The icing on the cake: baked goods, AI, and lessons in user adoption
What the evolution of cake mix could tell us about future trends in AI.
What do cake and AI have to do with each other?
Comparing the adoption of cake with AI might seem pointless, but a closer look reveals parallels that could predict future trends in user adoption for these emerging technologies. No one’s life is independent of food choices. Looking at what’s on the plate is an examination of culture, economics, class, politics, gender, and social life. Similarly, our relationship to technology is largely impacted by the same forces.
The story of boxed cake mix is one about marketing ingenuity but also shifting cultural values. At one point in time, this household staple was a technological advancement predicted to reinvent baking in the American kitchen. Seeking to innovate American cookery, cake mix served as a litmus test for the food industry to further ingratiate consumers to a wide variety of packaged food products promising convenience and ease. While the convenience of ready-made biscuits and pancake mixes were welcomed by consumers, cake mix was a different story. To reinvigorate sales, the food industry had to tackle the emotional and psychological ties consumers had to a product that had long held sentimental and symbolic meaning. That is, they had to redefine what it meant to be “homemade”.
The breadth and depth of AI impact is still and will continue to be defined for some time as we engage more meaningfully with these technologies. While AI has been integrated into our lives for some years, recent innovations in generative AI are forcing broader discussions and challenging our values around work, art, entrepreneurship, etc. If cake mix could alter our perception of “homemade” cookery, what other cultural and social norms can we expect AI to transform? Will we cry to AI-generated songs after a break-up? Will we do our annual checkups with AI medical assistants? Will we use AI to fight municipal tickets in traffic court?
It took cake mix decades to find it’s place in the American kitchen, but the current AI race indicates that we will not have the same amount of time to warm up to these technologies. However, the lesson of cake mix shows us that ease and convenience won’t be enough to drive this adoption. Regardless of how someone is introduced to this technology, there are psychological, social and cultural factors that will determine how they engage with it moving forward. There must be trust that the technology is reliable, accurate, secure and safe. There must be transparency around the recipe and how decisions are made. There must be personalization that addresses the unique accessibility barriers and challenges of users. More importantly, there must be the “je ne sais quoi” of human-like behavior that feels both comfortable and familiar to the user.
Nothing Says Lovin’ Like Something From the Oven
Cake has never been just a dessert. As Ms. Betty Crocker herself put it, “a gift you bake is a gift from the heart”. It’s no wonder that many of us turned to the simplicity of baking during the pandemic or tuned into watch other home bakers do their best on The Great British Bakeoff. But for most of its history, baking a cake has been incredibly laborious work that often took more than a day.
In the 18th century, home bakers would have had to proof their yeast, knead and hand-stretch dough, leaving it to rise for a further 24 hours to create the once popular celebration cake of yeast-bread laden with fruit. In the 1800s, sponge and pound cakes became popular but these still required incredible stamina to hand-whip the eggs. Cake changed shape again with the revolution of baking powder as bakers could create light, airy layers with fewer eggs and less kitchen labor.
By the 1920s flour consumption had taken a major hit as growing public interest in nutrition led people to eat less bread and potatoes. Gone were the days of hundred-pound sacks of flour and weekly bread-baking. The food industry responded to this decreased demand by letting go of bread and putting their hope in the other baked goods — biscuits, muffins, cookies, and cake.
“They weren’t in the flour-selling business anymore; now they were selling convenience.” — Michael Y Park
If you’ve ever baked a cake, you know that a lot can go wrong— the weather, age of ingredients, size of the eggs, oven temperature, your mood, the cosmic alignment…Conjuring Mary Berry, creating the “layers” of a well-risen cake demonstrate one’s cooking prowess and ability to triumph over these treacherous obstacles. Check the cooking section in any 1950s newspaper or magazine and there will be lengthy discussions troubleshooting what went wrong and recommending foolproof recipes for successful cakes. With high demand and a clear consumer need to be filled, it seemed so obvious that cake mix would fly off the shelves. Who wouldn’t want the assured victory of a perfectly baked layer cake for that special life occasion without the labor of hand-whipping eggs or uncertainty of home cooking?
