By Félix Chrétien
Do you want to carry out your first artificial intelligence (AI) project? It is an excellent idea. If you accumulate data, structured or not, it can hold a lot of value for our business (see our post " why now "). Be aware, however, that AI projects very often turn out to be failures. About 80% according to a handful of surveys (see for example thereof et that one). The good news is that we can significantly increase our chances of success by adopting certain good practices. Here are 3.
The first project should be the one that will be short and at the lowest possible risk. Even if it means leaving a potentially profitable project on the ice, but more uncertain. The objective of the first project is indeed to create momentum within the company. Often, a failed first project abruptly ends the adoption of AI. And this wasted moment risks causing the competition to fall behind, which is slowly but surely taking this path.
2. Added value
So the first draft should be as safe and short as possible. The only constraint: it must necessarily add measurable value to the company. Upstream, we must therefore set measurable objectives. A project can usually be separated into several models. Each of them must be done incrementally (one after the other) and each increment must add value, even if the contribution is small.
3. Technological accessibility
The first project is not the time to reinvent the wheel. The most powerful machine learning algorithms are well established, very well known and open access. No need to test a whole new algorithm, unless it is really, really, necessary. The revolution will be patient and the project will have a chance to work.
In summary: iterations
Good planning aims to maintain the momentum associated with the adoption of AI in the enterprise. She orders the projects by choosing the most accessible at the beginning. Yes, this is exciting technology and when we meet a new business, all of the teams are usually very excited to get started. But the best guarantee of success is careful management of expectations. Don't be the hare. You don't want your project to be added to the 80% shelf.