AI is not just an experiment
Many companies are running pilots applying artificial intelligence (AI) techniques across their business processes, hoping to fold them into their operating models later. What is scarce is information about the results.
Some technology vendors mention those results superficially while setting out the merits of their products, without being precise about what was achieved — which can end up discouraging companies. Even so, market research[1] carried out in regions with greater access to technology indicates that close to 93% of companies are interested in AI.
Fears about using these technologies will fade as more information about them becomes available, though the question of whether pilots and implementations succeed will probably linger.
The purpose of this article is to offer a few recommendations for making pilots successful.
The first is to establish clearly what the AI implementation is for within the company's operations. These tools interact with our customers, if they have been designed to, so the manner and style of that communication is expected to be friendly, effective and clear.
That leads to the second recommendation: adopt technologies that can scale with the operators' understanding and that are easy for the company's people to grasp. These tools have to add something in order to be adopted quickly, and they have to increase our people's productivity with noticeable impact.[2]
The third recommendation concerns where within the business model AI should be embedded. In other words, adopt technologies that are aligned with the transactional systems and that sit inside the company's systems. Otherwise there will be no permanent, up-to-date information.
Finally, and given that these are not mass-market technologies, we have to select the best strategic and technological partner for the job. That is an important consideration, since many of the company's critical variables are at stake — its customers, its information and the confidentiality of that information.
Have you heard of agile techniques for planning and developing projects? They are ideal for this kind of project, which has to be not only fast but also properly recorded, so that damage can be controlled and lessons carried into the next implementation. That is how we reduce the uncertainty and the impact of a poor AI pilot.
[1] Big Data Executive Survey 2018. [2] “The Problem With AI Pilots”, T. Davenport and R. Bean, MIT Sloan Management Review, July 2018.
