Our Blog

“Knowledge is knowing a tomato is a fruit. Wisdom is not putting it in a fruit salad.”


  • Categories

Search Blog

Multivariate testing the McDonald’s menu – an experiment

First, let’s establish some facts:

We humans are by and large, at our cores, creatures of habit.


Our inherent natural tendency to find patterns of behavior with which we are comfortable is one of the founding principles of modern therapeutic methods, and is a key factor in understanding why people do the things they do.

As a child, for many of us, it was a real treat to be taken to McDonald’s by our parents. This experience is by no means unique. In fact, It is quite common.

Our McDonald’s cravings, although they generally become rarer as we get older, often stick with us all the way from childhood into adult life.

Many still think it is delicious.


Somewhere along the way, for the folks here at Phrasee, our brains narrowed down our McDonald’s ordering options to two; Big Mac Meal or McChicken meal.

big macmcchicken

Burger, fries, coke.

Beef or chicken.

Image credit: Cartoon Network

It’s not rocket science, folks.

A problem arises…

Then, one day, a strange thing happened.

We arrived at McDonald’s, as we have so many times before, and reached our hands into our pockets.

We had what divided up to £4 a piece.

The Big Mac meal and the McChicken meal had suddenly become unattainable pipe dreams out of the reach of our financial reality.

We were hungry, so our instincts – deep, ancient instincts honed over millennia of our collective history as hunter-gatherers – kicked in.

Image Credit: SNL

A backup plan slowly took shape as our primitive brains perused their options.

We ordered McDoubles and chocolate sundaes.

They were amazing. They were satisfying. They were delicious.

It made us question everything we thought we knew.

Well, maybe not everything. Maybe just our McDonald’s ordering habits. But still…

Had we been going about things at the McDonald’s counter the wrong way all along?


Here at Phrasee, we have recently been experimenting with multivariate testing for email subject line optimization to increase click rates and conversions, which yielded extremely positive results.

Could the same principles be applied to optimize our McDonalds meal ordering habits to increase yum rates and deliciousness?

It was time for some multivariate testing

Image credit: Alaska Airlines

Fortunately, the McDonald’s menu is quite limited, with 19 sandwiches/wraps, 5 side options, 5 dessert options, and 18 beverage options; or 8,550 different meal combinations.

Several items from each category could be eliminated right off the bat, since we already knew we didn’t like them.

This left us with 2,250 possible meal combinations, were we to order 1 item from each category during each visit.

For the sake of simplicity, we eliminated beverages from consideration, and assumed that we would have water with each meal (since we do not believe that the beverage improves our overall enjoyment of a meal at McDonald’s) which cut the total number down to 225 possible variations, minus the 2 control meals.

The 2 control meals were our old standbys, the Big Mac meal and the McChicken meal.

That’s 223 alternative variation meals which must be tried and measured against the controls, to gather the data we need.

Data mining never tasted so good.


If we visit McDonalds once a week, trying a different variation each visit, this still means that the experiment would take us four years and just over three months to complete.

That’s some big data, even for greedy folks such as ourselves.

And all this data could potentially be skewed by several other variables:

  • How hungry are we on a particular day?
  • What else have we had to eat recently?
  • How have our personal tastes changed over the 4 ¼ years of the experiment?
  • What if McDonalds added / removed menu items during the experiment?

It was all getting pretty complicated.

Things were so much easier back at Phrasee headquarters, where high-tech learning machines could comb through the data in mere moments, accounting for all the variables, spitting out potentially delicious multiple variations, and sending them to hundreds of people to see which ones people liked…

Perhaps our colleagues would share our curiosity, and put Phrasee’s considerable resources behind this McDonald’s multivariate testing experiment?


Phrasee’s learning machines have bigger fish to fry. Our CEO, Parry Malm, made it crystal clear that we were on our own with this one.

So, this experiment will be an ongoing one.

However, we have managed to glean some significant learnings from the data already.

Image credit: Cartoon Network

Learning #1:

The beverages are a waste of time. Water does the job just fine, saves a few calories and saves a few pence.

Learning #2:

Stick with the fries. The salads are okay on their own, but, unless you are on a strict diet, fries are WAY better

Learning #3:

Working a dessert into the meal really spruces things up. The desserts are cheap, and they definitely add something. If one foregoes fries, the desserts are a great alternative. Salty and sweet together are boss.

Learning #4:

Stick with the classics. Most of the seasonal menu items are a waste of time, and don’t hit the comfort food sweet spot that a Big Mac, a Quarter Pounder, or a McChicken do. If you are feeling sassy, and want to try something entirely new, eat somewhere else.

There you go. Hopefully we have saved you some trouble.

We’ll keep you posted as we delve further into this experiment, so stay tuned.

Our goal is to maximize meal deliciousness and yum rates, and we won’t stop until we achieve it. The perfect meal, much like the perfect email subject line, is out there, and we WILL find it, because that is who we are, and this is what we do.

Trying to get ahead of the competition?

Check out our Learning Hub for everything you’ll need to know!