By the 1940s, there were more than 200 companies producing boxed cake mixes. The notable leaders were Betty Crocker and Pillsbury. There were two main types of mixes — the “just add water” and the “add your own eggs”. General Mills and Duncan Hines took the egg road while Pillsbury stubbornly stuck to the water only road. In her book, Something From the Over: Reinventing Dinner in the 1950s, food historian, Laura Shapiro quotes one Pillsbury executive:
These were the days when cake mixes were miracles; when using them was like having the essence of the modern world in your own kitchen.
If the 1940s and early 50s were the heyday for boxed cake mix, the music had died by the end of the decade when sales stagnated. There was enthusiastic support from farm wives, the early-adopters, for whom cake mix offered more frequent access to this baked good. However, surveys at the time revealed that homemakers would keep the mix in their pantries but always made the distinction between a box mix cake and one homemade. For them, the homemade cake was still the “real cake”. The food industry was at an impasse. How could they boost sales when they didn’t understand the degree to which cake mixes either supplemented or replaced traditional baking?
Trying to understand why women would balk at a product that could provide so much convenience, General Mills tapped consumer expert Ernest Dichter. He theorized the reason women didn’t respond to cake mix came down to eggs. That’s right, eggs. The “egg theory” posited that women didn’t find satisfaction in baking with box mixes because it wasn’t as difficult and therefore, not as fulfilling. It felt like cheating. By removing the powder eggs in the “just add water” complete mixes, women could add the eggs themselves and still feel like they were baking. Manufacturers jumped to reformulate the mixes, women whisked in some eggs and cake mixes were saved!
The egg theory has become one of the great consumer urban legends. We know that manufacturers were already offering “just add eggs” mixes long Dichter suggested it. Powdered eggs in the complete mixes resulted in inferior cakes. Adding the fresh egg improved the taste as well as the texture. So really this was the industry moving towards homogeneous quality standards. But with the egg theory, Dichter seemed to touch on a moral nerve. Women did frequently discuss feelings of guilt for using complete mixes because they felt they were taking shortcuts. However, this speaks more to their perceptions of their role in the home and the social imperative to bake from scratch, then some psychological relationship to eggs.
Advertising followed Dichter’s lead and began emphasizing that these mixes were merely another step in the baking process. It wasn’t the egg. It was the icing on the cake that saved the cake mix! Imagery in advertising became oriented around the lavishly decorated layer cake. Shapiro notes that:
“It was the willingness to focus on results — “at least it would come out being a cake” — was precisely the attitude the food industry hoped to foster among home cooks and was a harbinger of many such victories in the realm of gustatory revisionism.” — Something From the Oven, p.79
The difference between a box mix and homemade cake was more apparent to our grandparents than to us. As our taste memory changed, flavor and texture were no longer the obstacles when the cake was covered under frosting and adorned with fruits or nuts. Although cake mix has never fully lost its social stigma, it has continued to be adopted and adapted into homebaking and even had somewhat of a renaissance in the past few years. The personal touch of decorating the cake or incorporating the mix into a new recipe would transform a box mix into a individual “gift from the heart”. Regardless of the labor involved, it was all home cooking.
Adding the Egg or Icing the Cake?
Cake mix was a technological advancement that blew people’s minds in the same way that we’re being amazed by AI solutions today. Bursting onto the market, nearly 200 manufacturers raced to meet initial demand for cake mixes. But while manufacturers focused on convenience battling between complete mix and ‘adding the egg’ recipes, consumers lost interest and the stigma around the product, rooted in social norms, won out. A real cake was homemade while a box mix was somehow cheating.
