Question

I have a telephony app which has a prompt which requires user choice. I made the app select one of 10 different phone prompts based on the last digit of the caller’s phone number. Then I measure whether the user responds to the prompt (accept) or decides to skip to the next step (reject). I thought this would work well enough as a random selection, but I think I may be wrong.

What I’m finding is that the exact same prompt has a dramatically different response rate (25% vs 35%) for two different last digits. Now I’m curious why this is. Does anyone know how phone numbers are assigned and why the last digit would be significant?

Answer 1

I checked our billing database. We use Asterisk as PBX and store billing in PostgreSQL database.

select substring(cdr_callerid_norm from '[0-9]$') as last_digit, count(*)
from asterisk_cdr 
where cdr_callerid_norm is not null and length(cdr_callerid_norm) > 2
group by last_digit 
order by last_digit

Result:

 last_digit | count
------------+-------
 0          | 17919
 1          | 13811
 2          |  8257
 3          | 20708
 4          | 13492
 5          | 13708
 6          |  8813
 7          |  6943
 8          | 11693
 9          |  7942
                        |  2584
(11 rows)

To me those numbers are not random. You can do similar thing with your billing and check it. I think phone numbers can be random in general, but if few customers calls you much more often then callers number will not be random. Consider using something other to vary prompt: random number, use time etc.