The technology works. The hype doesn’t. The gap between them is where the money goes.

AI is having its moment, and companies are spending real money to be part of it. Most of that money isn’t coming back.

An MIT study released in 2025 looked hard at enterprise AI and found that about 95% of generative AI pilots delivered no measurable return. None. That’s not a knock on the technology, which is genuinely impressive. It’s a knock on how the technology is being deployed. The companies aren’t failing because AI doesn’t work. They’re failing because they aimed it at the wrong things.

Why most of them miss

The same MIT research turned up a telling detail. More than half of AI budgets are going into sales and marketing tools, the flashy front-of-house stuff, while the biggest actual returns showed up in back-office automation. The unglamorous work. Cutting manual processes, reducing outside agency and outsourcing costs, cleaning up operations. The boring wins were the real wins, and most companies were spending everywhere except there.

There was a second finding worth repeating. Buying from specialist vendors and building partnerships worked about 67% of the time. Building everything in-house succeeded only about a third as often. The lesson is simple. You don’t have to build it all yourself, and unless this is your core business, you probably shouldn’t.

What the five percent do

The companies that win with AI tend to do the same handful of things. They start with a real problem instead of a tool they wanted an excuse to use. They pick something measurable, so they actually know whether it worked. They get their data in order first, because a smart model on top of bad data is just a faster way to be wrong. And they bring in people who have shipped this kind of work before, instead of learning on the most important project they’ve ever run.

They also start small and earn their way up. There’s a wide gap between a pilot that looks great in a demo and a system people actually rely on every day. The winners pick one narrow workflow, get it working in the real world, prove the number moved, and then expand from that proof. The losers try to boil the ocean with a giant transformation program and run out of patience and budget before anything ships.

And they don’t forget that AI is a people change as much as a technology one. A tool nobody trusts or knows how to use just sits there. The teams that get value bring their people along, show them what the tool does and doesn’t do, and build a little trust before they bet the process on it. That part has nothing to do with the model and everything to do with whether the investment pays off.

How we do it

Our AI Studio is built for that approach. On-demand data scientists, machine learning engineers, and AI developers who plug in and solve a defined problem. This isn’t theory for us. We helped a global manufacturer cut new product timelines by 24% and built recommendation chatbots for electronics retailers. Real work, measurable results.

And for smaller companies that want the back-office wins without an enterprise budget, we run newgig.ai, focused on the practical stuff: automating the paperwork, the document handling, and the repetitive sales and marketing tasks that eat a small team alive.

If you’re wondering where to even start, look at the work that’s high volume, rule-based, and quietly draining hours from your team. Invoice and document processing. Sorting and routing incoming customer requests. First-pass review of contracts or records. These aren’t glamorous, which is exactly why they tend to pay off. The work is repetitive enough that a machine can help with it, frequent enough that the savings add up, and contained enough that you can measure whether it actually worked. Start there, prove it, then move on to the next one.

The bottom line

AI works. The hype around it mostly doesn’t. The companies pulling ahead aren’t the ones with the biggest models or the loudest press releases. They’re the ones who picked a real problem, got the boring parts right, and brought in people who’d done it before. That’s the whole game.

Sources

1. MIT NANDA, The GenAI Divide: State of AI in Business 2025 (via Fortune): https://fortune.com/2025/08/18/mit-report-95-percent-generative-ai-pilots-at-companies-failing-cfo/