Smart scheduling: How to solve workforce-planning challenges with AI

AI-driven schedule optimizers can alleviate age-old scheduling headaches—reducing employee downtime, improving productivity, and minimizing schedule-related service disruptions.

Today, workforce planning has reached a turning point. Traditional workforce management processes, which rely heavily on time-consuming and inconsistent manual steps, can no longer provide the dynamic workforce scheduling needed in the face of ongoing labor market disruptions. The past two years have brought the inefficiencies of traditional processes to the surface more keenly than ever before as the COVID-19 pandemic placed an unforeseen strain on day-to-day operations across many sectors. The challenges of a constrained labor supply and higher wages remain at the fore.

In recent years, advanced data applications and AI have optimized many business processes. Yet until now, workforce planning hasn’t enjoyed the same level of digital transformation. Recent advances in technology and declining costs have made end-to-end, AI-driven schedule optimization a real possibility—and an opportunity. This article explores how it can drive a long-needed transformation by bringing greater speed, flexibility, and intelligence to bear on the problem of optimizing schedules, so companies can deploy the people they need when they need them and unlock new levels of efficiency.

The unsolved optimization challenge

Optimizing schedules is one of the most challenging of all optimization problems. Extreme variability—in workforce types and operations, as well as across sectors and businesses—makes these solutions hard to standardize.

Even within individual businesses, the complexity of workforce planning and the demand for dynamic action make agile decision making difficult. To operate with the greatest efficiency, businesses must deploy the right number of workers to meet demand and minimize employee downtime on any given day. The constantly changing picture and high number of decision variables generate complex traditional computer models that often take a long time to run. Factor in unforeseen changes—such as employees not showing up for work at a moment’s notice or spikes in demand—and the pressures on optimization models become even greater. New schedules must be calculated using fresh inputs very quickly, yet most all-in-one optimization models take hours to deliver updated schedules.

What’s more, existing scheduling tools are not always user friendly and may require a team of data scientists to maintain and update. And to be truly valuable, scheduling models must be integrated with other models, such as demand forecasting. As a result of these challenges, businesses can lose the opportunity to streamline their offerings and provide better service to customers—and thus lose income too.

Optimizing workforce management matters now more than ever. Three recent factors have forced it up the strategic corporate agenda. First (and least expected) was the impact of the COVID-19 pandemic on operations globally as abrupt swings in demand stretched spreadsheet-based workforce-scheduling models past their limits. A North American telco illustrates the challenge: amid skyrocketing demand for internet capacity, the organization struggled to reassign its technicians (who were long used to providing on-site installation and repair) to resolve problems remotely. Hampered by old technology, the business couldn’t overcome its problems in workforce management and personnel scheduling. Satisfaction fell among customers and field workers alike, while both customer churn and employee attrition increased.

Second, though COVID-19 may prove to be a one-time event in our lifetimes, more changes in the workforce landscape are expected because of, for example, high inflation, demographic turnover (as large numbers of highly experienced workers retire), and potential policy changes affecting labor terms and conditions. The resulting uncertainty could persistently complicate labor planning.

Proposed regulations may force organizations to change their operational and labor supply strategies. Given the manual nature of current workforce management systems, optimizing and changing day-to-day operations require a lot of time, as well as large teams of planners and capacity managers. Consequently, current scheduling processes are often inconsistent and heavily influenced by human bias, and that raises the potential for error, inefficiency, and regulatory risk. All these factors probably increase in tandem with the complexity of the labor force.

Third, and most encouraging, the advent of new technologies for AI and cloud-based computing has reduced the cost of deploying end-to-end, AI-driven solutions for optimizing schedules. During the past ten years, the organizational appetite for adopting digital solutions for workforce management has consistently increased.

The advent of new technologies for AI and cloud-based computing has reduced the cost of deploying end-to-end, AI-driven solutions for optimizing schedules.

Leveraging AI to manage and schedule the workforce

The current market context makes it more important than ever to optimize schedules. AI-driven tools offer optimal solutions for the range of interdependent constraints and changing demand. These solutions generate schedules that are as efficient as possible, so the right resources reach the right places at the right times.

