4 Ways to Avoid AI Pilot Purgatory

Enterprise adoption of AI has tripled since 2019, but executives are complaining they’re not seeing value on their investments. Is the problem the technology -- or is it you?

There are two common reasons that AI projects fail to show value:

Here are four things to do before starting the pilot to give you the best chance of success:

1. Define your use cases

Two or three use cases are best. That way you can prevent putting all your eggs in one basket, but not lose focus with too many use cases. Think carefully about the complexity of your use cases: Complex improvement initiatives typically left on the back burner are ideally suited for machine learning, while more basic day-to-day problems are best solved using simple analytics software or even Excel.

Last but not least, make sure you’ve accurately assessed your data maturity for each use case. This is a key reason why pilots don’t prove value. Do you measure and record the data that the software will be using? How much data history do you have available? AI and ML are not magic; they’re just tools that are only as good as the data you put into them.

2. Define a pilot success plan

Many people just start with data and forget to think about what success actually means. This is the key element of your pilot preparation. Knowing how you calculate ROI for your particular use case means you’ll know right away if you’ve achieved value.

Your success plan should include:

3. Build a project team

Gathering the right team is crucial for succeeding with a pilot. It’s best if participants come forward voluntarily, so they have the most buy-in. They should also ensure they have enough time to participate (no one wants delays).

Here’s a list of the key team members you’ll need. Sometimes the same person may be responsible for more than one of these roles:

Project manager: This is the main stakeholder who will execute the analysis. In other words: the person who will be running the pilot. Ideally, they are a trusted expert in the field of digitalization and are connected across different sites or business units. It’s also great if they are enthusiastic about selling projects internally. That can make a huge difference when it comes to trust in the new technology.

Domain expert(s): If you have a very technical process, it’s best to have at least one person with deep knowledge about that process who can help train the models. This function could be multiple people, for example, a process engineer and a well-trained operator.

Sponsor: This person has enough budget not only to buy a pilot, but also to move to the next contract stage. They should be enthusiastic about innovation.

Data person: Someone who can download data from relevant databases.

Data scientist (optional): It’s great to have someone who can interpret the data and help draw conclusions.

4. Define the pilot management process

It’s important to make sure everything works and there are no big delays in the middle of the pilot. Schedule Sessions with key users (weekly); progress check-ins with project manager (weekly); at least two meetings with the sponsor to provide high-level updates about value/progress (monthly); and a final meeting to present results and discuss next steps.

Now that you know what success will be, you have a better chance of reaching it! Once your pilot is finished, you can move on to scaling the success you’ve seen across your organization.