First, let’s define where we stand. Compare it with this question, “Will IT ever be able to offer an ROI for enterprises?” – that is, the same question about IT projects. Here the statistics is well-known to be about a 30% success rate. So, if AI, whose current success rate is about 10%, can reach that of the IT projects, we can consider it a success.
What are the best practices for achieving success with AI? First and foremost, the quality of the AI answers should be good enough. The measure of “good enough” is called “error rate.” For example, IBM’s translation in the 2010s was in the 80% error rate range, and that was not good enough. So, a project based on the IBM technology then would fail. Now, the error rate of Google Translate is in the 95% range. This is good enough, and people find it satisfactory. By “people” I mean lawyers who are using it to translate the documents in their cases.
Second, the results may be “good enough” but not needed by the people. Many of the unsuccessful AI projects fall into this category. Here too, it may not be needed by the people because the quality of the results is not good enough. In this case, the usual solution is to force its use through sales or through the judge telling that one must use it. It works for a while but fails in the long run. I am actually writing a book about this, where I review the technique for defining AI applications that may be useful to people. One possibility to promise a good ROI is to study my book when it comes out, in about 4 weeks. Another is to study the use cases of successful AI projects. See our course /course/ai-for-nlp-advanced/
Thirdly, and just as important as the previous two, the AI project must include the discipline of Machine Learning Engineering or ML Ops. Today, there are many AI scientists who can run their AI in their lab environment (called Jupyter notebook) but are clueless about putting it into the production pipeline, called AI pipeline. See our course, /course/ai-next-step-ml-ops/
In summary, AI will be able to offer ROI for the enterprise, and we are in the time period when we have examples of it being possible. Education and experience will have to precede it, because of “no pain – no gain.”