.svg)
Most AI projects never make it past testing. Here's how to know if yours is one of the few that actually made it to production, and if it's paying off.
TL;DR
Your company greenlit an AI project at some point, a chatbot, a deep learning tool, an automated workflow. Someone somewhere is quietly wondering whether it's actually doing anything, or whether it's become one of those initiatives that technically launched and technically still exists, without anyone being entirely sure it's worth the investment. That's a completely normal position to be in, and it's worth a straightforward answer rather than a vague sense of "probably fine."
The first thing worth knowing is that most AI projects never even reach the stage where this question applies. According to Deloitte's State of AI research, only 25% of AI experiments have moved into production to date. The other 75% are still stuck in pilot, testing, or somewhere in between. If your project actually made it to production and is running against real customers or real data, that alone puts you ahead of most companies attempting the same thing. The bar for "did this even launch properly" is lower than it feels, and worth acknowledging before jumping straight to judging performance.
For projects that do reach production, the outcomes tend to be genuinely encouraging. Google Cloud's research on agentic AI adoption found that 88% of early adopters report a positive return on investment on at least one use case. That's a strong number, and it suggests that once a project clears the production hurdle, it's more likely than not to actually deliver value, rather than sitting there as an expensive experiment nobody's willing to shut down.
This is useful context if your own project is past launch and you're trying to judge it fairly: the base rate for production AI projects delivering real ROI is high. If yours isn't, that's worth investigating specifically, rather than assuming it's just how these things go.
Here's the part that complicates things: knowing whether something is working and being able to actually measure and prove it are two different problems. Morgan Stanley Research found that the share of North American AI adopters citing quantifiable AI impact nearly doubled, from 16% in the fourth quarter of 2024 to 30% in the fourth quarter of 2025. That's genuinely fast progress. It also means that even now, roughly seven in ten companies with active AI adoption still can't point to a specific, quantifiable measure of impact.
If your business is in that majority, quietly assuming your project is fine because nobody's complained isn't the same as actually knowing. The gap between "seems fine" and "quantifiably fine" is exactly where most companies currently sit, and closing it is worth deliberate effort rather than something that happens automatically over time.
“The real question isn't whether your AI project feels like it's working, it's whether you have a specific, quantifiable answer you could give someone who asked you directly. If you don't, you're in the same position as roughly seventy percent of companies with active AI adoption, not a special case.”
A useful starting point is picking one specific, measurable outcome tied to why the project was approved in the first place:
Then checking whether you actually have a before-and-after number for it. If you do, you're already ahead of most companies on this specific question. If you don't, that's the actual gap to close, not a bigger rollout or a flashier feature, just a clear, specific measurement of the thing the project was supposed to improve.
If you're trying to get a clearer, more specific read on whether your customer-facing AI is actually delivering, that's worth a direct look rather than a guess. SalesAPE offers a free demo if you'd like to see what specific, measurable outcomes look like in practice, no pressure either way.
According to Deloitte, only 25% of AI experiments have moved into production to date. The majority remain in pilot or testing stages, so simply having a project live in production already puts a business ahead of most others attempting similar initiatives.
Often, yes. Google Cloud's research found that 88% of agentic AI early adopters report positive ROI on at least one use case, suggesting that once a project clears the production hurdle, strong results are the more common outcome rather than the exception.
Measuring impact and knowing something feels like it's working are different skills. According to Morgan Stanley Research, only 30% of North American AI adopters could cite quantifiable AI impact as of late 2025, up from 16% a year earlier. Most companies still lack a specific, measured answer, even when adoption itself is going well.
Pick one specific, measurable outcome tied to the original reason the project was approved, time saved, response rate, conversion rate, and check whether you have an actual before-and-after number for it. That single, concrete measurement matters more than a broader sense that things seem fine.