Professional knowledge about the problem itself proves most useful in writing a search algorithm when the standard procedures prove insufficient. In real-world problems (unlike purely theoretical problems) the hardest parts are actually often confined to one sole part of the model. In consequence, you have to start searching for a partial solution in that region in order to minimize the exponential factor.

For example, when you're building a schedule, you have to assign the most difficult or the most critical items first. The easier parts fall into place around them.

This kind of solution clearly shows that the procedure for solving a problem depends very much on the instance of the problem under consideration, not merely on its formal statement.