Hugo: Building consumer simulation
Over the past few months we have gone down the path of consumer simulation. This is a change from what we originally set out to do, building the world’s best marketing agent. As founder I therefore feel an obligation to outline our technology and product direction, and the following is my attempt to do exactly that.
Hugo is a consumer simulation company that creates technology which enables you to know the outcome of any decision before it happens. To accomplish this we train our own highly specific models against revenue impact, and on top of those models we create products that are paired with LLMs and deliver consumer simulations to our customers in an integrated way. This might seem far from building a marketing agent, but it was actually a very natural evolution: first mapping out consumers, then analyzing them, and finally landing on simulating them.
The ability to simulate the outcome of a decision before it happens sounds like it belongs in a science fiction book, but so did all great technology before it became part of our everyday lives. We believe the same will happen for consumer simulations. In fact, the consumer simulation industry is comparable to where personal computers were in the early 1970s. We are starting to put the technology together and will soon reach a first massive financial outburst.
In a not so distant future, as consumer simulations become better, cheaper, and faster, we will see simulations become part of every single decision a company makes. Daily decisions like partnering with an influencer will be backed by a thousand simulations. Big decisions like sponsoring the World Cup, or where to open a new hotel, will be put through highly sophisticated and expensive simulations. And small businesses will unlock the executive power and strategy previously only accessible to big budgets.
Creating great products and technology usually means removing everything that doesn’t truly matter and having an extreme focus on what actually does. This often means taking a different route than the broader industry, and that is certainly true in consumer simulations. Every other company in the space is trying to create a replacement for traditional consumer research: artificial populations that respond to surveys the way real populations would. The problem with consumer research surveys is that the answers are vague, highly uncertain because people don’t mean what they say, and hard to measure and train models against. Even if you succeed in getting close to perfect survey answers, you still do not know whether it translates into real revenue. At Hugo we are taking a different approach, focusing on what is most measurable (and therefore easiest to train models against) and at the same time most directly impactful on the business: we are training models to simulate revenue impact. This gives us clarity on what progress looks like. Getting closer and closer to simulating the exact dollar amount of a decision compared to how it actually turned out, and we shall continue to care about nothing else.
Our approach leaves us with two factors to optimize for: the accuracy of our simulations, and the range of decisions we can simulate. Every incremental improvement on either side opens up new markets and allows for greatly improved products built on top of our simulation technology.
Consumer simulation relies on training models on previous behavior data, so access to great and reliable data will remain a constant priority for us. Besides the obvious solutions like scraping and buying data, we are in a position to create a strong moat by adopting the “Tesla method”. Tesla was never just a car company but rather a data company, where every time you drive a Tesla you are creating training data for their models. Similarly, every time a company makes a decision, we see it as our mission to record what led up to that decision and the result of it, making every new customer into a data supplier. This approach only becomes truly useful with hundreds of customers, but it will lead to an enduring moat.
Our simulation technology is only as great as the product we put it inside and supply to customers. We believe that having simulation be a separate platform, where users have to put in a decision, get an outcome, and then implement it themselves, is like buying a computer where you have to assemble the parts before you can use it. It’s a terrible user experience. Simulation should fit seamlessly into companies’ workflows and live where they already are. Today we do this in the shape of AI agents with access to the consumer simulation, integrated with companies’ internal data and tools, living where they already work across Slack, Teams, and Claude. Agents turn out to be a very good approach because the simulation doesn’t have to be perfect if the agent has a defined goal (for example, increase presence in Germany) and can iterate and actively work towards it using the simulation as its base. Today, and for the foreseeable future, consumer brands will be our only ICP, as they are constantly making decisions on pure guesswork and always looking for ways to outpace their competition.
In the future we see our simulation technology integrated everywhere a company works, for example into a meeting where participants simply say “simulate that idea” and get an almost instant result, transforming meetings from a discussion into a live testing ground. We also see an API as a viable future option, enabling every company’s AI agent to run simulations.
I have now covered the majority of my core ideas on the technology and product direction Hugo is headed in. While I hold my beliefs with enormous intensity, I will change them if new information shows it is beneficial to do so.
Olle.