the organization, which can make the subsequent rollout easier. In the US utility example used above, schedulers—who were spending four to seven hours a week building and updating the manual schedule—saw that the new technologies could build automated schedules that closely matched their own within minutes. After a successful pilot, it is important to execute a well-thought-out scale-up plan. This plan should take into account factors such as overall deployment speed, deployment across work types (that is, there may be different considerations for electric versus gas jobs or for short-cycle versus long-cycle jobs), the differing challenges of rural and urban service centers, and resourcing the scale-up (for example, potentially hiring change champions or trainers). Tools and processes can be scaled up in an agile fashion because making iterative improvements over time is generally preferable to trying to perfect the algorithms during the pilot period.
A key metric during the pilot period is the frequency of manual schedule overrides by schedulers. These overrides can indicate an issue with the underlying model and should therefore happen as seldom as possible. However, some manual intervention will always be required to address last-minute contextual changes such as sick days or employee holidays. In our experience, it takes at least four to six weeks for smart-scheduling algorithms to reach a 70 to 80 percent match with the final schedules previously created by schedulers, as measured by the percentage of jobs and crew pairings that are the same in each (Exhibit 2). While an optimal schedule is unlikely to exactly match the existing manual schedule, a relatively close match is a good indication that the new algorithm is factoring in the right parameters and will not require frequent manual overrides.
Pilots can also be an important way to build support for the new scheduling methods within
Exhibit 2 Schedule optimizers can improve to build 70 to 80 percent of the final weekly schedules in just four to six weeks.
Typical improvements in scheduling by AI-driven schedule optimizer
% of jobs from final schedule that were also suggested by optimizer
% of crew pairings from final schedule that were also suggested by optimizer
70–80 70–80 1
60
40
25
10
Week 5–6
Week 3–4
Week 1–2
1 Improvement of about 50 percentage points from week 3–4, driven by favored crew pairings.
McKinsey & Company
Accelerating the journey to net zero
129
Made with FlippingBook Online newsletter maker