Be more selective with your data, 3 months is not indicative of useage over 12/18/24 month contract
-
martin commented
Also have option to use maximum minutes used per month rather than average.
Say I use 180+180+240 minutes in 3 months, average is 200, but if I bought a new contract based on that the extra 40 minutes in month 3 would be an extra £15 or so.I would probably be better getting a 300 min contact in the first place. -
Hans van Baalen commented
Agreed, would like to have the option to have a longer period evaluated!
-
Anonymous commented
you seem to average things including my travel over the last three months. if you looked at my bill over the whole year you would see that there is no such thing as an average travel month. There are some holidays I do take at the same time each year. There is also usually a couple of random trips where I use the phone a lot. I do not think your recommendations take account of this usage pattern.
-
WobWob commented
Rather than fiddling with the sample period maybe you could allow users to nudge up or down the average values if they feel they are not representative.
-
mac commented
Hi guys,
This is a quote from Chris, one of the Billmonitor mathematicians. He explains it better than I could. Hope this clears up why we've gone for 3 months a bit.
"We did quite a bit of research on the optimal time-window of bills to analyse. 12 months is optimal if users behaviour remains more or less static over the year as it smooths out erratic outlier months. However, what we found is that for many users we see an increasing trend in usage, with people using more {minutes, texts, data} month on month. For these users, their usage 12 months ago is less representative of next months usage than say more recent months. When we did the research we found that 3 months (actually it was 2.8 months) provided the most accurate results on average across our user base.
It's a delicate balance. Ideally we would want to tailor the number of months to use to each user depending on individual usage trends -- but this is very hard to optimise and implement on an individual basis! Hence we had to settle on a compromise that optimises against the majority of the population, although I recognise that for some people this might not be the absolute best."
We are considering allowing users to adjust which time periods are included in their analysis but we haven't hit on anything we're happy with yet.
-
Tim Coote commented
I just made almost exactly the same point. For a random error distribution, 3 isn't enough, but I suspect that most billing patterns follow cycles and so are not randomly distributed. Just sampling from part of one year is dangerous.
-
PM commented
I agree. I spend weeks at our house in France over the year. Period of analysis needs to be longer or at least an option for it to be longer