Your AI platform is too data-hungry

“Your AI platform is too data-hungry.” This comment from a potential user at a conference caught me off guard. Not because it was true, but because of why they thought it was.

A couple of years ago, I was chatting with someone at a conference about conversational AI platforms when they hit me with that line: “Yeah, but Rasa is really data-hungry.”

My immediate thought was 🤔 Have you run benchmarks? Is there research I’ve missed?

It turned out their conclusion came from an unexpected place: our documentation.

We talked (a lot) about the importance of high-quality training data, while other platforms didn’t. They assumed this was just a deficiency of Rasa — that our product must be data-hungry because we were the ones emphasizing the need for data.

The irony is, if you’re using supervised learning for NLU (or anything else!), quality training data that reflects real-world user inputs should be your first, second, and third priority. It’s far more important than your choice of model.

But leading with this and advocating for best practices actually hurt the perception of our product, at least in some users’ eyes.

It was a tough lesson: sometimes the market will punish you for trying to be helpful.