Revenue Management Audit
Overview
We were asked to conduct a technical audit of a revenue management system and an organizational and business process assessment of revenue management decision-making for a passenger railroad. We identified deficiencies that were reducing the effectiveness of revenue management decisions and made a number of recommendations to correct the problems. Our findings included:
- Systematic inaccuracies in the demand forecasts resulted in inappropriate feeds to the optimization model, leading to poor inventory control recommendations.
- The firm’s approach to pricing was leading to poor customer service and was not well structured to support revenue management.
- A lack of senior management support for revenue management was leading to departmental conflict, inconsistent implementation of revenue management controls, and lowered the priority of obtaining needed system maintenance and enhancements from the Information Technology Department.
- The Revenue Management Department suffered from poor leadership, often leading to poor revenue management decisions.
- Expanded training and educational efforts for Revenue Management and other staff was needed.
- Current plans for reorganizing the Revenue Management Department were likely to further deteriorate the quality of revenue management decisions.
- A lack of credible performance measures resulted in conflicting signals on the impacts of revenue management decision-making and often led to inappropriate “corrective” action.
For each of these findings, and many others, detailed recommendations for corrective action were made.
Approach
To carry out this assignment, we interviewed staff in many departments and at multiple levels within the organization. In addition, we reviewed system documentation, training materials, performance reports, and internal correspondence and reporting. We also designed a database into which historical data was downloaded and from which we conducted a variety of data analyses. We provided interim updates to the client during the audit, obtaining feedback to determine whether and how we should shift the focus of our investigation.
The following discussion provides additional detail on one of our focus areas, the demand forecasting system.
Reviewing system documentation, in conjunction with information we gathered from our staff interviews, we began to form a hypothesis that the forecasting component of the system was systematically providing estimates of demand that could undermine the validity of the system’s recommendations. As we examined the situation in greater detail, including the technical specifications of the forecasting algorithms, we evaluated how the systematic bias that we suspected to be inherent in the forecast would affect the optimization module and the resulting inventory control recommendations.
The demand forecasting module combined a time-series forecast based on historical demand patterns with a booking profile based forecast to get a final forecast estimate of demand. We found a systematic overestimation bias in the booking profile forecast. The booking profile forecast estimated the demand at departure based on the historical average of the percentage of incremental bookings received between a data reading date and departure. The functional form of the booking profile forecast is a “multiplicative” model. Our experience in several industries over the past 10 years has shown this form to be relatively unstable; for example, it tends to overforecast demand when there is a spike in demand. In addition, this forecasting method also tends to overforecast early in the booking process if only a few more reservations than average are accepted. To counteract this situation, a variety of filters and self-correction mechanisms can be implemented. As we reviewed the forecasting methodology with our client’s staff, we found no evidence that the number and range of self-correcting mechanisms in the forecasting system would be sufficient to result in reliably accurate booking profile forecasts.
Using historical departure and reservations data, we then conducted a series of data analyses to see if our expectations would be borne out. The results were quite conclusive. The analyses demonstrated an extremely strong bias in the forecast, a bias that our client had not previously recognized. Armed with this information, we then demonstrated how this forecasting bias would affect the results of the optimization process. Our expectations were consistent with the client’s experience and led us to recommendations on how the forecasting module could be improved.