Generic selectors
Exact matches only
Search in title
Search in content

Reply To: Call Forecasting

Vijay Swami

Call volume forecasting methods can be very simple like using a regression method or could be very complex (like using Box Jenkins, or using complex seasonality algorithms etc)
However the very simple set of methods seem to be very powerfully working for me
You may consider doing this simple method
1. Take recent two years worth of data (please make sure that the business context has not changed in that time of data you are selecting. For example if there is a huge system change like a cut over of new technology, having this data will bias heavily)
2. Observe a typical week and find out the day for which the call volumes peak. For example Mondays may be the typical week you will see this high volume
3. After doing that take each month and calculate how many days are there in that particular month. For example 2000-Aug may have 4 sundays, 4 mondays etc
4. Now express the months in terms of the heaviest call volume day. For example 2000-Aug may have call volumes corresponding to 30 typical mondays and so on
5. After you have that data find the factor calls received/Equivalent day factor
6. Once you have those set of values for the two years do a linera Regression to find out the Slope, and the Constant
in the equation Y = mx+C
7. If you slove this equaltion and obtain m and C you can use that for future projections
Amazingly this simple model has worked out very well
But again, this may not work in all the situation

I have been in this situation once
What I did was the following
1. I took the business logic of the product
2. In this case it was an airline company
3. To the call I mapped out the list of all the activities that could possibly lead the generation of the call
4. For eaxmple if its an airline help desk
—Passengers may call
—employees within the comapny may call
–If there is any software in the Airport those technical people may call
like this we listed out all the factors
This may appear simple but we have to carefully think as there is no data
4. Now we attached a probability for each factor. What is the probabilty of having X number of calls in one half hour from a passenger (This will be influenced by having a promotion, Ad time in the TV etc etc)
5. Now you get set of probability with a set of calls associated with each factor categorised into each sub factors
You can do an additive model from that point to estimate the call volume

If any one would like further clarifications (as these are pretty lengthy stuff and I dont know how much I have explained) please feel free to email at