Pruning Is Not Retreating
Three-quarters of tech leaders say their data and AI costs are climbing and only 38 percent report significant ROI, so the interesting number is what they are doing about it.
What the survey found
An Ocient survey of 500-plus U.S. IT and data leaders (reported by CIO Dive on Nov. 5, and worth remembering that Ocient sells data infrastructure) found that nearly a quarter have reduced or paused underperforming AI projects, while about 45 percent have put more money behind the ones that work. Unpredictable cost is now a top data challenge for 41 percent, up from 35 percent two years ago. The report's own line is that pruning isn't pulling back, it's making room for what's worth growing.
Write the stop criteria first
We've seen clients kill projects that were still in the technical prove-out because someone else's project was outperforming (or because theirs was never measured at all). This is one of the key criteria for a successful implementation: you have to set the metrics and ROI before you start. Then you know along the way whether you're winning and whether it makes sense to continue. If you make them up at the end and rationalize the effort after the fact, you never had a chance along the way. (I covered the pilot-versus-production version of this in You're Paying for the Pilot. You're Not Getting the Production.)
The same report says security risk, financial return, accuracy, and cost at scale carry nearly equal weight when leaders decide what to keep, which means a project can fail on any one of them. So step one is to write the stop criteria before the project starts. Step two is to schedule a review where someone has to defend each project to keep it, instead of waiting for someone to propose killing it.
What to do
Pull up your list of active AI projects and check which ones have a written stop criterion. What is the last AI project you shut down on purpose?