02-04-2015 11:51 AM
I am trying to build a customer profile and collected 12 months historical
data. The data set contains CustID, 12 month avg number of deposits and average $ deposited.
The customer profile contains multivariate numerical summaries behavior pattern, so future
behavior of the client can then be compared with his/her profile in order to
exam the consistency with it (normal behavior) or any deviation from client
profile on monthly basis.
Any thoughts on this?
02-04-2015 01:47 PM
The type of customer profiles you build will depend on the purpose for these. It appears that so far you just focused on customer deposits. It would suggest you should also look at withdrawals and transactions in total, along with account balance. Whether the account as an overdraft facility (a limit) would also be useful.
If the profiles are to be used for marketing purposes, then you would also need to focus on account behaviour scores (if lending accounts) and also track over limit behaviour.
Don't forget customer personal attributes as well, like age, time with bank, employment status, income etc.
02-04-2015 03:30 PM
What if the client has a different periodicity for a behavior than monthly. For example self-employed and paying estimated income tax quarterly. You would see a "spike" in withdrawal totals every three months and those months would not "match" a monthly profile.
Also some type of payment may be received every 6 months or annually, especially if it is a big percentage of total deposits.
You might need to segment your client based on some criteria before using a simpler monthly comparison.
And don't forget that the average working stiff getting paid every two weeks has 2 months every year with 3 paychecks instead of 2.