Airline Metrics and Slippage
Airline seat inventory metrics, also known as KPI's or Key Performance Indicators, measure industry standard quantitative seat inventory metrics. A quantitative metric provides a measurement as opposed to a qualitative or intangible like user experience, quality and goodwill. One metric we discuss is a leading indicator of passenger behavior. Another is a lagging indicator of passenger behavior.
Think of how your price sensitivity changes when making tradeoffs for flexibility of travel plans. These are reflected by the class of fare's terms and conditions driving your purchase price. Another of our blog posts discussed the differences of direct and indirect measurement of a KPI which we consider in this post in more detail.
Metrics are meaningful for making comparisons. Our comparison basis uses time periods for comparing historical seat inventory performance at difference levels like comparing how the same aircraft inventory performs on different travel segments between cities. Other comparisons can be made inter-airline by evaluating seat inventory profitability, or on a single airline by comparing the load-factor of a flight by time of departure.
Analysis and comparison at the flight level relies on historical data instead of real-time data. This results in a missed opportunity in the timeliness of per-passenger preference, behavior and experience per-flight. The passenger experience and per-flight seat preference data were never captured because you were never asked or offered it.
Airlines decided to forego the ability to interact with the passengers for seating reassignments immediately prior to departure. The reservation systems produce per-flight metrics only when using historical performance data like load-factors and inventory performance. However, airplanes and the seats on them are not passengers because we have preferences with missed opportunity costs. We also need to make last minute changes or adjustments in the seating arrangement (for example pilot rebalancing of the aircraft).
We also wish to consider individual passenger metrics by using historical observation. We can produce a common basis for inter-airline evaluation from the passenger's point of view. Seating preference can vary by route, departure time, season, airline or even by specific aircraft model or tail number for a passenger. People behave differently when given choices previously unavailable.
This post introduces a passenger-centric metric called spillage. We first consider the current airline seat inventory performance metrics which are defined on a per-flight basis.
If we share a common definition about a metric we better interpret what they represent. Interpretation leads to analysis which leads to actionable information resulting in a decision. We would like to consider metric definitions that have the following attributes:
Another measurement topic is observability. An external observer, as a mathematical system model, may require directly measured values to calculate the underlying measured value. These techniques are frequently used in closed loop feedback control systems. Engineering projects often use indirect sensing from the transducer output with transformations to minimize system interaction.
Think of having the flu and going to a doctor. We expect a time difference between using an oral analog thermometer (the old glass style with red dyed alcohol) initially at room temperature to take body temperature versus using a contactless digital thermometer using infrared light to remotely sense ear temperature. You sit on the examination table for different durations as the assistant covertly counts your respiratory rate to avoid yet another interaction.
Our definitions include a well defined event, a method of capturing event occurrence and a well defined duration of measurement. This implies a sort of batch process (e.g. a weekly paycheck) rather than a continuous process (e.g. the taxi meter). If you were to think about it, you realize that unsold and oversold seats on a flight represent indirect measures of the supply-demand curve. These measures correlate to price point, on a given flight. The batch definition applies as defined by the passengers aboard a route. The flight duration is defined by the scheduled departure and arrival time. The flight event is defined as a batch process between the departure and arrival cities.
Baseline metrics provide reference measurement points from which we can derive or produce a variety of extended metrics (which are different from measurements). A derivative metric example could include aggregate measures of flight performance like load-factor averages and standard deviations. When we compare metrics with varying duration dimension we would "normalize" the data to a common value. This allows the data variance to reflect correlation to and identification of metric influencers like seasonal variation of load-factor compared year-over-year for the same flight. Adopting the airline revenue point of view, there would be no additional seat inventory spoilage or slippage indicating unsold or oversold seats on a given flight.
Spoilage and spillage metrics capture a numerator as event occurrences over a duration. Our metrics receive durations when we express the number of occurrences of either event. This would be reported as a time basis annually, seasonally, historically, and so on. The passenger experiences another definition for duration. The passenger experience metric for any given flight is the time difference between boarding an aircraft and departing the aircraft. The aircraft entry defines a physical boundary with regard to the passenger. We simply the model of passenger experience on a flight by reducing the definition to goodness of fit to the location of their seat and their preference for location.
A "cabin door closure" gives a crisp and well defined event. It can be thought of as an event that starts or stops your measurement duration. Think of the train conductor calls for "all aboard" or a bartender making "last call" at a pub. These notices to patrons and passengers prepare for the events of train departure or pub closing time. The Hopscotch SeatBot API value proposition to airlines really involves metrics taken a step closer to the passenger level to maximize revenue rather than the aircraft basis, which is a sunk cost.
