Asset management can account for a significant percentage of a transmission and distribution (T&D) company’s operating expenses and capital expenditures, with optimized operations and investments key to generating savings. New technologies can enable companies to capture these efficiencies. In fact, a recent McKinsey article explained how T&D utilities can leverage advanced analytics in their asset management strategies to unlock 10 to 20 percent in savings while improving overall reliability and performance of their networks.1
This article builds on that thinking and takes a close look at a North American T&D utility, which we refer to as UtilityCo. In 2021, UtilityCo leveraged advanced analytics in asset management to unlock savings of 20 to 25 percent in operating expenses and 40 to 60 percent in capital expenditures, which could then flow as savings into the profit-and-loss (P&L) statement or be reinvested to deliver significant reliability improvement. These savings and increased investment capacity are particularly relevant given today’s increasing constraints, including pressure from customers on affordability, inflation growth, supply chain bottlenecks, and the growing need for investments in the energy transition, such as renewable-energy solutions, electric-vehicle charging infrastructure, and cybersecurity. Based on the success of the initial model, UtilityCo developed a road map that scales the asset management risk–based approach to two-thirds of the capital portfolio over two years.
The following case study highlights the results of implementing advanced analytics at UtilityCo, including the approach taken, the lessons learned, and the best practices to adopt for others embarking on a similar journey. Although this article is presented as a stand-alone example, our experience shows that the results from applying advanced analytics to asset management are accelerated when deployed as part of a broader organizational transformation.
UtilityCo: An overview
UtilityCo faced a number of key challenges that are common in the industry. For example, the utility didn’t take a risk-based approach when making asset replacement decisions or prioritizing preventive-maintenance activities, and it had decentralized asset management operations, with each operating company taking a distinct approach and methodology. In addition, although UtilityCo was able to collect valuable data, the data were underused and stored in multiple systems. Finally, UtilityCo relied on rules that oversimplified asset management decisions—for example, the “three strikes” rule, which called for replacing cables after they experienced three outages.
The results from advanced analytics
UtilityCo was able to effectively use advanced analytics in asset management in four ways. First, it optimized capital expenditures either by maintaining current risk and spending less—and letting the excess capital expenditures flow into the P&L or be reinvested to deliver more reliability—or by spending the same amount and achieving higher reliability through replacing the riskiest assets. Second, it lowered preventive-maintenance (PM) operating expenses by optimizing PM activities. When successful, this optimization can deliver similar or better reliability at lower cost. Third, it lowered corrective-maintenance (CM) operating expenses by lowering spending on CM after those riskiest assets had been replaced. And fourth, it replaced the riskiest assets to help achieve higher reliability (measured as lower SAIDI and SAIFI2 performance) due to fewer failures.
Regarding capital expenditures for UtilityCo’s transmission transformers, the company underwent a paradigm shift, collecting data about each dollar’s impact on interrupted customer minutes. With this new perspective, UtilityCo determined it could reduce risk approximately two to three times over while spending the same amount, maintaining the same capital expenditures, and reducing customer interruptions. Alternatively, UtilityCo could maintain the same level of risk as determined by the current plan while spending 40 to 60 percent less, thus creating capital headroom for reinvestment, maintaining the same level of customer interruptions but reducing capital expenditures (Exhibit 1). Another option was to select a pathway that both reduced risk and required less spending.
On operating expenses, UtilityCo had the option of spending the same amount on PM—removing 1.5 to 2.0 times the level of risk from the system compared with the current baseline—or maintaining the same level of risk as determined by the current plan while spending 20 to 25 percent less on PM.
For the underground-cables asset class, UtilityCo was able to avoid up to 70 percent more outages as compared with the baseline by replacing its riskiest cables. The optimization model also gave UtilityCo the flexibility to achieve either higher reliability at current spending levels or P&L savings at current reliability.
Finally, UtilityCo developed a visualization platform that displayed the results produced from advanced analytics (see sidebar “Unlocking value as part of a broader transformation”). This dashboard allowed UtilityCo to visualize, prioritize, and implement new maintenance activities (Exhibit 2).
UtilityCo’s advanced analytics–led approach entailed leveraging both internal and external asset data to calculate a health score (the probability of failure) and criticality (the cost of failure) of a given asset. From there, it used the health score and criticality to estimate asset risk and prioritize asset replacement and maintenance activities based on risk (Exhibit 3).
Early on, UtilityCo had a clear plan to scale advanced analytics across all its assets and operating companies. It prioritized assets based on impact, including operating expenses and capital expenditures; time to impact; and feasibility, such as quantity and quality of data and the technical difficulty of building models. In addition, when sequencing distribution assets, UtilityCo followed the concept of “lead” versus “follower” (Exhibit 4). For example, it ensured that poles (lead) were modeled before cross arms (follower).
Health score: Estimating the probability of failure
Depending on the asset, UtilityCo considered more than 100 variables to estimate the probability of failure. A machine-learning model was trained on internal data (such as the age of the asset, work orders, and failure history) as well as on external data (such as weather data, which stretched back a few years). A holdout data set3 was used to test the model performance. For example, when looking at the transmission transformers asset class, the model was able to predict approximately 45 percent of failures in approximately 20 percent of the data (Exhibit 5).
