How should one call centers use big data, artificial intelligence (AI), and what they know about risk to create more accurate, less risky tickets?
Answering this question is the next challenge for one call centers and aligns with how transformative geographic information system (GIS) mapping was to the industry 15-20 years ago.
At that time, the industry was trying to figure out how to build a system to send out a ticket accurately. Fax machines and, to a lesser extent, emails were the “systems” of the day. While GIS was in its infancy, early adopters experienced a dramatic drop in outbound tickets because companies could accurately see where their infrastructure lay. Yet for years, when you created a web-based ticket, it was a manual process that started by filling out a form and sending it to an operator who would key in the information to create the ticket.
Today, we know AI is real, and big data is important. So, how do we embrace that to create accurate 811 tickets?
Understanding Risk Begins with Data
We receive about 120 million tickets into our platform each year – the most in the industry. In comparison, contract locator USIC does about 100 million transactions. Why is that critical? The amount of data Irth has access to gives us the ability to see trends. This insight allows us to put proactive measures in place to create better tickets.
Risk Algorithms: Learning Model and Artificial Intelligence
For effective risk management, you must take a two-pronged approach when analyzing risk factors. Our risk algorithms include:
This model is built from the experience and knowledge of people who know the industry well and is the model most people are familiar with when talking about risk. This includes what excavators typically cause the most damage, accounts for locators failing audits, and what digs are more challenging due to weather conditions or type of work. Ticket management systems are then populated with these details so field users make better decisions.
- Artificial intelligence model
Technology is a partner in damage prevention with an artificial intelligence model to assess risk. Data is key in this model. We use 811 ticket data, damage data, and external data sources from IBM Research and Boston Geospatial to give 360-degree situational awareness. When an incident or damage occurs, AI will look at the data to find trends. Then, that information is used to make better decisions when creating tickets.
AI’s performance is impressive. Using only data from the one call center, AI can predict more than 50% of incidents in the top 10% of tickets you receive (highest-risk locate requests). When you leverage additional data sources such as GIS data, weather patterns, etc., we can increase predictive performance even higher.
We help you understand where risk is greater and what actions you can take to minimize or at least lower that risk, such as sending out notifications to the field for tickets assessed as higher risk.
Action and Reaction: Great Formula for Understanding How Risk Works
In the world of damage prevention, creating a ticket is an action. The reaction we don’t want to experience in our industry is an incident or damage. However, when that does happen, we can look at the original data and find any trends that align. Then use that information to make better decisions when we’re creating tickets.
How to Leverage Data to Make a More Accurate 811 Ticket
The best thing you can do to create a more accurate 811 ticket is to blend AI and your learning model. This allows you to take the AI, which makes decisions based on a very simple action-reaction equation, and combine it with the learning model of gut feelings, what an individual knows, etc., for max accuracy when making a ticket.
With over 30 years of data and hundreds of millions of 811 tickets, our data is unparalleled.
We can create efficiency and accuracy by having the end-user create a ticket in real-time to send it out and eliminate the need to have an operator input the ticket into the system.
Studies show tickets aren’t as accurate when the operator translates what the person from the field says into an actual ticket. Issues can be introduced due to:
- Lack of communication
- An inaccurate description
- The operator not hearing the details correctly
One of the reasons tickets created in the field are more accurate is GIS is more accurate and accessible today.
Without needing to key in tickets, operators have bandwidth to do much more detailed work. Some of our centers automatically create 90% of their tickets by users in the field. This puts the onus on the system to deliver a high-quality ticket that doesn’t introduce any additional risk.
While we’re comfortable now with GIS and its advantages to the market, the next thing we need to build is comfort around the influence data has on risk management.
Examples of Risk Variables When Creating a Ticket to Identify Areas that Introduce Risk to a Project
- What excavators have caused the most damage?
- What work type has caused the most damage?
- Has boring caused a lot of damage?
- Have explosives caused a lot of damage or does it increase the risk?
- What facilities have the most damage?
- What locators have been the most at fault for damages?
- What is the consequence impact (i.e., asset footprint, infrastructure type, customer footprint such as urban vs. rural, schools, 911 centers, etc.)
These risk factors vary based on geographic regions.
Did you know directional drilling is the work type that causes the most damage? No call locates is the leading consequence for actual damage. When AI uncovers trends like these, it escalates these risk factors. When analyzing a ticket with directional drilling, the AI algorithm would advise being more precise in the information gathered since it associates this with higher risk. It will prompt more questions to make the ticket as accurate as possible.
Contrary to what you might expect, the data show damages are extremely low when explosives are used by utilities. While using explosives is risky, we see less damage because people put more effort into doing it correctly. When people are aware, they perform and behave better.
Once our system identifies trends where risk occurs using a learning model and AI model, we can educate our customers so the action after the reaction equates to success. For example, a truck was rolled for a high-risk ticket and as a result no damages occurred.
AI is dynamic and learns based on the adjustments you make and the results you experience to always surface what is currently the highest risk activity and, therefore, the riskiest tickets.
Challenges of Using Risk When Creating 811 Notifications
There are some considerations before we can automate the process of using risk in our systems. For example, when we automatically classify some tickets as “risky,” we don’t want to imply that others are less important. Additionally, risk factors will vary based on one call centers, or different utility owners may have different levels of acceptable risk.
With AI models, there is also “drift” we need to account for — the long view of our AI model action and reaction. When the action starts to drift, we must update our learning model to consider the results of the previous models.
At Irth, our North American AI model is based on 120 million tickets over 10 years to give us the baseline. We apply that model equally to everyone. The baseline works on a very generic level, but how do regional differences affect the model? As we add new factors, we see different actions and reactions, and what was once risky is now less risky; new activities appear at the top of the risk scale. That’s when we retrain our model.
Ultimately, the goal is to discern the level of comfort one call centers and the industry has with automating the process of using risk. One call centers would like to use the knowledge that they understand where risk is most valuable to guide them in creating a more accurate, less risky ticket. A couple of those factors include:
- Rechecking the validation of the GIS map data.
- Looking at the registration data of the utilities being notified on that ticket, the more accurate the registration data, the fewer damages we see.
So, when you create a ticket, the system can also create a follow-up notification that goes out to the end-user saying their maps haven’t been updated in two years prompting them to update their registration information. The objective is to build assurances that we’re building a safe continuation of the process.
What are Leading Companies Doing to Better Mitigate Risk?
These companies use high-quality data and then build and send notifications based on what the data tells them. As a result, they experience these benefits:
- Proactively identifying risk with predictive models
- Immediate visibility into high-consequence/profile facilities
- Focus damage prevention resources using real data
- Automate steps to notify and assign the correct individuals of risk