Maybe you've also heard about a metric called "seat miles", which measures the distance traveled by unoccupied seats. This is also a lagging indicator related to capacity planning.
Seat spillage happens when all passengers booked on a flight arrive for the same flight. It is known to the public as over booking when a passenger gets "bumped" to another flight. Minimizing spillage events means preserving per flight revenue by avoiding rebooking, reimbursement and associated travel disruption expenses.
The household terms of spoilage and spillage describe their metric from as seat inventory. We like the household theme and propose a term for inter-passenger transactions.
Slippage would be a type of micro-economy of its own. Most domestic passengers travel with a mobile device as the interfaces move to support mobile devices. Given a transactional structure, a formal means of valuation and an exchange medium, airlines can mine a golden opportunity.
H₀: Null hypothesis. No additional revenue opportunity exists because all passengers are happy with their assigned seat.
H₁: There exists a population subset of passengers unhappy with their current seat.
You could make dozens of additional hypothesis but we focus on these based on our passenger survey data.
Hopscotch assumes that the passenger-to-passenger micro-economic interactions support a micro-transaction marketplace. We want to work with the airline to remove and reduct transaction friction causing slippage.
A macro-economic profit motive generally seeks to maximize return on investment on aircraft with efficient expenditures on operations (e.g. economies of scale for fuel contracts). The finer granularity at the flight level still inherits these economies of scale. A paradox arises where aggregate flight revenue is driven on a fixed supply on the supply-demand curve and measured by unit seat sales which in turn cause passengers to react differently (price sensitivity) and drive the seat price-demand curve. This is well established and handled brilliantly by the airline "fare class" tied to terms and conditions of duration of stay and advance purchase. Business travelers pay for flexibility with vacationers making advance decisions or making a plan and sticking to it.
Hopscotch Solutions wonders what happens when we consider slippage as having a marginal propensity to drive demand (i.e. consumer action). If this is a function of the lead-time before the experience consequence, then we want to find out if as the time approaches for the journey, might we be able to drive adjustments to the amount of passenger compromise based on a hidden availability of other seats assigned to existing passengers.
Think of how your price sensitivity changes when making tradeoffs for flexibility of travel plans. These are reflected by the class of fare's terms and conditions driving your purchase price. Another of our blog posts discussed the differences of direct and indirect measurement of a KPI which we consider in this post in more detail.
Metrics are meaningful for making comparisons. Our comparison basis uses time periods for comparing historical seat inventory performance at difference levels like comparing how the same aircraft inventory performs on different travel segments between cities. Other comparisons can be made inter-airline by evaluating seat inventory profitability, or on a single airline by comparing the load-factor of a flight by time of departure.
Analysis and comparison at the flight level relies on historical data instead of real-time data. This results in a missed opportunity in the timeliness of per-passenger preference, behavior and experience per-flight. The passenger experience and per-flight seat preference data were never captured because you were never asked or offered it.
Airlines decided to forego the ability to interact with the passengers for seating reassignments immediately prior to departure. The reservation systems produce per-flight metrics only when using historical performance data like load-factors and inventory performance. However, airplanes and the seats on them are not passengers because we have preferences with missed opportunity costs. We also need to make last minute changes or adjustments in the seating arrangement (for example pilot rebalancing of the aircraft).
We also wish to consider individual passenger metrics by using historical observation. We can produce a common basis for inter-airline evaluation from the passenger's point of view. Seating preference can vary by route, departure time, season, airline or even by specific aircraft model or tail number for a passenger. People behave differently when given choices previously unavailable.
Seat Inventory Metrics
This post follows up on our metrics discussion for airline seat inventory. Airlines, via software algorithms and recursive feedback loops, capture, measure, test and adjust the effectiveness of seat prices in many ways through several different channels. These two tangible metrics have dimensions, or attribute definitions, that allow us to directly count occurrences of empty seat and oversubscribed seat events for a flight.- metric time duration,
- metric event definition,
- method of measurement,
- direct and indirect observation,
- leading or lagging correlation to business metrics,
- and relation to the supply-demand and price-demand macroeconomic models.
- Statistical correlation to a model being positive (related) or negative (inversely related),
- Strength of correlation to other metrics as weak (closer to 0.0) or strong (closer to 1.0),
- Frequency histogram analysis,
- Underlying probability model,
- Analysis of Variance (aka ANOVA),
- Others.
Is The Metric The Measurement?
There is something about measurements and metrics to understand. The process of measurement affects the system under observation by the act of observation. This was originally expressed as the Heisenberg Uncertainty Principle of the wave function. Relativistic physics experiments bear this out.Another measurement topic is observability. An external observer, as a mathematical system model, may require directly measured values to calculate the underlying measured value. These techniques are frequently used in closed loop feedback control systems. Engineering projects often use indirect sensing from the transducer output with transformations to minimize system interaction.
