What is a best way to forecast sales based on limited period of historical data? Let's say we have data of company sales for 4 months in the format like:

How to set a minimum plan for each salesperson based on its performance for 4 months? In my understanding, it is not possible to use time series model, due to the lack of data for the rest months of the year, we will not be able to take into account the seasonal effect. So what is the best way to do it? I would simply go with the mean for 4 months. For 1st row, for example, it would be 299.00
Thanks!
 A: Let's begin with some observations about your data and your goal.  To start with, we should begin with management's responsibility.  Research into organizational behavior would assign over 90% of the variance in employee behavior to management decisions, so the first thing to do is make sure your decision-making process reflects the proper attribution of management and employee behavior.
Since this data is being driven by the existing policy rules, the question should be what adding a new policy rule such as a minimum sales floor be on the outcomes?
The existing policy rules, the limited data, and the need to properly attribute the source of variance to its proper location limits your choices.  If you decided to do a time series for each actor you would almost exhaust your degrees of freedom and it isn't clear it would be meaningful.  What if the store sold spring and summer camping supplies and nearly nothing in the winter.
Because you only have three salespeople, looking at the horizontal average isn't meaningful because the variance is so high.  Likewise, the vertical averages don't tell you much either, again due to the variances.
There are a couple of things you could do, though.
For starters, if you would presume a normal distribution or would do a quick check with a kernel density estimate to see if it could be, then note that the grand sample mean is 313.1 and your sample standard deviation is 28.2.  A quick and dirty estimator is $\text{mean}\pm{3}\text{ standard deviations}$.
All of your observations are clearly within that range which is $(228.4,397.8)$.  The observed range is $(270.2,360.1)$.  This implies that the policies are driving your variability, based on research in organizational behavior.  None of these transactions should be attributable to differences in skill.  It isn't that skills don't matter, but they are interacting with the policies. With different policies in place or difference chance factors, the rankings may change.
You could look at Student's t-table and set the minimum at some level which would be an extreme event.  This begs the question of the appropriateness of normality, but a simple kernel estimate shows it isn't a bad guess.  You could perform ANOVA to further verify that no column or row is "special."
A less sophisticated and even more quick and dirty, but less distribution dependent solution would be to consider the "ranks" involved.  The range is 89.91.  If you would divide it by 12 your units are about 7.5.  If then remove 7.5 from the bottom, then you are somewhere above the 7th percentile, or 1/14th of 100%.
Finally, you could note that the monthly variance is comparatively very small in June and it is the lowest observed month.  If you were near some "base" demand, then you would expect the variability of the separate actors to contract.  This is a dangerous assumption if there is no outside information.
Looking for a base is risky with little data.  The reason is that people respond to the incentives in policies.  If it was a budgeting lower bounds rather than an employee planning lower bound, then it would be a different question altogether.  As a policy matter, it would be considered unwise particularly in light of Demming's 14 points and the related research.
Still, if you must, then you should be prejudiced against management and in favor of the sales staff.  Instead, if the observed lower bound of sales is unacceptable, then sales training, commissions and work environment should be looked at.
Given no other data, the sample mean should be your projected next month's value and the sample mean minus three or four standard deviations, due to the small sample size, should be your lower boundary. 
Since the type and source of randomness aren't known or understood, you shouldn't be holding employees responsible for the things you do not understand.  The research on "superstitious learning" shows that imposing a bad goal or bad incentives could have adverse long-run effects on sales.
A: With your data ... the mean would be the best model. However if one of the series  was 280,290,300,310 .. my "best forecast would be 320 ... Additionally if the data was 100,80,100,80 , I would want the software to predict 100 . So it all depends on the ratio of signal to noise ! The stronger the signal the less data is needed to empirically identify the model and vice-versa . Time series methods are not limited to a certain number of data points just some time series algorithm's are . 
Finally consider the series 120,90,90,90 ... I would want the forecast to be 90 essentially disregarding the errant 120. Similarly 90.100.90.100.90,100,50,100 .. I would expect 90 as the next value and seriously challenge/authenticate the "50".
