Is time-series decomposition always the most accurate forecasting method to use when you are introducing a new product?
In general, time-series decomposition is a good forecasting method if you are trying to track some sort of ongoing trend. You take all the past measurements as a time-series, and break them down into a trend, seasonal variations, cycles, and random fluctuations. Usually it's the trend you're interested in, but sometimes the seasonal effects and cycles can be useful as well. Random fluctuations are essentially noise that you want to remove as much as possible.
The problem with this is that it assumes that the process you're dealing with is in some sense unchanging; it's not literally constant (or your time-series would be trivial) and may not even be stationary in the statistical sense (a time-series is stationary if it does not have any unit roots), but it's in some sense "the same process". You are using past outcomes to predict future outcomes, so you are in effect assuming that the future will be like the past.
But if you introduce a new product---especially if it is very new, something quite different from what your company has produced in the past---you really can't assume that the future will be like the past.
There are some other ways you might be able to forecast that would be superior to a simple time-series analysis.
One would be to analyze your competitors: What happened when they introduced new products? Another would be to do market research on that particular product, using surveys or focus groups: How much do people really seem to want this new product? You might still want to do some time-series forecasting, if you have introduced other new products in the past and you think this launch might turn out similarly.
But ultimately, there is some inherent uncertainty that you may not be able to remove at all; introducing new products always carries risk. Most new products do not succeed---but the few that do are often profitable enough to make up the difference.