When doing artificial intelligence (AI) at an enterprise, you will have a large number of potential use cases brought to you. Many more than the internal AI team can implement themselves or even supervise from outsourcing companies. So naturally, the critical question is which projects to do first versus later and which projects to drop altogether. Let’s look at some options for doing this and their consequences.

Nearly every enterprise prioritizes projects in one of three ways: By political pressure, by difficulty or by risk. I advocate that you prioritize by value.

Prioritization Option 1: Political Pressure or Whoever Shouts the Loudest

A common method of prioritization is to focus on projects suggested or supported by senior and powerful leaders in the company. A variant is to focus on those projects that get actively pushed and followed up on. The thought behind this method is a desire to avoid conflict and the ability to delegate responsibility to these senior leaders.

Senior leaders in most enterprises are not AI experts and so cannot evaluate the specifics of what they are asking for. Typically, they have also not carefully evaluated the business outcomes but rather focus on problems that are visible to them. People who shout do this based on their own personality and their strong feelings about the issue, not because the problem is particularly acute.

AI leaders do not serve the enterprise by choosing this option. They also do not serve themselves because they effectively demote their own leadership positions into an order taker role at the AI drive-through.

Prioritization Option 2: Difficulty or the Myth of the Low Hanging Fruit

Another popular option is to focus on projects that are easy, cheap, or fast to do. These are known as the low hanging fruits popular in management literature. They are supposed to quickly demonstrate that something or someone can accomplish something good and thus establish that they have earned their place and can be further supported.

AI projects in an enterprise are rarely low hanging fruits, especially at the start of an AI function. The reason is simple. Before an AI project realizes any value at all, before it can be used by a business user in any way, the AI function must deliver fully functional software that does something useful, even if it is small. To do this, a great many moving parts of data pipelines, cloud services, models, software and process interfaces must be stitched together.

Even then, significant domain knowledge needs to be collected to make sure that the software does something the business users find useful. This does not happen quickly. Certainly not when the AI function is new and small. The AI tree does not bear fruit on low branches.

Prioritization Option 3: Risk or the Tale of the Sure Bet

Any new technology is risky, especially one that is very actively evolving, such as AI. It is reasonable to want to focus on projects that are low risk. There are two flaws in the details of this otherwise good idea.

First, it is hard to evaluate risk because there are several (almost) independent sources of risk. There is scientific risk of the AI model not being accurate enough or being accurate for only a very narrow distribution of input values. There is the software risk of the data pipeline or cloud ecosystem not working according to specifications. There is the economic risk of consumption costs being too high. There is a competency risk of the team lacking some skills required to do the project. There is time risk because it is seldom clear in the beginning how long this will take. Most importantly, there is scope creep risk because the users never know what they want before they do not get what they need.

Second, most enterprises will pursue several projects simultaneously. A portfolio approach to these is appropriate in which there would be some sure bets but also some more risky options promising a high reward. That would make sense from a wider lens of a diversified investment strategy.

Essentially, ordering an AI project list by risk is not practical. This is especially true in large enterprises where that list easily contains several hundred entries that would each have to be carefully evaluated.

Prioritization Option 4: Value

I propose to sort the project list by the value provided to the business. That sounds trite at first, but it includes a few caveats. The obvious reason is that the value is a single number that can be meaningfully compared. That number is relevant to the enterprise’s bottom line, which is, of course, the primary purpose of the enterprise. Generally, all the leaders can get behind doing something that’s valuable.

So, that leaves us with the question of how this number is obtained. The value must come from the business, not from the AI function. It must also come with a detailed explanation. STEM people might call it a derivation. This would include some basic assumptions and numbers around the context. One might say that this project is relevant to a certain number of machines, assuming certain market prices and so on. Finally, the leader of the business unit would formally sign off on this assessment.

The formal signoff is crucial in this process. It provides buy-in from the business and formal support for the project. The fact that this was done signals that the business has understood what it is asking for and has thought through, at least superficially, what this application will look like in everyday business reality. This implies that the project is not an alibi innovation project, but something that will yield true change.

The process of value derivation and signoff can take weeks, but it is an investment that is well worth the effort and will pay dividends during the project and after its completion.

Road to Success: Value

One of the most important choices of an enterprise AI function is which projects it will focus on. That choice is not a single event but an ongoing activity as projects surface and get added to an ever growing and soon to be very long, list of options. Faced with limited staff and budget, only a small fraction of the options can be pursued.

Instead of bowing to politics, trying to do the easy stuff, or taking sure bets, I have argued that the road to success lies with value. If the value is well assessed and attested by business leaders, you will have an excellent sorting criterion alongside business support with clear expectations. That will set you up for success in the project and the enterprise. Last but not least, it will allow clear and transparent communication on why these projects are being pursued.

Save 30% on Leading Enterprise AI Programs with code ABS30.