How we run our in-house generative AI accelerator: Framework for ideation

At some point, every scaling business faces the decision: Do we stick to our core activity or venture into uncharted territory? It’s a choice between enhancing existing products and branching out to create something entirely new.

For tech giants willing to expand into uncharted territory, the common approach often involves acquiring early-stage companies or even teams and integrating them into their ecosystems. This is because for mature organizations, breaking away from their core activity is no easy feat. In practice, going from zero to one is always more difficult than scaling from one to 100. In fact, less than 8% of ventures launched internally reach scale.

And the mergers and acquisitions (M&A) strategy may not necessarily be the answer for growth either. This approach demands deep pockets, a luxury primarily enjoyed by industry titans such as Adobe, Google, or Meta, along with the expertise to acquire and effectively integrate ventures. And still, 70% to 90% of acquisitions end in failure.

At Reface, we embarked on a difficult challenge to learn how to build new businesses within the company. In this article, I’ll unveil our framework for evaluating new product ideas and how we determine whether a concept merits our pursuit.

Enlightenment or methodology?

Classic Y Combinator advice on “noticing” ideas is undoubtedly a great way to discover ideas; however, it cannot be the main approach for a company that needs to constantly generate ideas to work on. Therefore, through trial and error, external advice, and extensive research on the topic, we have been developing our own methodology. This methodology can result in a higher number of ideas that are good enough for testing compared to just following the “noticing” path.

The first essential step to cultivate innovation within an established company is to create a dedicated space where new ideas will be born. In the daily grind of a functioning business, employees are usually occupied with their daily tasks and tend to stick to established processes and clear objectives, rather than exploring new ideas or initiating new projects. Therefore, it becomes essential to establish conditions that encourage idea generation.

This can involve setting up a small, dedicated team (ideally no more than five people) responsible for managing the idea backlog, creating an analytical feed of ideas, and developing a framework for evaluating and prioritizing ideas based on your company’s strategy. I believe in collective brainstorming during the initial stages of idea exploration, gradually transitioning to clear ownership during the idea execution phase to closely resemble the circumstances of a founder launching a startup.

Building an analytical feed of ideas is equally important. We designed a system that parses recent ML research papers, reacts to trends in social media and search spikes, and tracks various other sources of inspiration.

For the entire ideation process, we offer two key pieces of advice. First, adopt a laser-focused approach with concentrated sprints aimed at a group of ideas. When one idea is rejected, swiftly move on to the next. Second, focus on identifying promising idea spaces, not a specific idea. While the initial idea may not succeed, exploring the right space can lead to the discovery of viable alternatives. In essence, we hunt for fertile ground, trusting that within it, the seeds of innovation will naturally sprout.