The simplest—and highest-profit-margin—modeling project that Simitar founder Bob Kotcher has ever done demonstrates the power of simulation modeling for operations improvement.

Celebrate your inner Seinfeld

At a restaurant one day, my inner Seinfeld came out (we all have one—come on).

Since the Americans With Disabilities Act was passed in 1990, many restaurants had to remodel their restrooms to make them wheelchair accessible. This often required them to remove internal partitions that comprised the stalls. With privacy gone, they put locks on the restroom doors and made each restroom single-user.

BUT…many businesses failed to remove the male or female signs from the doors. Furthermore, even some new restaurants, building new single-user restrooms from scratch, put male and female signs on the respective doors. Why not put unisex signs on each door? We now often have the problem of going to such restrooms, seeing a line in front of “ours,” and having to wait (and wait, and wait, and wait…) while the other restroom sits enticingly empty (as our meal buddy out on the table grows increasingly bored).

This calls for an engineering analysis…

Suppose we did a time study on the restrooms (yes, I am an engineer) and used the results to make a static capacity model in a spreadsheet? We’d almost certainly find that each restroom is loaded far below capacity—maybe 75% at peak hours. If we made them unisex, average loading would remain the same. So what’s the problem with them being dedicated by gender?

Well, a computer simulation model would show how, by replacing the two restrooms’ signs with unisex signs, average queue time would decrease.  This is because, due to random variation, there are times when there are women in line when the men’s room is wide open, and vice versa.  Making the restrooms unisex would reduce waiting time in such situations.  Reduced waiting would perhaps result in more satisfied customers and more repeat customers—all for about $20 for new signs.

From restaurant to wafer fab (they both offer chips)

Now, I doubt that restaurants will be conducting simulation analyses to see if spending $20 for new signs would be a profitable investment or not. But one internal client at a wafer fab did something similar.  You see, in wafer fabs and other factories, there are often several identical machines processing in parallel at a particular operation.  If they’re running different recipes, process engineers often like to dedicate each machine to a particular recipe.  This makes process control easier.  But it increases average waiting time due to the above phenomenon.  What is the best tradeoff?

At this client, a process engineer oversaw a wet bench with two parallel identical baths, the only difference being that they were set at different temperatures (dedicated by gender, if you will). Recipe A required temperature A, and all other recipes required temperature B.  The engineer found that he could run the Recipe A wafers at the B temperature if he made Recipe A’s processing time longer.  This would reduce average queue time for all wafers…but increase the processing time for Recipe A wafers.  His question to me was: if I set both baths to the same temperature, will the reduced queue time for all wafers outweigh the increased processing time for Recipe A wafers?

I pretty quickly did a simulation analysis and found that, yes, setting both baths to temperature B and increasing Recipe A’s processing time would actually reduce  average cycle time.

Act on the analysis results, bank the savings

Armed with the analysis results, the engineer made the change. And the result was the biggest return on investment of any simulation project I’d ever done! Not because the savings were so mammoth, but because the cost of the rather simple analysis plus the cost of making the ensuing change were so low. But the change never would have been made had the simulation model not been available to test it out, because nobody in his right mind would approve a change that would increase loading on a machine in order to reduce  cycle time. Until a simulation analysis showed that it did.