Checking for consistent consumer behaviour

Question

Hi All,
I have a table which looks something like this (the data is not accurate):
CUSTOMER     CREDIT ACTION     TIME_GAP
123     15-Aug-11     
123     18-Aug-11     3
123     24-Aug-11     6
123     9-Nov-11     77
123     14-Nov-11     5
123     18-Nov-11     4
123     30-Nov-11     12
123     14-Aug-12     258
123     20-Aug-12     6
123     24-Aug-12     4
The table shows me for a list of customers, the date on which some sort of credit action was necessary (reminder letter, phone call, threat of legal action etc.) and the time gap (days) between one action and the preceding action.
I am looking for consistent behaviour, e.g. if a customer has 3 credit actions required each within 90 days of the other, then we say that they are a credit risk as they consistently show that it is difficult to collect debt from them.
However if there is a break (e.g. 180 days) between the string of credit actions then we say that they have broken the habit of not paying on time and we would start counting the 3 consecutive credit actions (within 90 days of each other) again.
What is the most efficient way to write a query that will return customers for who the above description hold true?
TIA
ec :-)

Answer 1

Interesting problem.
Initial thought is to create a derived data set that contains a list of customers, with consecutive 3 credit actions, and the total days between these 3 actions.
Basic approach:
// sample table
SQL> create table customer_actions(
  2          cust_id integer,
  3          day date,
  4          constraint pk_customer_actions primary key
  5          ( cust_id, day )
  6  ) organization index;
Table created.
// sample random data for 2 customers
SQL> insert into customer_actions
  2  select     distinct
  3          123,
  4          trunc(sysdate,'yyyy')+trunc(dbms_random.value(1,1000))
  5  from       dual
  6  connect by level < 11;
10 rows created.
SQL>
SQL> insert into customer_actions
  2  select     distinct
  3          456,
  4          trunc(sysdate,'yyyy')+trunc(dbms_random.value(1,200))
  5  from       dual
  6  connect by level < 21;
20 rows created.
// credit actions against customers
SQL> select * from customer_actions order by 1,2;
   CUST_ID DAY
       123 2012/05/20
       123 2012/05/28
       123 2012/06/05
       123 2012/10/05
       123 2013/04/03
       123 2013/07/25
       123 2014/01/20
       123 2014/05/19
       123 2014/05/25
       123 2014/05/26
       456 2012/01/07
       456 2012/01/18
       456 2012/01/30
       456 2012/02/03
       456 2012/02/23
       456 2012/02/29
       456 2012/03/05
       456 2012/03/23
       456 2012/04/05
       456 2012/04/11
       456 2012/04/16
       456 2012/04/24
       456 2012/04/26
       456 2012/05/15
       456 2012/05/21
       456 2012/06/23
       456 2012/06/30
       456 2012/07/02
       456 2012/07/04
       456 2012/07/10
30 rows selected.
// basic derived data set - creating rows with 3 consecutive actions
// against a customer (rows without 3 actual dates need to be discarded
// when processing the data set)
SQL> select
  2          cust_id,
  3          lag(day,2) over (partition by cust_id order by day)     as DAY1,
  4          lag(day,1) over (partition by cust_id order by day)     as DAY2,
  5          day                                                     as DAY3
  6  from       customer_actions
  7  order by 1,4,3,2;
   CUST_ID DAY1       DAY2       DAY3
       123                       2012/05/20
       123            2012/05/20 2012/05/28
       123 2012/05/20 2012/05/28 2012/06/05
       123 2012/05/28 2012/06/05 2012/10/05
       123 2012/06/05 2012/10/05 2013/04/03
       123 2012/10/05 2013/04/03 2013/07/25
       123 2013/04/03 2013/07/25 2014/01/20
       123 2013/07/25 2014/01/20 2014/05/19
       123 2014/01/20 2014/05/19 2014/05/25
       123 2014/05/19 2014/05/25 2014/05/26
       456                       2012/01/07
       456            2012/01/07 2012/01/18
       456 2012/01/07 2012/01/18 2012/01/30
       456 2012/01/18 2012/01/30 2012/02/03
       456 2012/01/30 2012/02/03 2012/02/23
       456 2012/02/03 2012/02/23 2012/02/29
       456 2012/02/23 2012/02/29 2012/03/05
       456 2012/02/29 2012/03/05 2012/03/23
       456 2012/03/05 2012/03/23 2012/04/05
       456 2012/03/23 2012/04/05 2012/04/11
       456 2012/04/05 2012/04/11 2012/04/16
       456 2012/04/11 2012/04/16 2012/04/24
       456 2012/04/16 2012/04/24 2012/04/26
       456 2012/04/24 2012/04/26 2012/05/15
       456 2012/04/26 2012/05/15 2012/05/21
       456 2012/05/15 2012/05/21 2012/06/23
       456 2012/05/21 2012/06/23 2012/06/30
       456 2012/06/23 2012/06/30 2012/07/02
       456 2012/06/30 2012/07/02 2012/07/04
       456 2012/07/02 2012/07/04 2012/07/10
30 rows selected.
// using this derived data set, we determine the total lapsed days
// for the 3 credit actions (period between day1 and day2, plus period
// between day2 and day3) - the result can then be used (via a case or
// filter clause) to determine the customer's behaviour
SQL> with data_set as(
  2          select
  3                  cust_id,
  4                  lag(day,2) over (partition by cust_id order by day)     as DAY1,
  5                  lag(day,1) over (partition by cust_id order by day)     as DAY2,
  6                  day                                                     as DAY3
  7          from    customer_actions
  8  )
  9  select
10          cust_id,
11          day1, day2, day3,
12          round( (day2-day1)+(day3-day2) )        as DAYS
13  from       data_set
14  where      day1 is not null
15  and        day2 is not null
16  and        day3 is not null
17  order by 1,2;
   CUST_ID DAY1       DAY2       DAY3             DAYS
       123 2012/05/20 2012/05/28 2012/06/05         16
       123 2012/05/28 2012/06/05 2012/10/05        130
       123 2012/06/05 2012/10/05 2013/04/03        302
       123 2012/10/05 2013/04/03 2013/07/25        293
       123 2013/04/03 2013/07/25 2014/01/20        292
       123 2013/07/25 2014/01/20 2014/05/19        298
       123 2014/01/20 2014/05/19 2014/05/25        125
       123 2014/05/19 2014/05/25 2014/05/26          7
       456 2012/01/07 2012/01/18 2012/01/30         23
       456 2012/01/18 2012/01/30 2012/02/03         16
       456 2012/01/30 2012/02/03 2012/02/23         24
       456 2012/02/03 2012/02/23 2012/02/29         26
       456 2012/02/23 2012/02/29 2012/03/05         11
       456 2012/02/29 2012/03/05 2012/03/23         23
       456 2012/03/05 2012/03/23 2012/04/05         31
       456 2012/03/23 2012/04/05 2012/04/11         19
       456 2012/04/05 2012/04/11 2012/04/16         11
       456 2012/04/11 2012/04/16 2012/04/24         13
       456 2012/04/16 2012/04/24 2012/04/26         10
       456 2012/04/24 2012/04/26 2012/05/15         21
       456 2012/04/26 2012/05/15 2012/05/21         25
       456 2012/05/15 2012/05/21 2012/06/23         39
       456 2012/05/21 2012/06/23 2012/06/30         40
       456 2012/06/23 2012/06/30 2012/07/02          9
       456 2012/06/30 2012/07/02 2012/07/04          4
       456 2012/07/02 2012/07/04 2012/07/10          8
26 rows selected.I kept on wanting to write Day Z... as in World War Z.... ;-)