Creating a tolerance range for measuring performance Apologies for my ignorance, I've been trying to find an answer to my question but I think my understanding of statistical analysis is so rudimentary it has been confusing.
I want to use a statistical method to determine what is an acceptable range of performance variance (actual vs forecast) for different products we forecast, modifying the acceptable range for different variables such as time of year, day of the week etc. I know I can use all types of variance modelling to determine what variables contribute to variances in forecast and are therefore significant. Do these models also assist in determining what are acceptable ranges of 'tolerance'? Is that what confidence intervals are? 
Thanks in advance.
 A: Tolerance has many facets so if you need very specific advice you need a much more specific question. However my reading is that you need to get a handle on the general concept in order to fully frame a more specific question. With that in mind I will try and highlight some key things to consider for devising tolerance limits.
Types of Significance
When we obtain a result we need to understand its significance. This has two side; statistical (probability of arising by chance) and practical (is it meaningful). Practical is irrelevant if a result is not statistically significant and the same is true vice versa.
Scientific literature is awash with examples of statistically significant results that are not practically significant. An example. 50% of the population eats food A. 1 per 100,000 get OrganX cancer. A study of 10,000,000 people finds that A increases risk of OrganX cancer by 25%! It is statistically significant! Big headlines, band food A immediately! Launch class action legal proceedings! Risk of OrganX cancer rising from 1 per 100,000 to 1.25 per 100,000 would not (speaking personally here) be not a meaningful factor in deciding whether I as an individual eat A. However, to a clinician running OrganX clinic a 10% reduction in patients (which would require over half the current A eating population gives up A) will be a practical difference. But what impact on the health care system for the thousands out of work because demand for A has dropped off a cliff edge? Practicality depends on who in affected by the result and can extend well beyond immediate consequences
Statistical Significance
Statistical measures are all focused on probability and their aim to to allow us some (not total) objectivity on how likely it is we are fooling ourselves. There are 2 types of errors in analysis - missing a true result and believing a false result. See http://www.statisticssolutions.com/to-err-is-human-what-are-type-i-and-ii-errors/ for more on this.
You need to define the relative impact of each type of error (your comment indicates there is a difference in impact) and adjust your approach to balance the need of each. 
Practical Significance
If, as it sounds you intend to, you want your model to interact with the real world, you also need to determine what is a practically significant result. What size of difference will make a noticeable difference to subsequent events? You need to think about what different events may be triggered and what impact they will have. An isolated 5% error that goes no further is very different to a 5% error that propagates through other events and can escalate into a much bigger error at the end of a process.
Which to use
Ideally the more stringent. However, in real life there may be other factors that demand a compromise. It may not be feasible either in economic terms or amount of time/resource required to meet the ideal threshold. If this is the case you max out your stringency and quantify the risk of doing this compromise.
Ultimately you need to realise that statistics is not about getting the 'right' answer, or 'proving' things 'correct'. It is about quantifying the risk associated with alternatives to allow us to make more informed decisions. You want to minimise the risk of bad decisions as much as actual constraints allow. 
Answering your questions (finally)

Do these models also assist in determining what are acceptable ranges
  of 'tolerance'?

They do assist, but are not the sum total.

Is that what confidence intervals are?

It is good practice to quantify uncertainty in any final outcome, which brings us to the question you asked about confidence intervals. Confidence intervals are an attempt to provide an indication of uncertainty in the final result, but no they do not define what is an 'acceptable' or 'tolerable' result. They instead communicate how much leeway there is in the result to help the reader evaluate the confidence they have in it (hence the name).
