Sometimes you can't forecast worth a darn because something is just not forecastable.

Being "unforecastable" doesn't mean you can't create a forecast, because you can always create a forecast.  It just means there is so much instability or randomness in your demand patterns that even sophisticated forecasting methods don't help you achieve the level of accuracy desired.

Also, being unforecastable doesn't mean you can't do anything about it:

• You can shape demand into more forecastable patterns, by pricing or promotional activities (or simply stop doing those activities that are causing demand volatility).
• You can manage operations to account for wildly inaccurate forecasts, by carrying extra inventory or flexible production capacity.
• You can protect yourself against the unforecastable by carrying insurance (because we can't predict when we'll crash our car (Lindsay Lohan, Rowan Atkinson), or our house will burn down (Jack Nicholson, Richard Branson), or we'll have our house burned down by a disgruntled girlfriend (Andre "Bad Moon" Rison and Lisa "Left Eye" Lopes)).

Management may be unwilling to acknowledge that we can't forecast something with any reasonable level of accuracy.  (Often: New product forecasts.) Unrealistic accuracy expectations can lead to overconfidence in our forecasting ability and bad business decisions. (Per the Solyndra press release announcing suspension of operations, "...uncertainties in recent months created significant near-term excess supply and price erosion...This was an unexpected outcome and is most unfortunate.")

A forecast is simply a best guess at what is going to happen in the future. Your forecasting software may provide some expression of confidence, or a likely range of values in addition to the point estimate.  The organization should have an appreciation of the uncertainty inherent in any forecast. Knowing that demand is likely to be 100 +/- 10 units instead of 100 +/- 100 units can lead to very different plans of action.

Unrealistic accuracy expectations can also lead to inappropriate forecasting performance targets.  Unachievable targets, such as correctly calling the toss of a fair coin 60% of the time, can discourage forecasters, who may wonder why they bother trying when the goal is unreachable.  Even worse, an unreachable goal may encourage forecasters to manipulate the metrics to achieve the necessary results (or in layman's terms, to cheat).

See the 2009 BFD post "A New Favorite Forecasting Article" for discussion of a great piece by Makridakis and Taleb on "Living in a world of low levels of predictability."