AI-driven solutions take significantly less time to schedule the workforce than current spreadsheet-based models do and can capture unexpected changes in operations more efficiently. The technology allows for a consistent and systematic approach, eliminating human bias and error, creating fairer planning schedules, and reducing the managerial bandwidth required to oversee the scheduling process.

Exhibit 1, for instance, depicts how smart scheduling could optimize the daily schedules of crew members at a utility service center by streamlining daily activities, reducing travel time, and increasing overall productivity and efficiency in the field. The left-hand side shows how much time crew members have traditionally spent on jobs, travel, and unassigned or nonjob work, such as training sessions. The right-hand side shows an optimized schedule—how smart scheduling could have allocated these employees’ time. Job time increases significantly thanks to geographic optimization and the use of better estimates for job durations. Unassigned times fall.

Smart scheduling can transform the daily experience for the front line and reduce unassigned time.
We strive to provide individuals with disabilities equal access to our website. If you would like information about this content we will be happy to work with you. Please email us at: McKinsey_Website_Accessibility@mckinsey.com

Turning this promise into reality—especially across different sectors—requires a new way to approach the workforce and demand. Three possible approaches could help smart scheduling succeed through optimization: generalizing schedules across operation types, developing a modular approach, and integrating user-friendly and end-to-end digital solutions.

Generalizing schedules across operation types

Across sectors, generalizing optimization to all types of operations is a major challenge in scaling up solutions to optimize schedules. Some operations, for example, require workers to travel between different locations to finish jobs; others don’t. These two possibilities require significantly different modeling. Part of the solution involves identifying different types of operations and creating an inclusive categorization system. The majority of operations across different sectors can be grouped in five categories:

  • Job stages. A single job (in construction projects, for example) might comprise multiple stages that cannot start until the earlier ones have finished. These stages can stretch out over multiple days. Other jobs, such as calls answered by call center agents, involve only one stage: the job is done when the call ends.
  • Crew allocation. A job might require more than one worker and skill type, so workforce managers must ensure that a crew with the right skills is allocated. Decisions about which crew is required for which job depend on skills, availability, and distance to the job’s location. Of course, that becomes more complicated the more crew members are geographically dispersed.
  • Demand type. In some cases—such as fast-food restaurants or call centers—demand types fluctuate, and the volume of work is not known ahead of time. In other cases, such as mining projects, the amount of work is known in advance, and the scheduling system must address a backlog of work in an optimal way.
  • Shift type. In some instances, shifts can change from week to week. Managers in a call center, for example, could decide to have a different number of agents on the load in different weeks. In other instances, shifts are fixed.
  • Mobility. In areas such as field-force operations, workers must travel from one location to another. This adds a level of complexity, since driving times must be factored in.

A single operation can fall into multiple categories. Consider a fast-food-delivery operation. It could have fluctuating demand, be mobility centered, and employ both full- and part-time workers. Depending on the product, the jobs could involve one stage or several stages. Optimization systems must be flexible enough to handle all the different job types relevant to an operation’s needs.

Developing a modular approach

Solving scheduling problems for all operation types requires a set of predesigned modules that can be assembled to address specific scheduling problems. A modular approach helps with run times and computational aspects because it breaks down the optimization into multiple smaller steps.

Four modules can handle a majority of schedule optimization problems.

  • Demand and supply balancing. This is a core module in most operations. The optimization model is an integer programming problem 1 in which the input is sub-daily demand and the output is required shifts. The module decides on the number of shifts needed and therefore addresses all or a portion of the demand (depending on the user’s setting) while minimizing total costs.
  • Job-to-work-center allocation. In some cases, the allocation of jobs to work centers must be decided before shifts can be optimized—for example, call centers where calls must be routed to different centers or field-force operations that must distribute jobs in different locations among a number of technician centers.
  • Heuristic dispatching. Particularly if assigning jobs is complex, a heuristic 2 approach can be successfully applied to dispatching problems. Some jobs might need to be prioritized, for example, or workers may have different competency levels. In these cases, heuristic optimization is the most powerful approach because it can apply all custom rules in a significantly flexible way. As a result of this approach’s iterative nature, the user controls how optimal the response should be. That makes run times and required computational resources more flexible.
  • The traveling-salesman problem. Mobile workforces such as field service operations can benefit from the module for traveling-salesman problems. 3 Once jobs are assigned to workers based on priority and skill type, this module can work out the right sequence of stops to minimize overall travel time.

