It’s often said that AI consumes data, so without data, is it impossible to initiate an AI project for your company?
Not quite.
With CRIM, you can start your AI project even before you have the data.
Two trends are converging. On the one hand, the arrival of pre-trained models – such as those of generative AI – which can be useful “as is”, without requiring a vast input dataset. On the other, the possibility of leveraging existing data, however imperfect, and building iterative projects around them.
“Five years ago, barely one organization in 20 had enough adequate data for predictive AI purposes. Today, all those with an existing culture of capturing their data can give it a reasonable shot, and soon, those without data too,” explains Michel Savard, Data Science Practice Leader at CRIM.
It’s a real paradigm shift: AI is becoming accessible to organizations that previously had no chance of taking advantage of it.
Today, anyone with an existing culture of capturing their data can give it a reasonable try
This approach is based on an incremental logic: start small, demonstrate feasibility on a similar corpus or with partial data, before investing in collection and annotation – two costly steps. “Often, our value at CRIM is to prevent an organization from wasting two years’ work”.
CRIM’s role is to help companies translate their business objectives into relevant analytical tasks.
“Too often, we’re asked for a prediction, when the real need is either optimization or scheduling”. By designing an AI initiative in line with a data acquisition plan, even if this seems limited in the early stages, new perspectives open up.
Here’s a concrete example: a venue booking platform thought it had a recommendation project. By rethinking how to capture interactions between its users, it discovered levers to measure quality, generate specialized performance indicators and automate several steps well beyond the initial recommendation. “They came out saying: now we’re doing AI, because we’re thinking about our business as a source of valuable data.”
So it’s no longer a question of data quantity, but a new way of launching AI projects, focusing first on the business problem, then on the data strategy.