Human errors cost big money. While banks aren’t required to report how much human error costs them each year, we can guess that human error makes up a significant portion of the total operational losses reported. From 2014-2019, 83 banks reported 778,639 loss events, totalling €482 billion.1 Even if you multiply that figure by 10, you’re still only estimating the losses for 830 financial institutions — out of a universe of tens of thousands.
Human error is such an accepted part of doing business that organizational frameworks, risk metrics and capital calculations have been established to identify, measure and pay for it.
778,639 operational loss events suffered by member banks, 2014-2019
€482 billion estimated value of losses
Source: ORX, 2014-2019
This is the Human Error Tax — the expected cost of human error paid by companies in the form of money and time.
The Human Error Tax can also be levied in indirect ways, such as poor customer experience, reputational damage, lower share price, or money and time spent responding to regulators.
What is Human Error?
We all make errors in our personal lives: We get off at the wrong exit; trip going up the stairs; or input passwords incorrectly. Most of the time we perform these tasks correctly, so what is different about the times we make errors?
Distraction, pace and muscle memory are common causes. Maybe we were distracted by a phone call and passed the exit. Maybe we ran up the stairs instead of walking. Maybe we had just changed a password, but muscle memory caused us to input the old one.
At work, we make human errors when executing manual steps in processes. These often take the form of input errors (sometimes called ‘fat-finger events’), missed execution (forgetting to do something), and mis-communication. Distraction, pace and muscle memory are contributing factors.
You have probably encountered several examples of human error in your business, such as somebody inputting incorrect information or forgetting to confirm information during a transaction. You have probably also seen the consequences of errors, such as needing to dedicate time and money to rectifying the error, delivering of poor client experience; or attracting unwanted attention from regulators.
Reducing the Human Error Tax
We can reduce the Human Error Tax by focusing on the human, the process, or both.
Given the myriad factors influencing human behavior, and the cost and complexity of end-to-end process redesign or automation, it is most practical to focus on both the human and the process. By targeting areas of the business where human error is common, instead of focusing on all areas, we can increase return on investment by reducing the Human Error Tax. This early return can help to self-fund more complex process redesign.
Changing Human Behavior
Let’s focus on the human first.
In order to reduce the Human Error Tax, we need to prevent the human from making the error. We do this by changing human behavior — the behavior of inputting the wrong information (input error), forgetting to do something (missed execution), or mis-communicating.
To do this, we must first predict when an error is likely and then provide a specific prompt to the human to offset the behavior that leads to an error. For example, to prevent input errors, we may prompt the human to ‘slow your input’. The prompt must be given only when there is a high risk of error and with enough time for the human to act.
With machine learning, we can now do exactly that. We can train machine learning models to find the factors that collectively contribute to a higher likelihood of human error. Unlike linear regression, which finds that x is correlated with y, machine learning can determine that when a, b, c, d and e move in a certain manner, then f and g are likely.
For example, think of a 75 year old person who regularly takes the subway to move about their day. On a given day, the likelihood of them slipping may be 5% based on age alone. However, machine learning can analyze large data sets such as health, weather, footwear and subway maintenance. Looking at those collective factors, we may find an 80% likelihood of a slip: a person older than 75, with impaired vision, on a rainy day, traveling through an above ground subway platform that has poor maintenance. A response could be that patients over 75 traveling to an eye-care center from an area with low subway maintenance, on a rainy day could be prompted to walk slowly, wear proper shoes, hold the handrail or take a cab.
We can apply this to a business by training a model to find days when specific lines of business have a higher likelihood of human error, alerting specific employees to change their behavior to avoid an input error, missed execution and mis-communication. As a result we can reduce the Human Error Tax.
Now let’s talk about process. Another way to reduce the Human Error Tax is to make a targeted change to a process, using machine learning to identify when those processes have a higher likelihood of human error. This will give us options:
- Process change: if we know human error increases when the volume of transactions hits a certain threshold, then we can adjust the process to automatically redirect transactions to other employees upon hitting that threshold. We can also identify which employees might have capacity to ensure that the redirected transactions can be absorbed.
- Process automation: if we can identify what parts of, or types of, process are likely to have an error on which days, we can target process automation. For example, some processes require a human to re-input or re-key information, but not all re-keying results in human error. Determining the factors that lead to re-keying errors in one process and not another helps to target investment to areas where the return is highest.
- Process elimination: if we know where human error is likely (and not likely), we may have an opportunity to reduce monitoring and reporting processes. For example, we may use risk assessments, key risk indicators, monitoring and testing to identify and report risk of human error. However, if we know that human error is likely in specific lines of businesses and not in others, then we can eliminate the execution or frequency of the processes where the risk is low. This enables businesses to align risk processes with where the risk is.
So how do we know when human error increases or is likely and identify what processes are more likely to have human error?
We can use machine learning to identify when there is a higher likelihood of human error in specific lines of business, in specific processes, and even in specific activities across processes. We do this by training the model to find the factors that collectively reflect a higher likelihood of human error. In response, we can allocate money and time to process change, automation or elimination that reduces the Human Error Tax.
OLI: Using Machine Learning to Predict and Help Prevent Human Error
OLI is the Operational Loss Intelligence solution that we built at BMO to predict when operational losses, often caused by human error, are likely so that they can be prevented.
OLI combines a company’s internal data with external data and uses machine learning to identify the likelihood of a human error resulting in a loss. This likelihood is based on patterns that the model has identified in the internal and external data. Once the likelihood of an error reaches a threshold, OLI generates an alert, prompting the team to take specific action in order to prevent the loss.
OLI, our Operational Loss Intelligence solution, combines your company’s internal data with external market data, and uses machine learning to identify the likelihood of a human error resulting in a loss.
For example, in a trading business, OLI may predict 75% likelihood of a human error on a specific trading desk on a given day, alerting the trader of the higher risk and prompting them to “slow your input.” Repeated high signal may prompt risk and trading colleagues to review a trading process for adjustments like rerouting trades to another trader when volumes are high to avoid human error caused by high volumes.
We can apply the same approach to other areas that pay the Human Error Tax, such as payments businesses, operations functions, fraud departments, and third-party vendors to which companies have outsourced.
1 ORX, 2014-2019