This interview is with Anastasia Trofimova, CEO Founder, Sundae Foods.
Looking back, what pivotal moment moved you from advising grocery CEOs at BCG to founding Sundae Foods?
There was not one dramatic boardroom epiphany. It was a Tuesday night.
I had just come off a ten-hour day, and I was standing in a grocery store scrolling through a healthy recipe I had saved on Instagram weeks earlier, trying to figure out how to actually make it happen. I did not have the ingredients, and I did not have the energy to piece it all together. I ended up grabbing something processed and going home with that familiar guilt that I think so many busy women carry quietly every single day.
But what was different that night was that I could not let it pass. I had spent years at BCG sitting across from the CEOs of some of the largest grocery retailers in the world, and I knew exactly how that industry thought, what it prioritized, and crucially, what it had never once been designed to solve. It was built for transactions, not for the way people actually live, eat, and get inspired in 2026.
The gap between those two things, between a $12 trillion industry and 70 million people searching for healthy recipe inspiration every single day with no way to shop it, was so obvious and so enormous that I remember thinking: if I do not build this, I will spend the rest of my career wondering why I did not.
So I stopped advising the industry and started disrupting it.
Zooming in on product performance, drawing on your management consulting and data analysis background, what single metric has proven most predictive of a recipe’s shopability on Sundae?
The metric that has proven most predictive is what we internally call save-to-shop conversion: the rate at which a saved or shared recipe actually converts into a completed grocery cart.
What makes it so powerful is that it cuts through all the vanity metrics the creator economy is drowning in. Views, likes, and saves all tell you about inspiration and intent. Save-to-shop conversion tells you about execution, and execution is exactly where the food discovery funnel has always broken down.
Recipes with the highest conversion share three characteristics consistently:
- A short, recognizable ingredient list of fifteen items or fewer.
- A clear nutritional outcome, such as high protein, low carb, or a specific calorie target, because people shopping with health intent are far more likely to follow through.
- They come from creators whose audience trusts them enough to act, not just scroll.
That last point is particularly striking because it makes follower count a surprisingly weak predictor. Micro-creators with highly engaged, health-focused audiences consistently outperform creators with ten times the following on pure conversion. The trust signal matters far more than the reach signal.
But the most important thing this metric revealed is something more fundamental. People genuinely want to eat healthily and improve how they feel. That desire was never the problem. What was missing was a frictionless way to act on it. The moment we removed that barrier, conversion followed naturally. They were not waiting to be convinced. They were waiting for the infrastructure to catch up with their intentions. Sundæ gave them that, and they ran with it.
That insight now drives everything: which creator partnerships we prioritize, which recipes we surface, and where the real commercial value in the creator economy actually sits.
On the AI side, how do you structure the workflow between AI-generated meal plans and nutritionist verification at Sundae?
The way we think about it is that AI handles the scale, and nutritionists have built the intelligence behind it. Those are two very different jobs, so we keep them deliberately separate but tightly connected.
On the AI side, the engine is doing the heavy lifting of personalization at scale. It ingests the user’s dietary preferences, health goals, and shopping history, cross-references that against our full recipe library, and generates meal plans that are both nutritionally coherent across the week and shoppable in a single cart. Every single recipe that moves through Sundæ gets a full micro-nutrition breakdown powered by the USDA database, and the algorithm itself was designed by nutritionists from the ground up to recommend balanced, nutritionally dense whole food meals. Not just calorically adequate, but genuinely nourishing.
This is where we differ from a lot of food tech products that bolt nutrition on as an afterthought. For us, it was a first principle. The nutritionist expertise is not a layer we added on top of the AI; it is baked into the logic of how the algorithm thinks and recommends from day one.
The result is a system that personalizes at the speed and scale modern consumers expect, while carrying the credibility that health and nutrition content demands. In a space where we are directly influencing what millions of people put in their bodies, that scientific foundation is not a nice-to-have. It is the entire point.
Operationally, what was the hardest edge case you had to solve—such as SKU mapping, substitutions, or inventory variability—to make instant recipe-to-cart work at scale?
The hardest part explains why this solution has never been built before, despite social commerce platforms in fashion and beauty becoming billion-dollar businesses. LTK, which connects creator content to shoppable fashion, is valued at $2 billion. The reason nobody cracked grocery first is simple: grocery retail data is a mess.
In fashion and cosmetics, product databases are clean and structured. Grocery is the opposite. Retailer databases are built on unstructured string text, meaning ingredients are stored as free-form descriptions with no standardization and a massive margin for error. When you are algorithmically matching recipe ingredients to a live grocery catalog at scale, that messiness becomes a serious problem fast.
We were not willing to take that risk. A bad match in fashion means the wrong shade of lipstick. A bad match in grocery means someone with a dietary restriction gets the wrong product, or a recipe calling for plain yogurt gets matched to yogurt-covered chocolates. That is not a minor inconvenience; it is a trust-destroying failure that kills the product.
