A 2027 AI budget works best as a small, recurring line item aimed at one measurable workflow — not a big one-time project fund spread across departments. Research on enterprise AI shows broad rollouts fail 80–95% of the time. Narrow, deeply integrated systems are the exception that actually pays back — and that's the model worth budgeting for.
It's budget season, and most of that money is about to be wasted
Fall is when a lot of businesses — from Fortune 500s down to a five-person shop in Kalispell — sit down and decide what next year's tools and systems spend actually looks like. At the enterprise end, that conversation increasingly means a bigger AI line item. The problem: the research on how that money performs is not good. RAND Corporation, interviewing 65 data scientists and engineers across more than 50 organizations, found that more than 80% of AI projects fail — roughly twice the failure rate of comparable IT projects that don't involve AI (RAND, 2024).
MIT's Project NANDA went further. Their July 2025 study — built from a review of 300+ AI initiatives, 52 structured interviews, and 153 survey responses from senior leaders — found that only about 5% of custom enterprise AI tools ever reach production with a measurable return. The other 95% stall out (MIT NANDA, 2025). That's not a small-business problem. That's what happens when companies with real budgets and dedicated staff try to do this.
What's actually failing isn't AI — it's the size of the swing
The MIT NANDA researchers didn't find that AI doesn't work. They found a specific pattern in the 5% that did: those tools got embedded into one real, high-volume workflow deeply enough to retain feedback and improve over time, instead of being bolted on top of a dozen departments at once (MIT NANDA, 2025). The failures, by contrast, tend to be broad by design — a platform rollout meant to touch sales, support, marketing, and ops simultaneously, measured (if at all) by adoption rather than by any single number that moves.
That's a budgeting mistake before it's a technology mistake. A budget built around "AI transformation" as a category has no natural stopping point and no obvious way to know if it worked. A budget built around "answer every inbound call" or "follow up on every quote within an hour" has both.
Small businesses aren't the ones getting this wrong — yet
Here's the part that should change how a small operator reads all of the above: most small businesses aren't overspending on AI. A 2026 survey of small and mid-sized business owners found that a third spend nothing on AI tools at all, 28% spend $25–$99 a month, 16% spend $100–$249, and only 10% spend $250 or more (Bluevine, 2026). The same report notes that when small business owners cite cost as a barrier, it's rarely about the price of getting started — it's uncertainty over whether going deeper, with more integration into how the business actually runs, is worth the money (Bluevine, 2026).
That instinct — start small, stay skeptical of the platform pitch — happens to be exactly what the RAND and MIT NANDA research says works. The risk heading into 2027 budget season isn't that Montana and Northwest small businesses spend too little on AI. It's that budget-planning conversations, pushed by the same "transformation" language enterprises use, talk them into a bigger, broader line item than the narrow approach that was already working.
| Broad "AI transformation" budget | Single-workflow AI budget | |
|---|---|---|
| Scope | Multiple departments/tools at once | One high-volume workflow (e.g. phones, follow-up) |
| Time to see results | Months to years, often never | Days to weeks |
| How success is measured | Adoption, license usage | A number that moves: calls answered, jobs booked |
| Typical outcome | Stalls with no P&L impact (MIT NANDA, 2025) | Live system, immediate before/after comparison |
Where a Montana operator's next AI line item should actually go
Early September is a real planning window for a lot of Flathead Valley and Northwest service businesses — the summer tourist and building season is winding down, next year's budget isn't locked yet, and there's a little breathing room before winter work (or, for some trades, the spring rush) demands full attention. That makes it the right time to set a 2027 AI line item — but the research above is a strong argument for keeping it narrow.
Concretely: pick the single workflow with the clearest, most countable cost of failure — usually the phone, since a missed call is a lost job with a dollar figure attached — and budget for that one thing, live and measurable, before adding anything else. That's a smaller number than a "digital transformation" budget line, and it's the one the research actually says pays back.