The Troublesome Yet Terrific Tortilla (Part 3)
In part 3 of our look at the troublesome yet terrific tortilla, following as it does from a general description of the tortilla quandary in part 1, and some specific details regarding the tortilla challenges anticipated in part 2, finally I’ll begin to explain how I approached the task of optimizing the formulation, or in simple terms, how I made the very best tortilla.
What Is “Best?”
As mentioned previously, I’ve already defined that I am speaking of wheat-flour-based tortillas; so my goal in optimizing a tortilla formulation is to produce the wheat-flour-based tortilla with the best organoleptic characteristics. It’s worth taking a moment to consider what I mean by “best.” In this case, I am considering only my own preferences, whether I’m talking about aroma, overall flavor, mouthfeel, or even appearance. However, if I were developing this tortilla for a regional restaurant chain, it would make sense to do consumer evaluation that includes a sample of the regular customers of that chain, and perhaps also consumers that the chain is trying to convert to customers. My specific preferences, for example, may or may not overlap well with those of the restaurant customers. But luckily, since the tortillas in question are for me, I can more simply consider my own preferences.
All Wheat-Flour Tortillas Are Not Created Equal.
Also, it is important for me to clarify that I want a tortilla that tastes great when fresh. I’m not at all concerned about storing the finished product on the shelf, or in the fridge or freezer. I want a tortilla that I can griddle now, eat, enjoy thoroughly, and then move on. If I were developing a tortilla for a grocery chain, I would have to reconsider this approach, because in that case the end consumer has little chance of buying the tortillas fresh and hot. They will need to be reheated out of necessity, and they will likely need to be able to remain free of mold for at least a week on the shelves, and perhaps longer. Such a tortilla would have characteristics that are dramatically different from the type I want to eat for my own enjoyment.
Are there Additional Constraints?
There are plenty of other details to consider. Do I want to work with ingredients available at the local grocery? Am I interested milling grain into flour? Does the tortilla need to be vegetarian? What types of equipment are available for processing and cooking the tortillas? And speaking of production constraints, while I’m willing to deal with dough that is a bit sticky or stiff and difficult to roll out in order to achieve a better tortilla, there is a limit to my patience. I don’t want to make my life 50% more difficult just to squeak in an extra 1% increase in improved mouthfeel. So, as I’m rating the various sensory characteristics of each tortilla, I can also consider rating things like dough stickiness, dough toughness, and perhaps overall workability.
Pulling Everything Together.
I’ve said that I want to optimize the formulation for a wheat-flour-based tortilla that is best according to my own preferences. I want to be able to make it at home with a 4’x4′ maple butcher block serving as the work surface, a wooden rolling pin, a 6-qt stand mixer, an induction range, a cast-iron skillet, and a spatula. I would also generally like to use ingredients that are readily available (e.g., King Arthur AP wheat flour, sea salt, and filtered tap water), but I’m willing to put in a little extra work to improve the quality of the added fat.
Ingredients.
For the fat in the tortillas, I’ll render my own lard from the backfat of locally raised hogs. As part of the rendering process, I’ll use a temperature high enough to brown the cracklings once they are void of fat, which adds additional roasted complexity to the lard’s flavor profile. At the end of the rendering process, I’ll also add some smoked bacon to give the lard a subtle smoky note. I will not be using plant-based oils in this recipe, as I know from experience that they have bland or off flavors in tortillas. I will also not be using baking powder. As mentioned in part 2, baking powder is quite a common ingredient in modern tortillas, but I would prefer to depend upon steam–from the added water–to leaven and lighten the texture of the tortillas. In my experience, the mouthfeel of the end product is better than when baking powder is used. So now I’ve clearly defined the constraints for this project. What’s next?

A Product Optimization Experiment.
We now have to take our known constraints and design an experiment that will allow us to optimize the tortilla formulation. How should we proceed? One variable at a time? Often seen as the acronym OVAT, this is thought by most non-scientists to describe how experiments are run according to the scientific method. Actually, even some scientists believe this. But here’s the thing, it’s dead wrong. Even in a relatively simple food like our tortillas with only four ingredients (i.e., flour, water, lard, salt), changing the amount of one variable at a time, let’s say the flour, while holding all else constant, is likely to be, at best, an incredibly inefficient way to determine what the optimal tortilla recipe is, and at worst, will never find it at all. In short, OVAT has significant disadvantages. Arguably, the most important disadvantage is that important “interactions” usually exist in foods, and will be missed with the OVAT approach.
Interactions!
In this project, if we simplify for a moment to say that our only goal is to improve the flavor of the tortilla on a scale from 1-10, then any positive main effect would be one related to any of the individual ingredients that increases quality. For example, maybe more lard increases quality on average. That’s lard’s main effect. But what happens when water and lard are both present? Does water, in addition to having its own main effect on quality also impact the degree to which lard has a positive impact? If so, then that would be an interaction effect. It might be the case that with smaller amounts of water, lard is hugely positive in its impact, but with more water it makes a much smaller positive difference. It is immediately clear why interactions can cause huge problems with the OVAT approach; because if you only change one variable/ingredient at a time, you can’t determine if interactions exist. And guess what? Interactions almost always exist to some degree, and sometimes they are hugely influential. So what can we do about this?
Optimal Design of Experiments.
When the OVAT approach can’t be trusted to work, which is usually, instead we can rely on the considerable theory in the field of statistics that supports what is called “design of experiments,” along with modern computers and software, to produce an optimal experimental design. Such a design is able to help determine what the optimal formulation would be of a tortilla, again from the flavor quality standpoint defined above, but more than that, such a design even allows the experienced product development expert to determine the optimal formulation while considering several characteristics at the same time. This is sometimes called multiple objective optimization. How it does this is easier to show with an example than to explain in prose, so we’ll take a look at a tortilla-development optimal design in the final part of this series (part 4).