AL - airline
AP - airport
MS = metasearch = partner - web where people search for airline tickets (e.g. skyscanner, kayak, momondo), they usually don’t have content, they just calls APIs of its content providers (ALs, us, …)
src, dst - source and destination cities
Route - (src, dst) pair
Search - when a user search for how to get from src to dst
SERP - search engine result page
Itinerary - one particular search result, schedule of particular flights
Convenient itinerary - itinerary that is reasonable to buy in terms of price, time, stopover number, …
Stopover - when you get out of a plane, wait on AP few hours and get on following (connecting) flight of an itinerary
OW / RT - oneway / return itinerary - if you want to go only A->B or A->B and later back B->A
IATA = International Air Transport Association - every AP has its 3 digit IATA code (E.g. Prague - PRG), sometimes abbreviation misused as IATA = AP code
Click - when someone click on particular itinerary on MS SERP and is redirected to Kiwi.com booking page
Payout - the price we pay to MS for each click, usually around 1$, no matter if the user buys the itinerary
BP = booking page - web page of a particular itinerary where you fill name, surname, passport number, credit card and click “Pay”
Booking - purchase of air ticket
CR = conversion rate - number_of_bookings / number_of_clicks, usually between 5 and 10 %.
Margin = markup - difference between price for which Kiwi.com buys the tickets and price customer pays us (covers our company expenses, salaries, customer support, guarantee, payout, …), usually around 10%
Feature (signal) - a value of some property of data (e.g. price, duration, number of passengers) from which ML creates a model
Ranker = model - the result of ML, a piece of software that predicts whatever it was trained for
Rank - answer of a model, number, usually float, in our case usually between 0 and 1, in our case can be interpreted as probability
Split = (Feature, Threshold) - node in tree, examples with value of feature less than threshold goes left, other right
Threshold - number that is used to interpret float rank as bool: if (rank > threshold) then True else False
(so we’ll use term ‘threshold’ in those two different meanings)
FP, FN - false positives, false negatives, examples misclassified as positive / negative while they are of opposite type
Bias - some property of a (training) data that you’d like not to influence the model (= be ignored by the model) (e.g. more expensive ticket are less likely to be bought, tomorrow tickets are more likely to be bought than those with departure in 6 months, …)
AL industry - different to any other, flexible demand and non-flexible supply result in price games (the later you buy the more you pay, RT tickets over weekend cheaper cause bought by tourists while the other by businessmen, same ticket at the same time more expensive for US ppl than for Czechs, ...)
Kiwi.com - pirate innovative company between ALs and MSs