Think of having the flu and going to a doctor. We expect a time difference between using an oral analog thermometer (the old glass style with red dyed alcohol) initially at room temperature to take body temperature versus using a contactless digital thermometer using infrared light to remotely sense ear temperature. You sit on the examination table for different durations as the assistant covertly counts your respiratory rate to avoid yet another interaction.
Our definitions include a well defined event, a method of capturing event occurrence and a well defined duration of measurement. This implies a sort of batch process (e.g. a weekly paycheck) rather than a continuous process (e.g. the taxi meter). If you were to think about it, you realize that unsold and oversold seats on a flight represent indirect measures of the supply-demand curve. These measures correlate to price point, on a given flight. The batch definition applies as defined by the passengers aboard a route. The flight duration is defined by the scheduled departure and arrival time. The flight event is defined as a batch process between the departure and arrival cities.
Seat Spoilage and Spillage
Seat spoilage and seat spillage are defined quantitative measures of seat inventory performance. The event will be an unoccupied seat, in the case of spoilage where the numerator represents the number of unsold or unoccupied seats on a flight. Seat spillage counts a slightly different event as the number of overcommitted seats on a flight. Spillage affects another flight in a latency between the time of purchase and the time of departure. Both metrics share a common definition of event duration as the flight time.Baseline metrics provide reference measurement points from which we can derive or produce a variety of extended metrics (which are different from measurements). A derivative metric example could include aggregate measures of flight performance like load-factor averages and standard deviations. When we compare metrics with varying duration dimension we would "normalize" the data to a common value. This allows the data variance to reflect correlation to and identification of metric influencers like seasonal variation of load-factor compared year-over-year for the same flight. Adopting the airline revenue point of view, there would be no additional seat inventory spoilage or slippage indicating unsold or oversold seats on a given flight.
Spoilage and spillage metrics capture a numerator as event occurrences over a duration. Our metrics receive durations when we express the number of occurrences of either event. This would be reported as a time basis annually, seasonally, historically, and so on. The passenger experiences another definition for duration. The passenger experience metric for any given flight is the time difference between boarding an aircraft and departing the aircraft. The aircraft entry defines a physical boundary with regard to the passenger. We simply the model of passenger experience on a flight by reducing the definition to goodness of fit to the location of their seat and their preference for location.
A "cabin door closure" gives a crisp and well defined event. It can be thought of as an event that starts or stops your measurement duration. Think of the train conductor calls for "all aboard" or a bartender making "last call" at a pub. These notices to patrons and passengers prepare for the events of train departure or pub closing time. The Hopscotch SeatBot API value proposition to airlines really involves metrics taken a step closer to the passenger level to maximize revenue rather than the aircraft basis, which is a sunk cost.
Other Metrics
Spoilage represents missed revenue opportunity in terms of unsold seats per flight. This metric came up during our industry research of seat inventory performance. A lagging metric would be captured after or at the end of the event measured, which happens when an aircraft leaves the terminal. Seat utilization per flight naturally lags flight revenue metrics like spoilage. Determining missed revenue from lagging indicators presents a valuation challenge for leading metrics like projected revenue.Maybe you've also heard about a metric called "seat miles", which measures the distance traveled by unoccupied seats. This is also a lagging indicator related to capacity planning.
Seat spillage happens when all passengers booked on a flight arrive for the same flight. It is known to the public as over booking when a passenger gets "bumped" to another flight. Minimizing spillage events means preserving per flight revenue by avoiding rebooking, reimbursement and associated travel disruption expenses.
The household terms of spoilage and spillage describe their metric from as seat inventory. We like the household theme and propose a term for inter-passenger transactions.
A New Seat Inventory Metric
Hopscotch internally uses the term "slippage" as shorthand notation for what SeatBot captures. Slippage represents missed top line revenue to airlines as additional revenue per flight from and between passengers. The SeatBot API service for airlines is a way to capture lost revenue that slips on a per-flight basis.Slippage
What is slippage in terms of passengers and seat inventory? What are the missed revenue opportunities? Because slippages are per flight micro-transaction events, we want to enable airlines to automate inter-passenger reseating requests and negotiate directly between passengers using SeatBot with airline oversight.Slippage would be a type of micro-economy of its own. Most domestic passengers travel with a mobile device as the interfaces move to support mobile devices. Given a transactional structure, a formal means of valuation and an exchange medium, airlines can mine a golden opportunity.
