How much detail is necessary to deliver accurate forecasts?

Have you examined the reasons for your variance?

It is possible that the algorithm and assumptions you are using as the basis of your forecasts are flawed, so whether you ditch detail or not, you are not correcting the fundamentals.

In general, I believe the amount of effort or modeling put into an area of a forecast should be directly related to the materiality of that area. So if personnel costs are material, then it makes sense to model them at a detailed level, such as headcount, to gain accuracy in the overall forecast.

If you are not getting the accuracy you want in the personnel area, then there could be issues in your modeling. Personnel should be able to modeled quite accurately as the related costs (taxes, benefits, etc) are very calculable. Variances that come in the size and timing of items such as raises and bonuses are fairly easy to track. Also, if your forecasting tool is a good one and is setup properly, the maintenance to the Personnel data that drives the forecast in this area can be managed with proper procedures.

Obviously at some point this theory can break down if the volume of information gets too large. With 1000 employees, if one person is maintaining the data, it should be manageable. But if you have a distributed model and you have 10 departments who will each manage their 100 employees, it is very manageable. There is also the factor of the source data to feed the forecast with accurate personnel data. If you set up an interface with your HR system, things get much easier. So the cutoff as to when to switch from Headcount to something less granular is going to depend on many factors that will be unique for each organization.