Defining a good AI project is a fascinating challenge. Unless you’re doing pure research, it requires you to find a strong intersection between business and technology. Finding that intersection is hard for many reasons. Today I want to focus on a specific one: the two worlds of business and technology are often in the hands of different people within an organization, and making them work together effectively isn’t trivial.

Let’s put ourselves in the shoes of a hypothetical pharmaceutical company, Acme Bio (what we’ll say applies to every industry). In this overly simplified example, let’s assume Acme Bio has these different challenges:

  1. Improving the operational efficiency of their supply chain
  2. Cutting costs of their production systems
  3. Develop new digital solutions for patients
  4. Improve the efficacy of their marketing campaigns
  5. Speed up their drug development efforts

Potentially, each one of these areas could be tackled with AI. You could design AI systems to automate supply chain tasks, but which task is the best to start with? Assuming you have an idea of where to start, are you sure you have the right data? Is your project even feasible?

You could also use AI to cut costs of the production facilities. For instance, AI has been extremely useful to optimize energy consumption of industrial plants. Or maybe it’s better to use AI to do predictive maintenance instead? That can also work, right?

You could decide to start with a more customer-centric approach instead of looking at your operations. For instance you could develop a chatbot to answer patients’ questions and offer them a better therapy experience. But what if the AI gives the wrong advice? Will it work well enough to comply with regulations?

You can see where this is going. The problem here is that a specific set of people in Acme Bio has a good understanding of what the company needs, but another set of people know what Acme Bio can actually build!

Who should be in charge of defining what Acme Bio will end up building? In my experience, I’ve seen companies trying two different options: starting from the business or from the technical teams (IT, Data Science teams, etc.).

If the business is in charge of defining AI projects, I’ve seen two risks unfold:

  • Without a solid understanding of AI, business people often design projects that are unfeasible, extremely complex to build, or that the company doesn’t have the capabilities (or the data) to build (overshooting)
  • Sometimes business people underestimate the potential of AI and miss great opportunities (undershooting)

On the other hand, technical teams tend to make different mistakes. The main one is build technology that looks great but no one wants.

This is a true personal story: I once built a Data Science project designed by an IT team that got everyone excited on the IT floor. We thought we had found a multi-million dollar opportunity, and presented the proof of concept to the GM. He stopped us after 5 minutes because he already knew everything we had found in the data - turned out that we missed a key variable, and what we thought was a massive untapped opportunity was actually a very profitable business. We just looked at the data from the wrong point of view, without the right business understanding.

What’s the solution then?

From my experience, it’s very hard to make technical people “get” the business side. Obviously it’s not a problem of capabilities - technical people are smart enough to understand what business means - the issue is that it takes a different mindset to solve technical or business challenges. On top of that, tech teams are often too far from the business that they’re not exposed enough to what the business needs.

If the answer to the question “who should be in charge of defining AI project” is not “the tech team”, it must be “the business”. There’s a catch though. We mentioned before that business people tend to overshoot or undershoot their AI projects, and you need to have in place strategies to contain that.

The two strategies are:

  1. Invest in AI education for business leaders. It’s hard to teach business to a tech person, but it’s much easier to teach basics of AI to a business person.
  2. Form effective cross-functional teams

Both strategies are not trivial - you need to implement them properly. I’ll talk about my learnings on how to do it in a future blogpost, subscribe in the box below to be updated.