Other packaged food products like tinned vegetables, biscuit and pancake mixes or frozen orange juice, didn’t have the same bumpy adoption curve that cake mix followed. That’s because, again, cake has never been just a dessert. The food industry made the same mistake that many tech companies continue to make — treating complex products like simple, convenient solutions. Using tinned vegetables or frozen orange juice didn’t have the same social or cultural norms of replacing homemade cake with a packaged product. It wasn’t until consumers changed their perspective around personalizing the cake mix through icing, and clever advertising, that things changed.
The right adoption approach should be chosen within the context of the application, user beliefs and behaviors, and complexity of the solution. There certainly will not be a one-size-fits-all approach for AI adoption because the breadth and depth of application will vary so much. What worked for tinned tomatoes won’t work for cake mix. Similarly, what works for a simple AI app won’t work for more complex use cases. In this, we’re all agreed. However, what that looks like in application is often missing from the conversation. The secret to AI adoption continues to be…user experience.
Examples of Technology Adoption Approaches for AI
In the late 1980s, research began focusing exclusively on user adoption of technology as people began interacting more with computers, productivity applications and personal software. I wrote a brief history of UI/UX here if you’d like to learn more. Many theories emerged to help model and explain adoption of technology, but let’s focus on mainly on TAM and TPB for this comparison.
Technological Acceptance Model (TAM)
The Technological Acceptance Model (TAM) is a theory to explain why people choose to use a particular technology in a (work) context. When determining whether to adopt a new technology, people consider how useful it will be to them and how easy it is to use. However, TAM says very little about the technology itself — interface, user flow, processing speed, functionality , etc.— but more about what we believe or perceive the technology to do. So, whether technology is actually easy to use or useful is not a matter of how it’s built but rather what we think it will do for us. That’s not because the technology is different but because people are different.
TAM assumes that people plan their behavior and are rational when determining the usefulness of technology and their intention to use it. This is one limitation because people are obviously not rational. Think about the people who camped out in front of Apple stores when the iPhone came out but had never actually used one before. Similarly, people signed up for the Bard waitlist without every having tried it. Though camping in front of a store is notably more difficult than one-click sign-up.
TAM also doesn’t give us any design advice to improve or build better tech. The primary purpose was to give practitioners guidance about measures to take before implementing a system not after. Other alternate methods emerged to fill in these gaps between perceptions, behavior and action to understand how we form expectations and how those expectations differ after interacting with tech.
Theory of Planned Behavior (TPB)
Theory of Planned Behavior (TPB) has long been one of the best predictors of behavior used in marketing. TPB seeks to explain why we do what we do through three main components:
1. Attitude Toward Behavior:
You believe a certain behavior or activity will make a positive contribution to your life. For example, you think using AI-generated text would save you hours drafting copy for your website.
2. Subjective Norm:
You are influenced by everything around you — sociocultural norms, social network, professional status, etc. You worry that your subscribers won’t read words not written in your own voice.
3. Perceived Behavioral Change:
You believe a certain behavior is easy or difficulty to do in a specific situation. You’re able to easily prompt and receive the generated text you need for your website.
You need all three elements to actualize an intention to use a technology. In fact, if only 1 or 2 of these components are present then the likelihood of your acting on your intention to use that technology decreases. So if you’d like to use generated text for your website because it saves you time and you feel you can do that, you might still not do it if you think people will judge you for it. But say you write for AI-enthusiasts, your anxiety about subjective norms could be false because they might have a higher tolerance than say readers looking for advice on creative writing.
While both TAM and TPB consider attitudinal factors as key predictors towards intention and therefore behavior, TAM focuses more on the specific context of technology adoption and TPB can be applied more generally to a wide range of behaviors. When to use one model over the other largely comes down to the research question asked, the tech, and the context.
When may TAM be more appropriate?
- The focus is on understanding the factors that influence technology adoption rather than broader social norms.
- The technology in question is perceived to be straightforward and relatively easy to use.
