How to match parking dynamic pricing solutions with a worldwide (Covid-19) crisis

The activity of the airport parking sector has considerably dropped since the Covid-19 related lockdown decisions taken by the various governments around the world. This raises a number of questions for those Yield managers in charge of parking revenue optimization. Such atypical situations indeed impose to think outside of the box and adjust the classical Yield actions. 

The major questions parking Yield managers have to answer

As in other sectors, dynamic pricing solutions (or “yield” solutions) in the parking sector are primarily based on demand forecasts. Those forecasts are used by optimization engines that will ultimately produce dynamic and optimal prices so as to optimize the revenues of a set of parking places.

There are many aspects by which parking revenue optimization engines strongly differ from that of  other sectors (the major one being the timeline aspect that drives high complexity in the algorithms, somewhat similar to the O&D problem in the airline industry). However, as far as the forecasts are concerned, atypical situations like the Covid-19 crisis, raise the following major questions, whatever the sector:

• Will the yield forecasting and optimization models learn by themselves and simply adjust their price recommendations?

• What should the manual actions be to ensure a proper control of the dynamic prices after the crisis?

The hypothesis that one can derive enough information from past observations to forecast  the future demand (usually per customer segment or “yield class”) is likely to be wrong for a while unless yield expert actions are taken. This is where the role of the yield manager is crucial, together with his partner, the dynamic pricing solution provider. 

The various situations Yield managers have to face during the crisis

The four main situations the yield managers will most probably have to face during this Covid-19 crisis are:

• The drop down phase where decline is steep and demand vanishes

• The lockdown phase where almost no demand is observed

• The recovery phase where demand is progressively back but still very low

• The cruise phase where demand and customer behaviors are back to normal (from a parking buying perspective)

During each of these phases, the behavior and capacity for action of the yield manager are radically different.

During the drop-down and lockdown phases, it is important to properly adjust the offerings: it can be simple car park closing or booking parameter settings to properly ensure demand transfer from one car park to another, etc. (case of the airports, for example). When capacity is still open for parking, make sure adapted pricing policies are put in place. From a yield perspective, make sure all actions taken on the basic offerings (capacities, prices) are dated and recorded.

During this phase, at stake is customer services: capacity to ensure a high level of service, including parking, in spite of abnormal travel conditions.

During the recovery and the cruise phases, yield managers have to be back on the deck, making sure the pricing policy they propose is consistent with the evolving business situation, the idea being to move back from a temporary manual pricing (right after the lockdown phase, yield foundations based on forecasts are not valid anymore) to the dynamic pricing solution that was used prior the crisis.

To do so, a deep understanding of their market, combined with a high expertise in the dynamic pricing solution they are using is strongly required. In other words, one needs to not only tightly follow-up the customer successive behaviors (during those phases) but also be able to take a deep dive into the dynamic pricing module solutions themselves: forecasting models, optimization algorithms, etc., in order to switch back to the automatic pricing mode. 

Guidelines for the Yield manager during such atypical situations

Drop-down phase

During the drop-down phase, it is recommended to:Either switch to a manual mode with prices that are adjusted to the vanishing local demand: “adjusted” does not mean lowered (as one would lower prices when willing to stimulate a demand that still exists but has gone to competition, for example) but simply set to a standard level;Or keep the automated mode but forcing it to pre-identified “yield situations” like a low-demand mode (where usually, prices are by definition reasonably low).In all cases, during this very short phase, whatever the method, the prices should be manually or semi-manually generated and attention should be primarily paid to customer service.

Lockdown phase

During the lockdown phase, there is not much to do, even in the backstage because all the yield expert-based actions are conditioned by the timeline, in particular by the end of this exceptional phase.

Recovery phase

During the recovery phase, there is an intensive work to be done “behind the scenes” by:

• A yield expert very knowledgeable about the yield solution that is used: usually the product manager of the software provider,

• A yield expert very knowledgeable about the business challenges at stake: usually the yield manager.

It is the combination of those two profiles that best put together their forces to ideally accompany this recovery.

