How to rescue a failing AI initiative

Sponsored by

Header: 59% of professionals spend 11+ hours a week chasing information across systems.†

Copy: This guide breaks down how to connect context across tools and keep work moving.

†Quickbase, 2025

You can’t win them all—advice that applies to lottery tickets, national sweepstakes, and, lately, AI initiatives.

Let’s say you’ve finally charged ahead with a long-awaited and highly anticipated AI project at your organization. The hope is that the initiative goes as expected, paving the way for other AI projects down the line. But how do you recover if those efforts fail, burning lots of time and resources in the process?

Red flags. According to Rimini Street Global CIO Joe Locandro, you can consider an AI project failing if it isn’t delivering an expected outcome. There are usually signs that an AI project is floundering. For example, you might spot employees spending too much time fixing results “due to poor data quality,” he said. 

“What you find is that a lot of the engineers keep spending time and getting delayed because the data wasn’t right,” Locandro said. “The second thing that you find is the results are hallucinating, and what I mean by that is they’re not generating very good and accurate results—and those two things are what causes the time and cost to blow out.”

Another bad sign is when organizations spend more time and effort exploring AI concepts than deploying them, said Chris Ebert, SVP of GTM at SaaS Consulting Group.

“You’re never doing any deploying, and so you don’t know what your milestones are,” Ebert said. “You don’t know what you need to achieve in order to reach the next phase of actually deploying something.”

An additional telltale sign: employees are finding ways to work around a specific AI tool.

“Either the messaging isn’t correct, the process that you’ve defined isn’t actually providing that much benefit to the people that should be observing it or realizing that value. It’s just not there,” Ebert said. “If you have a change management problem, that’s a sign of a potential failing project.”

To the rescue! There are ways to salvage AI projects seemingly on the verge of going up in flames. IT pros can start by identifying what’s causing bottlenecks and shrinking the scope of a project to be more defined.

“By shrinking the scope to really focus on one or two or maybe a handful of different workflows versus a very broad project, it can really allow you to dial in your approach and hopefully rescue that project by delivering some sort of value,” Ebert said, adding that assigning ownership to projects that lack it can also be a possible resolution.

“That may help because it removes the ambiguity of who the decision-maker is, how to progress, [and] what the outcome should be,” Ebert said.

Jillian Powers, senior director of responsible AI at AI consulting firm Slalom, said companies can “scrap” working parts of an AI project and repurpose it for other use cases.

“Maybe we built a really good RAG system for accuracy. Maybe we connect it to different disparate data sources or platforms or systems,” Powers said. “There’s ways that we could take what we’ve done and reuse them in new places.”

When to throw in the towel. Sometimes, a project might be too far gone to be saved. Ben Schein, VP of AI and analytics at Progress Software, told IT Brew that high costs and risks can indicate that ending a project might be the best course of action.

“You might be better off if there’s a high risk, if it’s high cost, if trust is is waning, of just saying, ‘Look, we’re going to cut our losses on this one, not because we couldn’t maybe get it back on track…but [because] it’s going to disrupt our ability to do other things,’” Schein said.

Companies should also establish checkpoints for when they expect to see improvement when trying to revive an AI project, said Ebert.

“You [should] have a defined window for how long a project rescue attempt is going to be…You have to define what is going to be your indicator that that project’s back on track, and if it’s not, that means it might be time to stop the project,” he said.

Post-failure. The conversation doesn’t stop once an AI project is put to rest. Locandro said companies should look back at failures as learning opportunities that can be used for future deployments.

“You do a review, and from the review you get learnings, and from the learnings you then develop an action plan,” Locandro said. “And the action plan would be, ‘What is the checklist we need to modify before we start to attempt this again with another project?’”