Skip to content
Fast Company Executive Board

The Fast Company Executive Board is a private, fee-based network of influential leaders, experts, executives, and entrepreneurs who share their insights with our audience.

3 elements for a successful AI project

Most clients clamor for AI, but the smart ones combine it with ROI.

[Photo: Getty Images]

Profile photo of Alex Castro

A recent MIT Media Lab/Project NANDA study stated that 95% of generative AI pilots fail. Working with many clients on their first AI project, we have observed a pattern that offers clues to both AI success and failure.

Many AI launches go straight for the dazzle. A new, shiny UI feature. An attempt to revamp an entire sales pipeline. Going from one social post a week to having AI generate all content marketing.

Then? These AI launches fail. Executives calculate tech debt. Staff go back to the drawing board, without a single problem solved.

Why do these projects fail? Because the mandate “We need to launch AI yesterday” cannot compete with “Where can AI solve a real problem and deliver immediate ROI?”

So, how can you quickly identify your first AI project? How can you feel confident in your decision? How can you dramatically increase the likelihood that this first AI project delivers positive ROI quickly?

The answer is simple. Start with boring. Boring projects make money.

3 elements that unite successful AI projects

Clients who address these three elements are more likely to be among the 5% who succeed in their first AI project.

1. Find problematic tasks

Every business has tasks that are repetitive and error-prone. The first place to look for opportunity is where you have high employee attrition.

For example, one of our clients hired seasonal auditors during tax season. It was a specialized role, with extremely high turnover. Together, we found a way to deploy AI to automate many of these tasks, which minimized attrition and helped retain their core employees.

Which roles and tasks are burning out your employees the fastest? Look for these time-wasting workflows in your operations. That’s where you should first invest in AI.

2. Train yourself to train the model

Training AI models takes time.

Every single one of our 11:59 consultants knows that their clients’ IP lives in their processes, so that’s where we focus our time. And that is where you should, too.

In their daily lives, many of our clients use Claude or ChatGPT, models already trained on internet data. This sets the client’s expectation that their first AI deployment will already be trained and work without errors or hallucinations, on day one.

But for business optimization, we use client-specific data to train the models and keep client data safe. This requires time to train the models to learn from errors in the dataset and optimize outcomes.

Train your AI on your processes. This is the key to seeing a return on your investment.

3. Make your pilot synonymous with commercialization.

Many AI failures look promising during pilot but fail during commercialization. Pilot myopia is a mistake. Use commercial tools right away.

We specialize in AWS AI implementations because AWS offers native features that support AI security. That means security is built in from the first line of code. We’ve found this approach resonates with highly regulated industries, such as financial services and insurance, where data security is paramount.

Security and governance should be built into every line of code—even during the pilot phase.

Let customer demand determine when a product becomes commercial. Don’t rely on the product being ready. It should already be ready. The transition from pilot to commercial should be instantaneous.

If you follow these three simple steps, you are much more likely to go from zero to AI-driven successfully, all while seeing a return on your investment.

Alex Castro is the cofounder and CEO of 11:59.