Redefining Revenue Management, Marketing, Data and Analytics through the COVID-19 shock
Part 2 of 2: Adjustments to the analytical models
by Kelly McGuire and Naman Gandhi
In the part one of this two-part series, we discussed how to overcome the data disruption caused by the pandemic by fine-tuning the data signals fed into the models. Simply ensuring that models rely on more recent signals may be sufficient to improve the reliability of some models, whereas others may need to be temporarily replaced with a different methodology until patterns stabilize.
Premise: The crisis has temporarily rendered typical historical reference points for analytics practically meaningless, including historical booking patterns, market segmentation, buying behavior, price elasticity, and no show and cancellation patterns.
Solution: This drastic change warrants a closer look at the models that rely on these data points.
In this part of the series, we will focus on the analytical strategies and algorithmic adjustments that can help adapt to “Business Unusual” by discussing adjustments to the most impacted analytical models: demand forecasting, price sensitivity and consumer behavior.
Demand Forecasting
Time series forecasting is widely used in hospitality, not just because it is easy to access, but also because it is particularly effective when there is a strong and stable relationship between what happened in the past and what will happen (seasonality, trends, averages). Until the pandemic disruption, time series provided a stable and reliable forecasting method for most hospitality needs. Now that the immediate crisis is past, analysts may be tempted to continue to use these algorithms as-is by considering the initial pandemic months of muted demand (March, April, May, for example) as outliers and exclude them from the data the algorithm uses to predict the future.
Unfortunately, in today’s environment, this approach would still result in incorrect demand forecast. Tagging extreme pandemic impacts as outliers and continuing with “business unusual” data would only work if the level of demand changed, but the underlying patterns still existed (seasonality, peak days, events, segment mix). In the case of COVID, demand has fundamentally changed. For example, demand drivers like sporting events or festivals have been cancelled and people are no longer traveling for conventions and business, which were big drivers of seasonality patterns. Further, at least during the current business unusual phase, all demand is extremely volatile, subject to rapidly changing travel and lodging sentiment, infections/vaccinations trends, and re-opening procedures. Previous patterns no longer exist.
Which begs the question, should we even bother forecasting demand right now? Obviously, hotels still need some sense of expected demand for pricing, planning and operational purposes, and they will likely need to balance that forecast against other market (competitor rates, demand drivers) and operational factors (labor costs, debt servicing needs) when making any decisions based on the forecast.
In a sense, today’s demand planning can be thought of like opening a new property in an unknown market with limited competition. There is no historical data to rely on, so rather than analyzing previous patterns, we need to look extrinsically to understand expected demand levers and demand materialization, relying on leading market indicators and very recent demand observations. A methodology that heavily weighs current conditions, recent demand signals and that can handle a good amount of volatility is best suited for these current (and hopefully, short term) conditions. Given this, and the ongoing need for “expert” intervention to interpret outcomes, the exercise should really be thought of more as “demand sensing” as opposed to demand forecasting.
The demand sensing models would augment the volatile demand and occupancy signals of recent times with leading demand indicators from social trends and macroeconomic factors to more accurately estimate near-term demand. Businesses can build these demand sensing models via Bayesian probabilistic prediction models to get real-time expected demand broken down at region, property and segment level. Bayesian techniques can help tease out the uncertainty that is prevalent in the market. Priors can be set as revenue managers’ demand hypothesis (based on day-to-day operational insight) and posterior can be observed by measuring the market reaction to pricing and other evolving customer utility/preferences.
The advantage of this methodology is that it is tuned to current conditions, however, the unpredictable and unprecedented nature of this current crisis will still require human involvement. “Experts” will need to review the forecast with an eye to their understanding of market behavior, and evaluate the uncertainty and risk associated with the model outputs before they are implemented.
Pricing Elasticity and Revenue Optimization
Despite demand drops, revenue managers need to resist the urge to simply offer deep discounts to drive demand. Price sensitivity, or price elasticity, is a particularly important metric during low demand periods. Traditional yielding methods like setting thresholds and restricting availability work best during periods of high demand. If these methods are used in low demand periods, inventory would be wide open at the lowest rate. Understanding the true price sensitivity of demand, however, lets you find a few extra dollars in rate when the potential is there, but appropriately discount when the market dictates, regardless of whether the hotel even comes close to selling out.
Revenue managers have relied on revenue management systems to support pricing decisions, but, as these systems tend to rely on stable historical demand patterns, it will take time and human intervention in the short term to pick up on new patterns. In the meantime, careful tracking of demand characteristics like source market, rate segment, booking windows and booking patterns will provide some insight into changes in demand patterns and willingness to pay. However, this information is not enough to truly understand the shifts in price sensitivity.
“Elasticity should be measured continuously over the booking horizon.”
An incremental step towards assessing elasticity is to use pricing from your close competitor set to understand sensitivity in the market overall. Rather than just letting your rates ride at the lowest levels, the market can act as proxy for what guests are willing to pay and then targeted adjustments can be made, and their impact on market and hotel demand been tracked. Of course, this assumes the market is adapting to consumer willingness to pay, and not just deep discounting. Further, the composition of the competitive set may have shifted with pandemic closures and restrictions. So, market data is helpful, but certainly not perfect.
Experimentation is a more robust way of measuring elasticity, and provides the additional benefit of providing rapid, near-real time feedback. This “Test & Learn” approach should be designed to analytically explore the impact of pricing variations on buying behavior, and ultimately find the optimal price based on price elasticity at any given time. Intentionally varying price, rather than simply riding at floor levels, will give a much fuller picture of market potential.
