Task Completion Forecasting First off, I am not familiar with forecasting at all so I am kind of lost...
Since I am familiar with Excel, I have been put in charge with forecasting task completion by 30 minute time intervals at work to look for possible support needs.
Currently, the system they have in place is just taking averages of four weeks. Friday's intervals are an average of the last four Fridays. It forecasts most 30 minute intervals as above task completion targets and therefore we are wanting something better.
Edit:
We have a task completion goal of 17 minutes for our employees. We track this in 30 minute intervals. As of now, we are just going off of last four weeks of historical data. Say the last four Fridays from 2p-3p the task completion times averaged 14 minutes from 2-230 and 21 minutes 230-3. We are using that to provide more support at the 230-3 interval today.
Is this the best way of doing something like this? Or, should I be using an exponential smoothing model like holt winters on the historical data?
If I used a different model, what would be the best way to sesetup the historicals?
I'm thinking maybe something like:
11/28/14  2:30  23
12/05/14  2:30  13
12/12/14  2:30  18
12/19/14  2:30  21
12/26/14  2:30  ??
01/02/15  2:30  ??

To forecast completion time for the 2:30 interval on Fridays. Our goAL is to either determine if there's a staffing issue or if there is a support issue.
 A: The problem I believe is that you want to forecast out the expected value for 48 half-hour intervals for some foreseeable period of time. I am currently working with a major fast-food franchise to predict 96 15-minute intervals by day for the next 30 days. Now each day can have it's own trend or expected value and holidays can have an effect along with price promotions and such. The tool for you in my opinion is a mixed-frequency time series model that would incorporate intra-day forecasts and daily total forecasts. Care has to be taken to deal with unusual values, shift/level changes and trend changes and to optimally develop a useful model. Excel won't cut it nor will most other solution providers. Why don't you post 48 historical readings per day for as far back as you have perhaps no more than 3-4 years. If the data is confidential simply scale it .In this way you won'have to assume any model ( as you probably would be very wrong ! ) and let good analysis sort out what is the best way to set expectations and confidence limits.