When the four optimization modules are combined, the five optimization categories can address a majority of operation types. A modular approach not only adds flexibility but also reduces the required computation time (Exhibit 2).

Almost all schedule optimization problems can be resolved by combining a set of four optimization modules.
We strive to provide individuals with disabilities equal access to our website. If you would like information about this content we will be happy to work with you. Please email us at: McKinsey_Website_Accessibility@mckinsey.com

Integrating user-friendly, end-to-end digital solutions

Optimization must be integrated and provide a user-friendly, end-to-end digital solution through constant updating, accurate forecasting, and a helpful interactive front-end interface.

Constant updating is critical to ensure that incoming data are fresh and relevant for scheduling decisions. Certain data sets must be refreshed and input into the scheduling engine on a daily or weekly basis—for instance, incoming jobs, work booked for the coming two to four weeks, incomplete or carryover work, and upcoming crew availability. This allows the scheduling engine to output relevant schedules and planning calls between the scheduling and field execution teams.

In this way, scheduling teams can procure the right resources with sufficient lead times to prevent unplanned overtime, backlog, and incomplete orders. Short-range forecasts, for example, provide a view of expected work volumes and available crews. Any mismatch between supply and demand can therefore be adjusted ahead of time, preventing last-minute scrambles. Thanks to this approach, a North American telecom provider reached 80 to 85 percent accuracy levels by developing forecasts with daily granularity.

Finally, an effective, interactive front-end interface allows a new scheduling tool to be adopted more quickly and sustainably. Features such as drag-and-drop daily or weekly schedules, preloaded AI-optimized schedules, and metrics dashboards are superior to (and easier to use than) the current spreadsheet-based schedules.

Significant results in the electric and gas sector

The electric and gas utilities sector, by nature, presents a scheduling challenge given the variety of different work types and varying schedule dynamics. Schedules must take into account short-term, long-term, and unplanned emergency jobs, and demand must be matched with resources and supplies such as crews, materials, and equipment.

Smart scheduling has been shown to work to great effect within this sector. A US electric and gas utility, for example, deployed a smart-scheduling solution for six weeks at one of its service centers. It improved the productivity and user experience of schedulers and field crews alike.

To deal with the challenges of job prioritization, schedule preparation, and execution, the utility used a machine learning–based schedule optimizer that automated and optimized the creation of schedules, thus improving productivity in the field and reducing rework among schedulers. The service center made significant gains in productivity—break-ins (emergency jobs that disrupt schedules and demand real-time reworking) fell by 75 percent and job delays by 67 percent.

Smart scheduling also identified the optimal crews for emergency jobs, basing its choices on geographic proximity and the importance of the crew members’ current jobs. False truck rolls—when jobs can’t be started or completed on time because crews, equipment, or materials are not available—fell by 80 percent (Exhibit 3). This in turn made employees more available for jobs, and fewer break-ins meant that more work was completed—total on-job time increased by around 29 percent, and total jobs worked on by 6 percent (Exhibit 4). Smart scheduling also ensured that crews, equipment, and materials were available to optimize each job.

Smart scheduling solved several issues, including job delays and false starts.
We strive to provide individuals with disabilities equal access to our website. If you would like information about this content we will be happy to work with you. Please email us at: McKinsey_Website_Accessibility@mckinsey.com
Smart scheduling resulted in more job hours being completed, and more jobs worked on overall.
We strive to provide individuals with disabilities equal access to our website. If you would like information about this content we will be happy to work with you. Please email us at: McKinsey_Website_Accessibility@mckinsey.com

Overall, after accounting for seasonality and other confounding variables, the service center, over a period of six weeks, boosted the productivity of field workers by 20 to 30 percent and the productivity of schedulers by 10 to 20 percent. This equaled one to two hours a day—results confirming that smart scheduling is a way for businesses to get ahead.


Workforce optimization has long been among the most challenging problems for businesses. It is even more so now, given ongoing labor disruptions and higher wages. AI-driven schedule optimizers offer solutions that smooth out and speed up workforce management processes. By adopting customized AI-driven schedulers, businesses can optimize across all their spheres of operation, save time and money, and, ultimately, boost their productivity.

Explore a career with us

Related Articles