Up until now, we have had very little discussion of the costs of certain features that customers may wish to include in the system requirements. Part of the reason is that it made sense to separate that very complex issue from all of the other problems surrounding requirements elicitation, analysis, representation, validation, verification, and so on. Another reason for avoiding the issue is that it is a tricky one.
In the early 1980s, Barry Boehm proposed a model for software projects called the funnel curve that helped explain why software effort and cost prediction was so hard (Bohem 1981). The model had been around for some time, but Boehm first proposed its applicability to software (and later systems) projects. Steve McConnell refined the idea, calling the curve The Cone of Uncertainty (McConnell 1997). 
The model is very much relevant today, and it applies both to software-only systems and complex systems that may have little or no software in them. More significantly, the model helps explain why early requirements understanding is so important, which is why we recast the model as the Requirements Cone of Uncertainty (Figure 10.1). It’s easiest to understand the model by reading it from right to left.
Because requirements change, it is not until the project is delivered that the final list of requirements is known with certainty. Up until that time, the requirements can change. In fact, early in the life cycle of the project, at requirements definition time in particular, the requirements are highly volatile, that is, uncertain. We show the estimation variability of the requirements at the projects initial conception on a scale of 4 to 1, that is, at this point in time, our estimates of time to complete, and hence, cost of requirements could be as much as four times too conservative to four times too optimistic. Of course, this factor of four is entirely arbitrary.

The purpose of the cone model is to emphasize that it is only over time and experience that we gain precision in our ability to estimate the cost of the true requirements of the project, and so the variability of our estimates converge to their true values over time.