Agile estimations introduce a new method for forecasting the delivery of work. It can take some influence to win over managers into accepting relative estimates rather than time-specific estimates. Building more accurate estimates both builds confidence and predictability in delivery.
Distinguishing the difference between variation, as well as understanding its causes and predicting behaviour, is key to management’s ability to properly remove problems or barriers.  Before we attempt to establish accurate estimations, we should first value the importance of variation in work systems.
A process that is in statistical control, stable, funishes a rational basis for prediction.  Advantages of processes that are in statistical control (stable) include:
Building a baseline consists of various elements.
Measure and Capture Estimates – Teams should capture their estimates for work prior to commencing work on Product Backlog Items. Estimates may be captured prior or during Sprint Planning.
Measure Delivery Against Planned Work – At the end of the Sprint the team should review the completed work against the planned work. In most cases there will be some variance between planned and actual work. Metrics can be captured each Sprint to build a profile of velocity and throughput, as well as reliability of estimates.
Analyse Delivery – Teams should review delivery, such as in a Sprint Retrospective, to understand the root causes of any factors that resulted in any significant differences between estimate and actual delivery. Both effort and duration should be considered to better understand team velocity and throughput. Additional considerations such as staff unavailability should be taken into account.
Planning Poker is an activity for estimating work items, typically during Sprint Planning. The team selects a work item and each individual team member makes their own estimation. Estimations are then shared. Where there are different estimates, team members discuss their assumptions about the work. This helps to better understand factors in delivering the work. The team may then repeat the individual estimations until there is an agreement for the estimate.
There is no accepted standard of statistical significance for past data sets become reliable for estimating team velocity. A common standard is six to eight Sprints.  However, this is assuming that both the Sprint velocity and individual work item estimations become more accurate over time.
Even with a stable team with a strong track record of estimations, there will always be variance if work is not uniform (e.g. manufacturing physical products).