Simulating Only “Meaningful” Connections
It’s Not So Simple Connecting The Dots
Using advanced algorithms to better plan airline connections is only being smart in an intensely competitive modern transportation industry. There are sophisticated tools and methodologies that can help ferret out the truly “meaningful” connections.
An airline’s network planning and scheduling teams constantly evaluate new and existing markets to ensure that the carrier remains profitable in those markets, as well as across the network as a whole.
Analysts use various commercial tools to synthetically generate an airline network (taking into account an array of connections and itineraries), and carry out different estimations and analysis to determine the profitability potential of each route.
Building an airline network is the first essential step in that exercise.
To create connection simulations, different commercial tools are designed using different methods or algorithms (such as the well-used hill-climbing algorithm), as well as different heuristics.
Connection simulation is generally driven by sets of rules on different network parameters, such as level of “circuity” (the directness of the actual route through the necessary connections), maximum connect time and so on.
Old Method Versus New
|Connection-Building Metric||Standard Method (flat percentile)||New Method (reverse percentile)|
|Itinerary Match: Other Carriers||76%||86%|
|Overbuild: Other Carriers||5||4|
|Itinerary Match: Host Carrier||85%||89%|
|Overbuild: Host Carrier||4||3|
When comparing the standard method and the proposed new method using MIDT data of a large carrier, the important connection-building metrics are populated for both methods; however, the new method generate superior results in all instances.
The most common challenges faced by analysts during connection simulations include:
- Building too few connections or itineraries (leaving out the most important ones),
- Building too many connections, resulting in dilution of the share of important ones,
- Taking too long for the connection-building process to complete (result: loose parameter bounds).
However, there is an advanced (but not overly complicated) method of building only the “meaningful” connections, and hence a meaningful network in which the underperforming connections can actually be kept to a minimum.
On the part of the analyst, this process requires a clear understanding of comparative methodology relating to network parameters, as well as the ability to compare different connection-building metrics of diversely derived methods.
Let’s first try to more fully understand what the word “meaningful” really stands for in the context of a connection or an itinerary.
When thinking about it from the passenger’s point of view, it can encompass the characteristics that determine the passenger’s choice of one itinerary over another.
Circuity can be considered as one such characteristic, and can be used as a network parameter in a connection simulation.
Also, it’s sometimes quite a challenge to simulate a network for a carrier that flies to many continents, as the passenger-circuity choice patterns might be different from one region to another.
For example, in well-connected European markets, people can reach their destinations either with a nonstop service (circuity = 1), or perhaps with only one connection at most. But things could be quite different in Africa, where the connectivity is not as robust as in Europe or the Americas.
Advanced Way To Build Meaningful Connections
Today, there is an advanced (but not terribly complicated) method of building only the "meaningful" connections, which will produce a meaningful network in which the underperforming connections can actually be kept to a minimum.
This prompts a certain curiosity (or perhaps even an absolute necessity) to look at circuity by different geographies, which can either mean by market or by a group of markets that are similar in nature.
There are different ways of grouping markets, but the “market-grouping” discussion is really outside the scope of this article. So for simplicity, let’s assume there are many such market groups.
Once these market groups are defined, we can figure out each respective market group’s circuity percentile, based on any historical flown or booking data (for example, Marketing Information Data Transfer, or MIDT).
Using these circuity percentiles, as applied on a schedule of all carriers (OA or “other-airline” schedule), a global airline network can be simulated. This will generate all the possible itineraries that are less than or equal to the respective market groups’ circuity percentiles.
While generating this information, we also need to ensure that these itineraries cover at least a certain substantial percentage of global passengers, such that all subsequent modeling works well.
The standard practice (or at least a very common practice) is to take 95th-circuity percentiles, with the expectation of covering 95 percent of the global passenger totals (or 95 percent pax).
But this expectation is generally flawed, as the circuity densities are not the same for all of the market groups.
In other words, the “left-out” 5 percent pax (if we take 95th-circuity percentiles) in Europe will be much higher than the “left-out” 5 percent in Africa. Effectively, then, it’s not 95 percent of the global pax that are covered, but some percentage less than that.
A common solution to this problem might be to simply raise the market groups’ circuity percentiles to some even higher number (which, in any case, must be greater than 95).
While this would work in covering at least 95 percent of the global pax, it would generate too many “non-meaningful” itineraries, and in that process would bring a lot of what can only be referred to as “junk” into the system.
The most effective solution to this problem would be to set a global-pax coverage target (such as 95 percent), and then essentially work backwards to find out the respective market groups’ circuity percentiles using optimization methods.
Reverse percentile is one such method that can actually be tested very easily. This will curb the “junk-itineraries” issue to a larger extent, and will basically only generate the “reality-level” itineraries that are needed.
To test this method on a large carrier that flies pretty much all over the world, some figures have been generated.
Keeping all other conditions unchanged, here are the resultant comparisons among different connection-building metrics (comparing the two methods, which are the standard flat-percentile method in comparison to the new reverse-percentile method):
- If the connection-building metric is “itinerary match,” under the standard flat-percentile method the number comes out to 76 percent, but under the new reverse-percentile method the number lands somewhat significantly higher at 80 percent.
- If the connection-building metric is overbuild, under the standard flat-percentile method the number is 5, but under the new reverse-percentile method the number is only 4 (this is a “good” reduction)
- If the connection-building metric is “itinerary match” for the host carrier, under the standard flat-percentile method the number is 85 percent, but under the new reverse-percentile method the number is increased to a pretty impressive 89 percent.
- If the connection-building metric is overbuild for the host carrier, under the standard flat-percentile method the number is 4, whereas under the new reverse-percentile method the number is only 3 (again, a “good” reduction)
Note that “itinerary match” is defined by the percentage of historical pax covered by the simulated itineraries, and “overbuild” is defined by the extra itineraries that are created for every matching itinerary in history (which is why a lower “overbuild” number is actually the better number for that particular comparison).
The fundamental point is that the new method works much better in all comparative instances.
The new reverse-percentile method fits much more tightly with history.
However, from an operational standpoint this can sometimes be considered a challenge, and that’s especially true under circumstances in which systems (i.e., a calibration solution) are not refreshed regularly.
If there is a need to assess any new-connection market in a less-connected geography or market group, there might be a necessity for a manual “override,” and this is perhaps the only real drawback when using this new reverse-percentile method.
Similar treatment can also be extended to other network parameters, such as maximum connect time, service-operational flag, elapsed time and so forth.
Circuity can also be quite properly viewed as a function of distance. So there can be a significant correlation between origin-and-destination circuities and distances.
Quite arguably, the leading commercial simulation tool in the airline industry today is Sabre AirVision Profit Manager, on which Sabre continually updates the best and latest simulation technology.
In general, connection simulation methodology continues to evolve, and with each new development it comes closer to real-life scenarios, which represent vital information for airlines in planning and evaluating connections both present and future.