Most businesses think they know their capacity: Restaurants count seats in the dining room, hotels count rooms in the building, and airlines count seats on the aircraft and the numbers appear straightforward, but they rarely reflect the system’s real constraint.

In practice, the binding constraint often lies somewhere else entirely. In restaurants it may be the kitchen’s ability to produce meals during peak periods. In hotels it may be housekeeping availability.

In delivery systems it may be driver supply. In large venues it may be security throughput or entry bottlenecks. What appears in the system as “capacity” is often only a rough approximation of what the operation can actually deliver.

Revenue management has long recognized this distinction. Yet most systems still treat capacity as fixed. The relevant constraint is not theoretical capacity, but effective capacity: the amount of product the system can realistically deliver under actual operating conditions.

For decades, however, these operational constraints have been difficult to measure directly. Managers could observe reservations, sales, table counts, room inventory, and booking patterns, but the underlying drivers of effective capacity remained largely hidden.

During the Covid pandemic, this became particularly visible. Restaurants technically had the same number of tables and hotels the same number of rooms, but government policies, staffing shortages, and operational disruptions dramatically altered how much capacity could actually be used. The theoretical capacity of the system remained unchanged, but its effective capacity fluctuated constantly. Managers were forced to make decisions with only partial visibility into the constraints shaping the system.

What is beginning to change is our ability to observe these operational limits more directly. Advances in data systems and AI are making it possible to measure aspects of capacity that were previously invisible.

Reservation and POS systems contain signals of unconstrained demand, but those signals are indirect and easy to miss. They appear in patterns such as customers who search for a table but do not book, repeated requests for unavailable time slots, walk-in guests who leave after hearing long waits, or orders that are delayed or abandoned. The data exist, but they are fragmented across systems and rarely interpreted in a way that reveals the underlying constraint.

What AI makes possible is not simply collecting more data, but connecting and interpreting these signals at scale. By linking search behavior, booking patterns, table utilization, and service times, systems can begin to estimate how many customers would have been served if capacity had been available and where in the operation that constraint actually lies. In some cases, platforms are already using these signals to adjust availability, pacing, and staffing decisions in real time.

Computer vision tools can track guest flows, walkaways, and table utilization. Delivery platforms can observe driver availability and kitchen throughput across thousands of restaurants in real time. In effect, these technologies are beginning to illuminate the operational bottlenecks that determine how much capacity a system can actually deliver.

In some sectors, particularly large digital platforms, the system can already observe these constraints at scale. But in many service businesses, managers are only beginning to see how much operational information has historically remained hidden. In many cases, the data exist, but have never been structured or connected in a way that makes the constraint visible.

Many years ago, I had a research assistant sit in the lobby of a restaurant and simply count how many guests were turned away during dinner service. Each hour he would ask the host how many people had been turned away. The host would usually respond that very few guests had been refused.

My assistant would then tell them how many he had counted and point to the empty tables in the dining room.

That assistant, WillGuidara, went on to build one of the most thoughtful approaches to hospitality in the industry. Even at nineteen, he saw something many systems still miss: the operation was not actually full.

The exercise illustrated a simple but important point. Even in a relatively small operation, managers often lack clear visibility into the operational factors that determine effective capacity. Some guests were turned away because the host believed the restaurant was full. Others left after hearing an estimated wait time that turned out to be longer than necessary. Meanwhile, tables occasionally sat empty while the system adjusted.

What looked like a fixed capacity constraint was actually a moving boundary. And in a hospitality business, every guest who quietly walks away is not just lost revenue, but a missed opportunity to create an experience.

Anyone who has spent time in restaurant operations has seen some version of this phenomenon. Similar patterns appear in hotels when rooms remain unoccupied while housekeeping catches up, in delivery systems when orders wait for drivers, and in event venues when entry bottlenecks slow throughput. The dining room appears full, yet tables sit empty. Guests are told the wait is thirty minutes, yet a table opens ten minutes later.

Similar dynamics appear in professional services, where billable capacity is constrained not by hours available, but by how much time is absorbed by coordination, rework, and non-billable tasks. The question is not whether these situations occur. The question is how often they occur and how much revenue quietly disappears as a result.

This is precisely the type of operational friction that has historically been difficult to observe, and therefore difficult to manage. Today, however, AI-driven tools are beginning to make these dynamics visible. Systems can track walkaways, monitor table utilization patterns, and detect operational bottlenecks in real time. In doing so, they begin to reveal the true constraint governing the system.

For many years, revenue management systems have treated capacity as a fixed number in the model. The optimization problem assumes that this constraint is known and stable. Effective capacity often fluctuates throughout the day as operational conditions change.

If new data and AI tools begin to make these dynamics visible, revenue management may gradually evolve from managing demand against a fixed constraint to understanding, and perhaps even managing, the constraint itself.

Many service businesses believe they know their capacity. The systems they use reinforce that belief.

Often, the most important constraint is the one hidden in the operation itself.

Sheryl E. Kimes, Ph.D., is professor emerita of operations management at the Cornell University School of Hotel Administration, Cornell SC Johnson College of Business, and a visiting professor of analytics and operations at the Business School at the National University of Singapore (sk@sherrikimes.com; sek6@cornell.edu).

Her area of specialization is revenue management. She has been teaching, conducting research, and providing consulting services in this area for over 25 years. She has published over 100 articles and book chapters and has received multiple awards for her research including the Lifetime Achievement Award by the College of Service Operations of the Production and Operations Management Society and the Industry Relevance Award by the Cornell University Center for Hospitality Research in 2010, 2012, and 2014. In 2017, she was given the Hotel Sales and Marketing International Association Vanguard Award for Lifetime Achievement in Revenue Management. She was the third recipient of this award.

www.sherrikimes.com