Consider Using the 25-50-25 Rule
In their book Statistics for Experimenters: An Introduction to Design, Data Analysis, and Model Building (Wiley, New York, 1978), Box, Hunter, and Hunter recommend that only the middle 50% of the project budget be spent on elaborate designed experiments (such as response surface designs).
The first 25% of the budget should be spent getting more comfortable with the system (perhaps using screening designs to confirm important variables, or using the large variable size simplex to locate the region of the optimum).
The last 25% should be reserved to play "what if" games to take advantage of what has been learned from the previous 75% of the budget!
In our experience, we have found this 25-50-25 rule to be good advice.
Be Reasonable in Setting Factor Levels
In his book Design and Optimization in Organic Synthesis (Elsevier, Amsterdam, 1992), Rolf Carlson states, "I have heard of one experiment in an industrial laboratory where the experimenter had tried in vain to adjust the pH to be exactly as specified by the fourth decimal place in a computer-generated design. This was a rather senseless attempt to achieve unnecessary perfection."
Don't Be Timid About Using Designed Experiments
Perhaps you have been timid about using statistically designed experiments because you think that you need to be a statistical expert to use them. You don't.
Simple designs are often the most effective. Talk with persons who have used planned experiments before, and have them show you how to set up a simple 2x2 factorial design with three center-point replicates. You will be amazed at the quality of information you will obtain!
If You Can, Use Blocked Plans Sequentially
Remember that many classical experimental designs can be broken apart into smaller designs.
For example, a central composite design can be taken apart into two half-fractional factorial designs and a star design. One half-replicate can be carried out first; then (if necessary), the second half-replicate; and finally (if necessary), the star design.
It often happens that the information needed to solve a problem is obtained after only one or two of the subsets of experiments have been carried out. It is sometimes unnecessary to complete the full design.
Pay Attention to Your Measurement Methods
(a) Is your measurement method as precise as it needs to be?
It is discouraging to finish a designed experiment and find out that the results aren't worth much because the measurements are too noisy. Give some thought to this before the experiments are carried out. Is the measurement method sufficiently precise to see what is being looked for?
(b) What is your measurement method really measuring?
In one experimental design project to minimize the amount of contaminant in a product, a chance comment questioned the liquid chromatographic measurement method that was being used. As it turned out, the experimental design wasn't needed once it was discovered that the chromatographic peak of interest contained more than one compound. The process wasn't producing as much of the contaminant as was originally thought!