learns that employees going to Manhattan will pay $400 a night for a hotel during certain times of the year.
At the same time, it tells you about that one employee, who continues to use the company credit card for groceries, might need help. This nuanced judgment is dependable, but only if effective reason codes are used.
This is part of an ongoing blog series based on the recently released book, Robo Auditing: Using Artificial Intelligence to Optimize Corporate Finance Processes by Patrick J.D. Taylor, Manish Singh and Nathanael L’Heureux.