The general idea during this phase is to keep on controlling the situation on a manual mode (for example on-line prices remain semi-manually set, usually via the rule-based pricing module of the Yield solution) but in parallel, start to take all necessary data correction actions so as to prepare the next phase where the crisis is overcome and where the dynamic pricing solution will be on again.

Cruise phase

During this phase, the situation is supposed to be back to quasi normal from a business view point. For a crisis like that of Covid-19, we all know it will take time before, in the airport segment for example, travel behaviors and transport demand reach their original level.Still, during this phase, a parking dynamic pricing solution can be switched to an automated mode if, and only if, the previous phase has been properly conducted (data and yield model adjustments are done) and of course, if the Yield manager is given the proper tools to pilot the progressive return to normal. 

How to correct the lockdown related data and adjust the dynamic pricing models for the back-to-normal situation?

Yield management solutions are often based on several forecasting models usually imbricated with each other, some of them capturing the long-term trend (or seasonal trend), others more focusing on the latest demand observations, others specialized in the deep-learning parameters of the previous models, others again catching exogeneous data (competition or external demand driver data, etc.).

In all cases, the past observations combined with the actuals as well as complementary information like competitor pricing moves, airline passenger forecasts, etc. are digested by a sound dynamic pricing solution to produce the ad hoc dynamic prices.

Data corrections

In a crisis situation like that of Covid-19, the past observations have to be taken care of with specific attention:

• The lockdown period of time has to be excluded from the forecasting models: this can be done via a user interface if this yield management exclusion feature exists or in a degraded way, via a direct manual action on the database ;

• In case some forced demand transfers have been observed during the recovery phase, some manual a posteriori switches have to be done on the past sets of data. This can be the case when some parking lots were closed (usually long-term airport car parks) while others accepted the entire demand but with adjusted prices ;

• It is also possible that specific customer segments have been more affected than others (for example, depending on the destinations that were still active while others were closed).

Anyway, those 10 or 12 weeks of data (or more) definitely need to be flagged so as to be easily excluded from the forecasting models if needed.

Model adjustments

Direct data correction or exclusion are important but even more essential are the adjustments to be made on the forecasting models:

• One may be willing to switch from a “per-car-park” forecasting mode to a “per-set-of-equivalent-carparks” mode: the airport offerings being contracted for a while (airport or mall segments) or left as they are (but leading to an overcapacity situation), it may be of interest to have the forecasts be produced at a higher level (parking aggregates, yield class clusters, etc.)

• In case of AI forecasting models that adjust engine coefficients upon the day-to-day observations, it is important, during the recovery phase for example, to make sure that the weight put on the latest observations is high: the system will thus learn quicker from the trend than from historical data. It is only then that demand growth will be properly monitored and therefore exploited at its best efficiency (for example in the case of exponential smoothening models)

Price settings

Last but not least, even when the past data are properly corrected or excluded, even when the models are set to a mode where they learn better from the recovery phase than from ancient times, prices will have to be manually or semi-manually (rule based) set for a while (actually, as long as the recovery phase is not entirely over).

Manual price setting should not be done blindly: they need to be even more nourished by works on the analytics (daily observations of the entry volumes, customer behavior changes, etc.), parking competition tracking and airline forecasts computing for example. But one needs to leave time to the machine to learn the new demand again.

At this stage, from a business viewpoint, is necessary is to be highly “price agile”: be able to react to a sharp demand growth (even on limited volumes) for example.

The decision to switch back to the automatic price generation will be 100% linked to the reliability of the forecasts: during the recovery period where multiple yield-expert-works have to be done and where manual (or semi-manual) pricing is required, a continuous tracking of the forecasts vs the actual has to be done as well.

Only when forecast accuracy reaches back its original level can the dynamic pricing solution be made active again on an automated mode.

It is important to bear in mind, though, that when this phase is reached, even though the forecasts are stabilized again at a satisfying level of accuracy, the yield strategies may well have to be adjusted themselves. This occurs due to potential changes in the market behavior (for example, long stay tickets can easily be forecasted but they do not correspond to the level prior the crisis – because some competitors are not there anymore, for example – therefore leading to new and less risky yield strategies, etc.).

But after all, this is where yield managers are simply back to their daily lives!