To build the health model, UtilityCo aggregated data from several different systems, such as geographic information and outage-management systems. The utility then cleaned and unified the data in preparation for the machine-learning model and identified prediction targets. In some cases, the process was straightforward, such as labeling a transformer that had suffered an outage. In other cases, it was more difficult, such as when labeling a failed cable that was missing a serial or part number. Next, the data were divided into a training set, which was used to train the machine-learning model, and a test set, which was used to help test the performance of the model after training (for example, testing how often the model correctly predicted asset failure). Because UtilityCo was interested in going beyond a standard machine-learning algorithm, it also incorporated a failure-mode analysis into the transmission-transformer models to support detailed assessments of probability of failure by component and to help with the prioritization of condition-based maintenance.
Finally, UtilityCo combined the outcomes from a previous engineering health model with the machine-learning model to calculate the probability of failure and improve performance.
Criticality: Estimating the cost of failure
UtilityCo estimated the cost of failure across several dimensions, including repair, service, safety, and the environment (Exhibit 6). Repair costs are those related to bringing the asset back online after a failure, service costs are estimated based on lost revenue and other factors related to the “importance” of the customer (for example, a hospital is considered “more important” than a single household), and safety and environmental costs are dependent on location and asset type.
Depending on the asset, utilities will need to group the replacement of assets for operational efficiency. For example, UtilityCo found that grouping the replacement of underground cables and applying operational constraints made sense, but it wasn’t necessary to cluster the replacement for transformers. Typically, simple business rules—such as clustering cables to be repaired based on the protective device they are connected to—can be used to cluster assets.
Building an optimization engine
UtilityCo used an optimization engine to prioritize asset replacement and PM activities based on the risk of the asset. To build the optimization engine, UtilityCo first estimated the risk of each asset by multiplying its health score by its criticality. Asset replacement was then prioritized based on the resulting risk score and the asset’s replacement cost. For example, a high-risk transformer with a lower replacement cost was prioritized over the same type of transformer with a higher cost. To optimize PM, a detailed failure-mode analysis was incorporated into the optimization engine to enable estimates of how much risk was removed by each PM activity. In addition, the optimization engine factored in the cost required to perform each of these activities and prioritized activities that reduced the most risk at given costs.
To optimize PM, a detailed failure-mode analysis was incorporated into the optimization engine to enable estimates of how much risk was removed by each PM activity.
Lessons learned and the recipe for success
Implementing an advanced analytics–led asset management program resulted in several lessons learned, particularly with regard to getting started, data quality, talent and capabilities, change management, and implementation and governance.
First, the team faced internal resistance when getting started, including concerns about not having enough data, or the right data, to address regulatory considerations. The first key step to addressing this resistance was to implement a proof of concept, identifying assets that had good enough data to get started and developing a solution that was better than the current state. Success with the proof of concept gave UtilityCo the confidence to proceed with rolling out the solution across multiple assets and operating companies.
Another lesson involved data that were either siloed, scattered across several different systems, or incomplete or duplicated. For example, data from one asset class were missing installation dates. As a workaround, the manufacturing date was used instead. Developing data architecture and putting processes in place to capture and perform quality control checks on the right types of data were key to addressing this issue going forward.
Although the artificial-intelligence and machine-learning space is still emerging, it is growing quickly. The data scientists and engineers who are key to building solutions are scarce. Thus, UtilityCo built a digital center of excellence and leveraged it to manage the pipeline of talent and to develop processes and trainings to drive consistency across the organization.
During the advanced-analytics implementation, UtilityCo asset managers were asked to make changes to their management processes (see sidebar “Driving additional value from cross-asset optimization”). The key to addressing this issue was to engage the asset managers early on and bring them along as the solution was being developed.
Finally, incorporating advanced analytics into the processes for selecting assets for replacement and updating maintenance processes and policies based on model recommendation was an implementation and governance challenge. Engaging subject matter experts (SMEs) early and running pilots to test new processes gave UtilityCo confidence in its models’ abilities to meet its needs.
Several key ingredients contributed to the success of UtilityCo’s asset management transformation:
- Top-down leadership buy-in and push. Because implementing such a solution usually involves changes in processes across different departments, entailing new technologies, UtilityCo discovered that leadership buy-in and push from the top was critical for advanced-analytics adoption. For example, the vice presidents of T&D and IT joined biweekly sprint reviews to take stock of progress, encourage the team, and communicate the importance of the effort.
- Agile approach to working. By setting one- to two-week sprints and working collaboratively to achieve the goals for each sprint, the team had time to cocreate, bring along team members, and course-correct as needed without falling behind. As part of this, a cross-functional team of business and IT was key to making the implementation successful. For example, in a joint meeting with IT and business, a subject matter expert was able to provide guidance on which corrective work orders should be included in the data used to calculate the probability of failure for a given transmission transformer.
- Not letting perfection be the enemy of getting started. It was more important for UtilityCo to get started with the data that were available and demonstrate some economic impact than to get everything perfect (or build a big data lake) before starting. For example, UtilityCo started with proof of concept for two assets to demonstrate the value before scaling to more assets.
- Change management. Asset managers typically have years of experience managing assets and follow specific procedures and policies. As a result, UtilityCo found it difficult to convince its managers to take recommendations from machine-learning models. For example, when it came to replacing underground cables, UtilityCo asset managers were accustomed to the three-strike rule—in which a cable is replaced after three failures within a 24-month period—and it was a difficult task to convince them to replace a cable that had a high probability of failure but no previous failures. One successful strategy was to involve the team from the beginning and bring them along in the process of building the models.
UtilityCo transformed its asset management strategy from a manual process that used very limited data to make asset replacement decisions and a preventative, one-size-fits-all maintenance approach into a strategy that leverages extensive data and advanced analytics to make replacement and repair decisions based on asset risk. In making this change, UtilityCo unlocked significant savings in capital expenditures and operating expenses while increasing reliability for customers and regulators.