Monday, August 29, 2011

Does extrapolation in estimation work?

How do you estimate test automation efforts for building a regression suite for functionality that is already in Production?

The approach I am following is:
> Get some level of understanding about the domain
> Look at a subset of existing test cases of varying complexities and sizes
> Estimate the effort for each of them and also identify the complexity involved in each, with the risks and assumptions.
> Identify some obvious spikes / tech tasks that will be needed
> Extrapolate the estimates
> Add 10-15% buffer in the estimates for unknowns.

And presto!! .... I have my estimates for the task on hand.

Is this approach right? 
Is extrapolating the estimates going to give valid results?
What can be different / better ways of getting the results required?