Fare family designer

How it works

Customers

2,000 synthetic shoppers for one market, drawn once from a fixed random seed () and stored in farefamilies.json, so the page gives the same answer every time. Each belongs to one of four segments. Each customer has a willingness to pay W for a bare Basic seat on this flight rather than their next-best option (another airline, another day, not travelling), and a value v_a for each attribute a at its full level. Both are lognormal around the segment's median, W with a log-standard deviation of 0.3 and the attribute values with :

Loyal elites put almost no value on bags, seats and priority because their status usually gives them those already; what a fare adds for them is status credit. The travel policy of the business segment doesn't allow Basic fares.

Families and levels

A family includes each attribute at a level ℓ between 0 and 1: included (1) or not (0), with two in-between levels. Changes with a fee count as 0.4 of free changes. Reduced points and status count as 0.5 of full earning: Air Canada's Basic earns points but no Status Qualifying Credits, and United's Basic Economy earns miles and PQP but no Premier qualifying flights.

Choice

For customer i and family f priced p_f, the surplus is

S_if = W_i + Σ_a v_ia ℓ_fa − p_f + Σ_{e ∈ {bag, seat}} max(0, v_ie(1 − ℓ_fe) − c_e),

where c_e is the à la carte price: a customer buys a bag or a seat separately when the family doesn't include it and it's worth more than its fee. Not buying has a surplus of 0. The customer then picks with a multinomial logit:

P_i(f) = exp(S_if / μ) / (1 + Σ_g exp(S_ig / μ)),

with μ the randomness slider. As μ falls, everyone takes the option with the highest value minus price. Results add up these probabilities over all customers rather than drawing one choice each, so the charts move smoothly. Basic bought with a bag or a seat on top is shown as "Basic + extras".

What the tiles measure

• Revenue per passenger: expected fares plus à la carte fees, divided by expected passengers. Per 100 shoppers also counts those who don't buy, which is what the step chart maximises: a higher step can raise revenue per passenger while losing passengers.
• Upsell rate: passengers in any family above Basic, as a share of all passengers.
• Buy-down: passengers who take Basic although they would have bought a higher family had Basic not been on sale (their surplus for it is positive). It's the price of offering Basic at all, and it's shown as a share of all passengers and of Basic passengers.

The step chart

The chart re-runs the model for every step between Basic and the second family from $0 up to the third family's step (or $250), in $5 increments, holding everything else fixed, and marks the step with the most revenue per 100 shoppers. It's a one-dimensional search: the other steps and prices aren't re-optimised along the way.

Presets and sources

The presets take what each airline's economy families include from its public pages. Prices, the Basic fare and the à la carte fees in every preset are illustrative, and all presets run on this page's synthetic short-haul customers, so they show how each ladder design would sell here, not what it earns for the airline.

The segment structure (price-driven leisure, travellers who need bags, managed business travel, frequent flyers) follows the usual description of branded fares in the literature, for example Fiig, T., Isler, K., Hopperstad, C. and Belobaba, P. (2010), "Optimization of mixed fare structures: Theory and applications", Journal of Revenue and Pricing Management 9(1–2), 152–170, and Train, K. (2009), Discrete Choice Methods with Simulation, 2nd edition, Cambridge University Press (the logit model, chapter 3).

Limitations

• Every family is always on sale. Real airlines close cheap booking classes as the flight fills, which changes the mix a lot; that side is on the booking limits and buy-down pages.
• No competitor on screen. The "not buying" option stands for every alternative at once, so the model can't show a rival matching or undercutting a step.
• The logit treats all options as equally different from each other (independence of irrelevant alternatives): adding a family that's almost a copy of another one takes share from every option, not mostly from its twin.
• Attribute values are drawn independently of each other and of willingness to pay. Real customers who value refunds also tend to value changes, and richer customers value most things more.
• Points cost money. Here, loyalty earning is free for the airline; in reality every point issued is a liability (see the breakage page), so full earning in a cheap family has a cost this page leaves out.
• Only bags and seats are sold à la carte. The carry-on, changes, refunds and status can only be had in a family. Change fees and partial earning are simplified to one level each, and nobody actually changes or cancels.
• One market and one flight, with no connections, no groups, no corporate discounts and no loyalty tiers inside the elite segment.

Fences and branded fares in Revenue Management 101, chapter 01; buy-down and forecasting when customers pick the cheapest fare in chapter 05; ancillaries and offers in chapter 08; and the seat-control side in the buy-down simulator. This is a personal project and isn't affiliated with any airline.