So we built a matching algorithm specifically designed to navigate unstructured data, identify ingredients with high accuracy, and reject ambiguous matches rather than guess. We chose precision over speed at every decision point. That painstaking infrastructure work is exactly why Sundæ works reliably where every other attempt in this category has stalled. The messiness of grocery data is our moat because we did the hard work of solving it.
That said, we are still an early-stage startup, and we know we are not yet where we want to be. That is the journey ahead. Every positive signal from our users feeds back into the algorithm and makes it sharper. For us, being user-led in how we prioritize our technology roadmap is not just a product philosophy; it is our biggest advantage. Building Sundæ with our community involvement is honestly one of the greatest privileges of this whole journey.
Internationally, drawing on your project management experience, what was the most surprising consumer behavior difference you encountered when expanding from 40,000+ Swedish households to the United States?
The most surprising difference was not what we expected going in. We anticipated the obvious things: different grocery chains, different ingredient availability, and different price sensitivity. Those were knowable and plannable. What genuinely surprised us was the difference in the relationship people have with food inspiration itself.
In Sweden, our users tend to be quite intentional and structured about meal planning. They come to Sundæ with a clear weekly mindset; they want to plan ahead, shop efficiently, and reduce waste. There is a strong cultural current of practicality and sustainability running through how Swedes approach food. The app fit naturally into a habit that already existed.
In the US, the relationship with food inspiration is far more emotional, spontaneous, and identity-driven. American users are not just looking for a convenient way to execute a meal plan. They are looking for food that expresses who they are, aligns with a wellness identity they are building, or reflects a creator they trust and admire. The decision to cook something is as much about aspiration and self-expression as it is about nutrition or convenience.
The other surprise was the sheer intensity of the healthy eating conversation in the US right now. The GLP-1 and Ozempic wave has put nutrition at the absolute center of mainstream culture in a way that has no equivalent in Sweden. We landed in the US at exactly the right cultural moment, and that urgency in the market is something we could not have fully anticipated from the outside.
From a research lens, which method or signal has most accurately captured shifts like GLP-1 adoption and women’s nutrition priorities for your users?
The most accurate signal has not come from formal research. It has come from behavioral data inside the app itself, and it has consistently told us things faster and more honestly than any survey or focus group ever could.
What we have observed over time is a clear and sustained shift in the nutritional filters our users are applying. There is a growing appetite for high-protein, nutrient-dense, whole food recipes, and a move away from anything that feels processed or artificially diet-focused. Whether that reflects broader cultural conversations around health, the GLP-1 moment, or simply a generation of consumers who are more nutritionally literate than any before them, we cannot say definitively. Probably all three. But the direction of travel is unmistakable, and it is accelerating.
For women’s nutrition priorities specifically, the most revealing signal has been recipe save patterns combined with the nutritional categories driving the highest cart completion. What we consistently see is that women are not just optimizing for calories the way older diet culture would suggest. They are optimizing for energy, for gut health, and for ingredients that make them feel good throughout the day. Anti-inflammatory recipes, high-fiber meals, and protein-forward breakfasts—these are the categories that index dramatically higher among our female users and convert at the highest rate.
The broader lesson for us is that in a world drowning in survey data and focus group findings, the most honest research tool you have is a product that people use habitually. Behavior does not lie the way self-reported data does. What someone actually puts in their cart on a Wednesday night tells you infinitely more about their real priorities than what they say they care about in a questionnaire. That behavioral lens is now central to how we make every product and partnership decision at Sundæ.
Looking ahead 12–24 months, based on patterns you’re seeing in Sundae’s data, what AI capability will most transform shoppable recipes and meal planning?
The capability that will most transform this space is what I would call predictive personalization at the household level, and we are already seeing the early signals of where it is heading.
Right now, most meal planning tools operate on a reactive model. A user inputs their preferences, browses, or discovers a recipe, and the AI helps them execute it. That is already a massive improvement on the status quo. But the next leap is moving from reactive to anticipatory, where the AI understands a household’s rhythms, nutritional patterns, budget cadence, and taste evolution well enough to surface the right meal at the right moment before the user even goes looking.
Think about what that actually means in practice. The app knows it is Wednesday, that this household tends to cook something quick mid-week, that they have not had enough iron this week based on their recent shopping history, that one of their favorite creators just posted something that fits perfectly, and that three of the five ingredients are already likely in their pantry. It knows that the kids have soccer practice that evening and that they will need a higher protein, more nutritionally dense dinner to recover and refuel properly. That level of contextual intelligence, operating quietly in the background and surfacing exactly the right recommendation at exactly the right moment, is what transforms meal planning from a task into something that feels almost invisible.
The second capability that will be transformative is real-time nutritional optimization across a full weekly shop rather than individual recipes. Right now, we optimize at the recipe level. The next frontier is optimizing across an entire week of eating simultaneously, balancing macro and micro nutrition, variety, budget, and ingredient overlap to minimize waste, all in a single cart. That is a genuinely hard AI problem, and whoever solves it well will define the category.
For Sundæ, both of these capabilities point in the same direction: a future where the gap between wanting to eat well and actually eating well disappears entirely. That is the product we are building toward.