Machine learning models tend not to generalise from one task to another. Buying access to generic AI solutions is unlikely to give good results.

A host of companies are cashing in on the excitement and promise of AI and ML by offering so-called 'AI-as-a-service'. According to an article in TechCrunch, their customers are unlikely to get much value.

Machine learning models can work well when they are trained on a tightly defined task, using a large amount of data. However, these models tend to be dumb and somewhat fragile. If the data changes they can break, if the business context changes they can give unhelpful predictions. Using the technology successfully usually takes the dedicated efforts of highly trained teams of experts. Event then, sometimes it still fails to deliver value.

The idea that a plug and play ML service can create value without any data or understanding of the business content is, I'm afraid, a myth.

Value can be added by external service providers by either providing the expertise to train bespoke models, or by providing software that makes the management and deployment of models easier.