When a $100M+ company finally commits to AI, the instinct in the room is to swing big. The CEO wants the customer-facing copilot that shows up on the earnings call. The board wants the moonshot that reprices the whole category, and almost every time we've watched that instinct win, the program is quietly dead inside a year.That's not a hunch. MIT Project NANDA's 2025 study, The GenAI Divide, found that 95% of organizations are getting no measurable return on their generative AI spend, even after $30 to $40 billion poured into it. The companies on the right side of that divide tend to start somewhere far less exciting. They start with the project nobody would ever put in a press release.
The boring use case is the one that survives
The first AI workload that reaches production at a large enterprise is usually something like invoice processing or internal ticket routing. It's unglamorous, it's internal, and that's exactly why it works. When the blast radius is contained inside one team, a bug is an annoyance instead of an incident, and the people who built it can fix it on a Tuesday afternoon without anyone in legal or communications ever hearing about it.That containment is the entire point. A first project exists to prove one thing, that the company can ship AI at all, somewhere a stumble costs you a week instead of a reputation. Winning the year can come later.
Ask whether it can be killed quietly
The test we give clients before they greenlight anything is a single question: can this be killed quietly? If the project fails, does it fail inside the building, or does it fail on a customer's screen and land in a Slack channel the CEO reads?That question matters even more in the agent era. Gartner expects more than 40% of agentic AI projects to be canceled by the end of 2027, citing escalating costs, unclear business value, and weak risk controls. A use case you can shut down without a public apology is one you're allowed to learn from. The discipline lives in choosing the project whose downside you can absorb, even when a smarter, riskier one is sitting right next to it.
What a clean first win actually buys you
The reason the boring project matters has almost nothing to do with the project itself; instead, it's political capital. A focused first win that returns something a CFO can defend, in months rather than years, earns the program the runway to fund the ambitious work in year two.That's the sequence behind the results that eventually get written up. Avalara, the tax-compliance company Vista Equity Partners took private for $8.4 billion, scaled to 179 AI agents across 8 departments, with 84,176 hours saved and a 7x return. That scale wasn't approved in a single board vote. It was earned one defensible win at a time, until the budget for the bigger swings became easy to say yes to.
The failure that takes the whole program down
Leading with the flashy project puts more than that one project at risk. A visible AI miss in month four can freeze the entire AI program for the next 18 months while the organization quietly decides AI "isn't ready for us yet."That outcome is far more common than the success stories suggest, and MIT's researchers traced the cause to how the systems get integrated into real workflows, not to the models themselves. The teams that stall almost always picked something too visible, too early, and spent their credibility before they'd built any.
Pick the project your CFO could defend
So the right question for your first AI project isn't "what would impress the board." It's "what would my CFO feel comfortable defending if we had to shut it down next quarter." That project is almost always the boring one, and the discipline to choose it is the same discipline that keeps the program alive long enough to do the exciting work later.Boring first, ambitious second. In that order, AI tends to survive contact with the enterprise.If you're deciding where to point your first AI project, take our AI assessment and we'll help you find the use case that wins quietly.
