I’ve been thinking about this a lot lately after coming across a cluster of startups trying to upend market research using AI. Within the span of the last couple months, companies like Simile and Aaru have raised hundreds of millions of dollars chasing the same idea: replace traditional consumer research with AI-generated simulations. It caught my attention because it cuts right to a problem I’ve run into throughout my career in BD. One of the harder parts of business development and partnerships is figuring out how customers will respond to something before you’ve built it. Will they actually want this product? Will they adopt this feature? Will the new product/service cannibalize other areas of your business? These questions typically sit at the center of most BD decisions and yet we often have to answer them with surprisingly limited information using more “gut” instinct.
Real customer insight is often slow and expensive and If you want statistically meaningful studies, you’re looking at securing research from Gartner, Ipsos, McKinsey, etc, a process that can run well over $100K easily and take months by the time you’ve scoped it, got internal budget approved, recruited panelists, run focus groups, and waited on analysis to be complete. Oftentimes, that timeline doesn’t fit how deals actually transpire in my experience. Opportunities emerge fast and partnerships evolve in real time. Because of this, most BD teams do what they can which is they talk to a handful of customers, debate internally, and lean on pattern recognition, which usually works fine to some extent, until it doesn’t.
The concept behind companies like Simile and Aaru is intuitive once you understand how it works. Instead of surveying real people every time you have a question which can be expensive as discussed, you build AI models that “represent” real consumers. These models have been trained on survey data, demographics, behavioral patterns and purchase history. Simile, which spun out of Stanford and recently raised $100 million refers to these as “digital twins.” For example, instead of recruiting 1,000 people for a survey, BD teams could query 1,000 simulated consumers and get directional answers in minutes rather than weeks/months. A similar competitor to Simile, Aaru, is pursuing something similar, focused on how specific consumer segments might react to campaigns, brand positioning, or new products which could be immensely valuable for marketing teams as well, not just BD. Having tried Simile it’s not a crystal ball, and no one is claiming these models predict behavior with perfect accuracy, but fast and directional is genuinely useful when the alternative is slow and expensive or nothing at all.
Most of the conversation thus far around these tools has centered on marketing, which makes sense since marketing often owns research budgets. But I’d argue BD teams stand to benefit just as much, if not more. BD sits at the intersection of strategy, product, and distribution. We are constantly evaluating opportunities where the central question is: how will customers respond or do I have the right product/value proposition? And you’re usually doing it on a timeline and/or budget that doesn’t leave room for proper research.
I’ve been in that position more than once. When I was working on a platform competing with Uber, we spent a lot of time trying to understand why customers chose one service over the other in specific situations (e.g. was it price, availability, brand, loyalty, etc)? Those behavioral questions mattered for how we thought about partnerships and distribution, but getting clean answers quickly was hard. We occasionally consulted expert networks like Alphasights, GLG, etc, but that didn’t give us direct customer feedback, but rather insight from other experts who had been in our position before and was not always useful. Whereas, a simulated consumer model trained on users of both platforms could have let us pressure test hypotheses in days instead of months.
Another example was at Blue Apron: the category was getting squeezed by the large food delivery players (UberEats, DoorDash, etc). The question we had to answer wasn’t abstract, it was whether customers primarily valued the experience of cooking their meals or were increasingly just seeking convenience. Getting that wrong had real implications for product and partnership strategy, but the pace of the business (and budgets) didn’t allow for a full research study every time we were debating a particular direction.
When I was at Lime and we were seeking new ancillary high margin revenue streams, I recommended launching Lime Ads, but we debated internally whether wrapping our bikes/scooters in advertiser branding would affect how riders perceived the Lime brand. For example, would a Mastercard wrapped scooter still read as Lime, or would it dilute the identity causing people to search, for example, for Mastercard bikes in Apple’s App store vs Lime? That’s exactly the kind of questions a simulated audience could help test cheaply before you’ve committed to anything in the real world.
The place I think these tools become very powerful is early in the idea development process, before anyone’s committed resources. In my experience organizations have no shortage of ideas from new products, potential partnerships and strategic experiments, but rather struggle most for validation, not creative ideas. You can only run so many research projects, so ideas either die in limbo or move forward based often on instinct. If AI consumer simulations can lower the cost of a directional gut check, you get a better filter and that leads to weak ideas getting killed earlier rather than later. To be fair I’m not suggesting this replaces actual customers immediately. Human behavior is messy, and the companies building these tools are the first to say so and simulated insights still need to be validated against real-world data, but I would think of it less as a replacement for research and more like a prototyping environment for business ideas – a way to quickly gut check things. Our product/eng teams don’t skip testing because prototypes aren’t perfect, but rather they use prototypes to figure out what’s worth building before committing to the real thing. That’s the version of this I find compelling.