Model of Slippage
We would like to evaluate a model of passenger seat satisfaction level. This might be modeled using an Inverse Power Law, also called the "80-20 rule" or Pareto Principle, as defined in economic models. using the power law relationship shown in Equation 1: Pareto Probability Density function. Let's define a minimum seat satisfaction level
Let's say there is a seat satisfaction level, below which passengers would rather move to another seat.
Next, lets look to model seat satisfaction level overall, meaning for the passengers aboard a particular flight, with their current seat assignment. Here we need to search for a feasible point (i.e. feasible restricts meaning to be another whole seat, or integer) at which the passenger considers exchanging seats for a seat more congruent with their goal.
We can find the goodness of goal-to-seat fit where it reaches a reseating point, which we should label as the personal "worthwhile" valuation, and this would be aligned close enough to their location goal. Close enough requires relaxing some constraints such as to other nearby rows, for a column as side-to-side directed priority, or relaxed to include other nearby columns, for a row or front-to-back directed priority.
Below, we see an example histogram plot of the Pareto probability density distribution with a power of 3.0, or the alpha parameter from Equation 1. It uses the mode, m, located at 1.0 from the equation. We generated 200 random observations using the equation for the probability. Those values were binned to discretize the values into a frequency histogram plot.
Let's say there is a seat satisfaction level, below which passengers would rather move to another seat.
| Equation 1: Canonical Pareto Probability Density function, an Inverse Power Law |
We can find the goodness of goal-to-seat fit where it reaches a reseating point, which we should label as the personal "worthwhile" valuation, and this would be aligned close enough to their location goal. Close enough requires relaxing some constraints such as to other nearby rows, for a column as side-to-side directed priority, or relaxed to include other nearby columns, for a row or front-to-back directed priority.
Below, we see an example histogram plot of the Pareto probability density distribution with a power of 3.0, or the alpha parameter from Equation 1. It uses the mode, m, located at 1.0 from the equation. We generated 200 random observations using the equation for the probability. Those values were binned to discretize the values into a frequency histogram plot.
 |
| Figure 1: Frequency histogram with plot of Pareto Probability Density Function |
Pareto probability frequency of occurrence shows an inverse relation as the input tails off farther away from the mode. The "80-20 rule" refers to probabilities of 0.80 and 0.20 for the sum of observations above and below the mean. We would like to use this distribution as a behavior model of passenger behavior borrowed from the famous Pareto economic model.
Pareto Process Model
Passenger preferences and tastes change. Every passenger is micro-economically unique and yet many seat selection behaviors follow some established price sensitivity patterns. The Pareto probability distribution appears repeatedly across fields of study when analyzed from a probabilistic or predictive standpoint. We propose applying this probability distribution as a probabilistic model of the economics of slippage and therefore as an ancillary airline revenue model.Inputs, Outputs and Parameters
Assign the passenger satisfaction level with their assigned seat as a Pareto probability. Outside of economics, the Pareto distribution, also known as the "80-20 rule", or from mathematics, the Bradford distribution. We would like to form a statistical hypothesis test as the following:H₀: Null hypothesis. No additional revenue opportunity exists because all passengers are happy with their assigned seat.
H₁: There exists a population subset of passengers unhappy with their current seat.
You could make dozens of additional hypothesis but we focus on these based on our passenger survey data.
Data Sources
We draw on surveys from 2016 from a population sample of passengers who flew at least one flight in the last year. Two separate questions asked passengers to choose from fixed dollar amounts, including $0 and Not Applicable, as their maximum amount to offer another passenger for their seat and the minimum amount they would accept to exchange their own seat.Hopscotch assumes that the passenger-to-passenger micro-economic interactions support a micro-transaction marketplace. We want to work with the airline to remove and reduct transaction friction causing slippage.
A macro-economic profit motive generally seeks to maximize return on investment on aircraft with efficient expenditures on operations (e.g. economies of scale for fuel contracts). The finer granularity at the flight level still inherits these economies of scale. A paradox arises where aggregate flight revenue is driven on a fixed supply on the supply-demand curve and measured by unit seat sales which in turn cause passengers to react differently (price sensitivity) and drive the seat price-demand curve. This is well established and handled brilliantly by the airline "fare class" tied to terms and conditions of duration of stay and advance purchase. Business travelers pay for flexibility with vacationers making advance decisions or making a plan and sticking to it.
Hopscotch Solutions wonders what happens when we consider slippage as having a marginal propensity to drive demand (i.e. consumer action). If this is a function of the lead-time before the experience consequence, then we want to find out if as the time approaches for the journey, might we be able to drive adjustments to the amount of passenger compromise based on a hidden availability of other seats assigned to existing passengers.
Comments
Post a Comment