- The users are not as influenced by external or social factors, e.g. individuals versus organizations
When is it better to use TPB?
- The focus is broader behavioral intentions rather than just adoption of one specific tech.
- The tech is complex or specialized and requires specific skills or knowledge to use.
- The users are more likely to be influenced by external or social factors, organizations or groups with shared belief systems.
So you might use TAM for an AI solution that generates memes from text descriptions but you’d need deeper understanding of user behavior for an solution that triages whether or not you should see a GP.
If you have a personal health tracking app, you might use TAM if the functionality is straightforward. Think of how simple Fitbits from a few years ago were. They basically tracked your steps and heartrate. But AI health companion Ada, which learns your symptoms and guides you to take health actions, would require more robust adoption support through TPB because you need to understand social and external forces. For personal shopper chatbots, the public didn’t need that much convincing to use product recommendations due to seamless integration. The recommendations were already positioned as socially accepted for people also interested in the things you were interested in. However, taking product recommendations to a personalized styling experience requires more integration and therefore, deeper understanding of shoppers’ attitudes. Finally, in industries that have been slower to adopt data-driven technologies, the ease of adoption might mirror the amount of decision-making control and the user feels they still retain and trust around the underlying predictions. Bottom line, different contexts create different user experiences that should be taken into account when driving adoption.
While both models can provide valuable insights into factors that influence technology adoption and use, the tradeoff between these two is that TAM is more focused on the perceived usefulness and cannot provide insight into design. On the other hand, TPB can focus on broader behavioral intentions and social norms. However, if the focus is on specific factors that may influence the adoption and use of AI technologies, such as privacy concerns or trust in AI systems, then alternative methods that address the limitations of TAM and TPB should be applied.
As the tech landscape changes from seamless integration to direct contact with the underlying technology/algorithm, more emphasis will be put on user perceptions, behaviors, and social norms. Just as cake mix didn’t completely upheave baking, AI will augment not replace human behavior. There will always be a special place for “baked from scratch” cakes but our acceptance of what something means to be “homemade” has certainly expanded. The adoption of cake mix allowed many households to enjoy a special treat that would have otherwise been a rare occasion. Similarly, AI will expand access to services and solutions that were previously too expensive or too rare to use. We could very well see AI lawyers providing services to those for whom legal services remain unaffordable.
How we use AI and the stigma around the work generated from it will change as our values change through continued engagement with it. If I always had the time and energy to bake homemade cupcakes for my child’s class I would, and yet no one complains when I show up with some from a mix. A break-up song written after a messy divorce will hold more truth in it than a probabilistic prediction of generalized heartache, but that doesn’t mean you won’t still cry to it. The way your pediatrician recounts your child’s development will be a different beside manner experience than the AI medical assistant, but you might not want to pay the insurance bill for the doctor if all they need a check-up. There’s a personalization and human touch that won’t be easily replaced. The reality is that we’re all balancing constraints, and AI will offer solutions that reduce risks, improve efficiency, and democratize services. But the impact potential will all come down to choosing the right approach for the right context.
Sources
Park, Michael Y (2013) A History of the Cake Mix, the Invention That Redefined ‘Baking’ | Bon Appétit (bonappetit.com)
Shapiro, Laura (2004). Something from the Oven: Reinventing Dinner in 1950s America. Penguin Group
Fincher,Melanie (2022) The History of Boxed Cake Mix (allrecipes.com)
Pillsbury Kitchens (2017) The Incredible True History of the Pillsbury Bake-Off® Contest — Pillsbury.com
Marikyan, D. & Papagiannidis, S. (2022) Technology Acceptance Model: A review. In S. Papagiannidis (Ed), TheoryHub Book. Available at http://open.ncl.ac.uk / ISBN: 9781739604400
QUT IFB101 (2015, Mar 8) Youtube: Technology Acceptance Model — YouTube
QUT IFB101 (2015, Feb 26) Youtube: Theory of Planned Behaviour — YouTube