Price experiments can also tease out consumer preferences by segment. Purposefully testing how deep (or not) you need to incentivize each segment can inform marketing efforts as well as pricing strategies. Experiments could evaluate upsell or value-add promotions to loyalty guests to determine the best driver of conversions, or test discount levels to price-sensitive segments to avoid unnecessarily dramatic discounting. Price sensitivity experiments (see figure 1) will help analytical models in the revenue management system adapt much faster, as they will put a wider variety of prices into the market sooner than normal business operations might have, so the system would have more robust data to update sensitivity.
Revenue managers must be sure to invest the effort in channeling all results of manual manipulation, market evaluation and experimentation into the revenue management system. Overlaying human judgement and current market conditions will recalibrate the analytics and help the system to recover faster. Manual inventory management may seem the best path forward for now, but in the long-run it can never match the possibilities offered by algorithms that can manage millions of price decisions on a daily basis. We need to set these systems right by feeding the right demand signals and work with the machine in tandem until they can run again independently.
Guest Segmentation and Marketing Promotion recommendations
Given restricted budgets due to pandemic cutbacks, it is even more important for hospitality companies to invest their marketing budget wisely, targeting segments that are traveling now, while keeping relationships with loyal guests warm for when they are comfortable to travel again. However, much like the revenue management models we discussed previously, consumer behavior modelling traditionally has relied on historical data.
In order to overcome the lack of historical data, for an analytical evaluation of the behavior of the current traveling guests, we propose a framework (Figure 2) that analytically assesses guest behavior on three factors:
1. Recent price elasticity: spend on most recent bookings will update willingness-to-pay (represented by bubble size)
2. Recent Travel and Stay Intent: measuring guest’s recent search patterns and/or actual bookings can act as a proxy for customer’s degree of travel confidence (willingness to travel and drive vs. fly preferences) (represented by X axis)
a. Recent travel patterns are compared to pre-COVID travel share by markets to evaluate evolving guest’s intent over time (represented by pie-chart for each customer)
3. Loyalty Value: retain the measure of guest’s value from pre-COVID in order to keep our loyal customers engaged. (represented on Y axis)
Based on continuous analytical assessment of elasticity, combined with travel intent and prior loyalty behavior, promotional interventions can be designed that match the guest’s preferences against available supply. The below prioritization strategy places guests in the appropriate categories for action.
1. Pursue Aggressively the guests that have demonstrated local intent as markets have opened, irrespective of previous loyalty status. Offers can be designed based on elasticity and travel patterns
a. Guest/Segment A and B are rather price inelastic, presently have ‘Drive-to Market’ confidence/intent and are most loyal. Their pre-COVID travel patterns also corroborate with their Drive-to confidence, so clearly this segment is recovering faster. If “Upper Upscale or Luxury” properties are open in their drive-to market, this is the right offer for that guest.
b. Guest/Segment C have not been very price sensitive in recent months, but they had relatively low prior loyalty value. The marketer could offer them in Midscale properties in their local market and could experiment with or Upper Midscale/Up-scale property recommendations as well to assess if they would be willing to pay for those properties.
2. Maintain Relationship with the loyal customers, irrespective of their current travel intent, through relevant promotions. Engage with promotions according to current travel intent, or with relevant general travel content if they don’t appear to be traveling currently
a. Guest/Segment G are some of the many loyal customers, but don’t have demonstrated current travel intent. They are guests of tomorrow and will react as broader travel demand strengthens, so the marketer should keep them warm by sending advance 90–180 day booking promotional options with flexible cancellation policies. Alternatively, provide aspirational general travel content to maintain engagement.
3. Defend and Grow Customers that have regional or minor international travel intent and have demonstrated loyalty in the past. Find incremental revenue opportunities and exploit willingness-to-travel.
a. Guest/Segment D is largely inelastic right now and has medium loyalty value. It is best to invest in growing these customers by engaging them in up-sell offers. Whereas, marketing efforts on the Guest/Segment E can be for advance international offers or local domestic offers to match their travel confidence and intent.
4. Deprioritize customers that have low loyalty status, no current travel intent and dominant historical international travel patterns.
a. Guest/Segment F are guests of tomorrow with lowest historical loyalty and will react as international travel demand strengthens and borders opens, so the marketer should deprioritize any immediate investment.
This model relies on frequent evaluation of guest value and guest elasticity. Conducting pricing experiments, as we described earlier, can help measure elasticity so that marketing interventions avoid heavy discounting or wasted efforts in the short-run.
Conclusion
The demand shock induced by this pandemic will have lasting effects and as economies get ready to rebound with mass-inoculations, you need to be prepared to emerge into the new normal not just with a different operating model than in pre-COVID times, but also using different data and analytical approaches until steady state returns. Even if travel behaviors change forever, historical data and pre-pandemic analytical approaches will be valid again as patterns stabilize.
A few parting thoughts for the analytical leaders:
- Be ready to look under the hood until the patterns stabilize by quantifying the data drift and incorporating the leading indicators of demand recovery and suppressing longitudinal data elements.
- Continue to find avenues to rationalize and allocate your marketing budget by assessing guest’s elasticity, travel intent/confidence and loyalty to create a prioritization scheme. This will help tailor promotional recommendations to guests that have highest travel ability/confidence, lower price elasticity and are brand loyal.
- Finally, be agile! Keep an eye on leading indicators and continue to assess the volatility of the market. You need to be ready to take advantage of current market opportunities, as you prepare for the New Normal.
