Melissa Perri on Overcoming AI Data Bottlenecks with Product Intuition

In a recent LinkedIn post, Melissa Perri discusses a common pitfall for teams embarking on AI product development: the struggle with data acquisition. Perri highlights how many teams fall into a “chicken-and-egg problem,” where the need for user-generated data to train AI models clashes with the requirement of having AI models to attract users.

This common roadblock often leads to prolonged periods of waiting, as teams anticipate data magically appearing. However, Perri points to a solution discussed on the latest episode of the Product Thinking Podcast, featuring Vanessa Lee from Shopify. According to Perri, Lee’s team has found an innovative way to bypass this waiting game.

Cracking the AI Data Code

Perri explains that instead of waiting for years of real user interactions, teams can leverage AI to solve AI’s data challenges. This approach involves actively engineering the necessary training data rather than passively waiting for it.

“You’re basically imparting all of your product intuition into an LLM to then shape yet another LLM.”

As Melissa Perri elaborates on this strategy, the core idea is to use the team’s existing knowledge and expertise to bootstrap the AI systems. This involves taking deep insights into the product domain, customer behavior, and overall product strategy and using this knowledge to systematically generate the training data required for AI models.

From Bottleneck to Engineering Challenge

Perri argues that this proactive approach transforms the data challenge from a passive bottleneck into an active engineering task. This is a key differentiator for successful AI product teams.

“This is what separates teams that accelerate with AI from those that get stuck. You don’t find perfect data lying around. You create it using the deep product intuition you already have.”

According to Perri, the ingenuity lies in recognizing that the team’s current knowledge base is a powerful asset that can be used to build the foundational AI systems. This contrasts sharply with the common mistake of waiting for ideal data conditions that may never materialize.

“The ingenuity is in recognizing that your existing knowledge can bootstrap the AI systems you’re trying to build.”

Perri concludes by posing a critical question to her audience, prompting reflection on how others are tackling this prevalent data acquisition challenge in their AI initiatives. This highlights the ongoing and evolving nature of best practices in AI product development.