How to solve problems with the payment gateway: Airbnb Case
Today we decided to analyze the experience of IT specialists from Airbnb, which encountered problems in the work of the payment gateway.
/ Photo by frankieleon / CC
As part of this case we are talking about the problems that may arise with any international company. Airbnb is already working not only in the USA, but in dozens of other countries where there are particular features of the payment systems.
')
Currently, Airbnb not only accepts payments, but also transfers funds to apartment owners, gives gift certificates and makes settlements with other counterparties. All this complicates the conditions in which the team of IT project specialists has to work. Of the most common problems: there are difficulties with processing various types of currency or with loss of access to a specific payment gateway that processes transactions.
In addition to frequent A / B testing of various solutions, the team has developed a special anomaly detection system. It is adapted to work in real time and to assess the situation based on a search for emissions in a sample of time series.
For this system, Airbnb takes into account the effect of seasonality, which makes it possible to eliminate systematic errors in calculating simple regression using the least squares method .
Representation of the seasonality model is performed using the fast Fourier transform (FFT). After applying the FFT to the time series, it is possible to select the time intervals on which amplitude jumps are observed, indicating seasonality, and all others can be discarded.
To determine the trend of time series, analysts use the sliding median. For example, to identify a trend on a specific day. The advantage of using the median instead of the expectation is higher stability in the event of outliers.
Estimation of the magnitude of the error makes it possible to understand whether there are anomalies in the array from the time series. Depending on the allowed number of errors, it is possible to choose the number of standard deviations from zero. The anomaly warning system helps Airbnb to track most of the biases in error. Such tools may be useful in relation to other types of businesses that monitor certain statistical parameters of their work.
PS A little about the work of our IaaS provider:
/ Photo by frankieleon / CC
As part of this case we are talking about the problems that may arise with any international company. Airbnb is already working not only in the USA, but in dozens of other countries where there are particular features of the payment systems.
')
Currently, Airbnb not only accepts payments, but also transfers funds to apartment owners, gives gift certificates and makes settlements with other counterparties. All this complicates the conditions in which the team of IT project specialists has to work. Of the most common problems: there are difficulties with processing various types of currency or with loss of access to a specific payment gateway that processes transactions.
In addition to frequent A / B testing of various solutions, the team has developed a special anomaly detection system. It is adapted to work in real time and to assess the situation based on a search for emissions in a sample of time series.
For this system, Airbnb takes into account the effect of seasonality, which makes it possible to eliminate systematic errors in calculating simple regression using the least squares method .
Representation of the seasonality model is performed using the fast Fourier transform (FFT). After applying the FFT to the time series, it is possible to select the time intervals on which amplitude jumps are observed, indicating seasonality, and all others can be discarded.
To determine the trend of time series, analysts use the sliding median. For example, to identify a trend on a specific day. The advantage of using the median instead of the expectation is higher stability in the event of outliers.
Estimation of the magnitude of the error makes it possible to understand whether there are anomalies in the array from the time series. Depending on the allowed number of errors, it is possible to choose the number of standard deviations from zero. The anomaly warning system helps Airbnb to track most of the biases in error. Such tools may be useful in relation to other types of businesses that monitor certain statistical parameters of their work.
PS A little about the work of our IaaS provider:
Source: https://habr.com/ru/post/271567/
All Articles