I was in a meeting with a VP of Operations last week when she admitted something that made everyone in the room go quiet. Six months in, their AI tool was an $800K paperweight.
"We don't know how to use it," she said. "The documentation assumes we have engineers. We don't. And the vendor's onboarding is designed to sell, not teach."
She's not alone. Not even close.
The Numbers Are Brutal
A 2025 Benchmarkit study of 372 enterprise organizations found that 85% of companies miss their AI cost forecasts by more than 10%. Not a rounding error... a fundamental miscalculation of what it actually costs to get value from these tools. And 80% of enterprises miss their AI infrastructure forecasts by more than 25%.
Boston Consulting Group surveyed 1,000 C-level executives across 59 countries and found that only 26% of companies have the capabilities to generate tangible value from AI. The other 74% are stuck between proof-of-concept and actual results. And here's the kicker... about 70% of the challenges are people- and process-related. Only 10% involve the actual AI algorithms.
This isn't a technology problem. The tools work. The models are capable. The problem is everything that happens between signing the contract and getting business value... and nobody budgeted for that part.
You're Not Buying Software. You're Buying an Obligation.
Here's what most people miss about buying AI tools: you're not buying software. You're buying the obligation to rebuild how your team works, train people on new workflows, do deep integration work, and figure out all the things the marketing materials didn't mention.
Vendors price the software and assume you have people who can figure out the rest. Mid-market companies don't have those people. They don't have a team of ML engineers on staff. They don't have a data architecture group. They have smart operators who are already stretched thin running the business.
So companies buy the tool, pay for implementation, realize they need to rebuild the implementation anyway, and end up spending 3x what they budgeted. I see it happen every week.
Gartner found that through 2026, organizations will abandon 60% of AI projects that aren't supported by AI-ready data. And 63% of organizations either don't have or aren't sure they have the right data management practices for AI. The vendor never mentioned that during the demo, did they?
The Workflow Problem Nobody Talks About
McKinsey's 2025 State of AI report found something that should reframe how every company thinks about these purchases: only 21% of organizations using generative AI have redesigned any of their workflows. Nearly 80% are just layering AI on top of existing processes... like bolting a jet engine onto a bicycle.
And here's the part that matters: workflow redesign was the single strongest factor correlating with EBIT impact among the 25 attributes McKinsey tested. AI high performers were nearly 3x more likely to have fundamentally redesigned their workflows when deploying AI.
Think about that. The most expensive AI tool in the world won't move the needle if you're still running the same 12-step process you had before. The tool doesn't redesign your workflow for you. It can't. That's your job... and it's the job that nobody budgeted for, staffed for, or even realized was part of the deal.
What the Fast Movers Do Differently
The companies I see getting actual value from AI... they don't start with a tool purchase. They start with the problem.
Before they talk to a single vendor, they've answered three questions:
- What specific workflow are we trying to fix? Not "we want AI." That's not a strategy. Something like "we want to cut lead qualification time from 4 hours to 20 minutes" or "we need to reduce manual data entry in our operations pipeline by 80%."
- Who understands both the business side and how AI systems work? They put together a small team... maybe two or three people... that bridges the gap between the business problem and the technical solution. Those people might be internal. Might be external partners. But they exist before the tool purchase, not after.
- What does our data actually look like? Not the sanitized sample the vendor used in the demo. The real data. The messy, inconsistent, 14-years-of-accumulated-business-reality data that the AI will actually need to work with.
Then they pick the tool that fits how they work, not how the tool wants them to work. That distinction... it's the difference between the 26% getting value and the 74% wondering what went wrong.
The Hidden Cost Multiplier
Let's talk about what actually happens to the budget. You sign a contract for an AI platform. That's the number the board approved. But then:
- Data preparation: Your data isn't AI-ready. It needs cleaning, normalizing, deduplicating. Remember... Gartner says 63% of organizations don't have the data management practices for AI. That gap doesn't close itself. It closes with money and time nobody budgeted.
- Integration work: The pre-built connectors work for generic setups. Your Salesforce instance with 14 years of custom objects? That's custom integration work that no vendor demo ever accounts for.
- Training and change management: Your team doesn't have engineers. So someone needs to train them, build simplified interfaces, create documentation that actually makes sense for non-technical users. The vendor's onboarding won't cut it.
- Workflow redesign: The 80% of companies who skip this step are the ones sending me DMs about their $800K paperweight.
Add it up and you're looking at 2-3x the initial platform cost. That's not a worst case... that's the median outcome. The real cost of AI implementation is always downstream of the software license.
A Better Way to Spend That Money
If I had $800K and a mid-market company that wanted to get serious about AI, here's how I'd spend it:
First $100K: Discovery and strategy. Map the highest-value workflows. Audit the data. Identify the 2-3 use cases that will actually move the P&L. Build the small team... internal or external... that bridges business and AI. This phase eliminates the 60% of AI projects that fail because they started with the wrong problem.
Next $200K: Build the foundation. Data pipelines, integrations, workflow redesign. This is the unsexy work that makes everything else possible. No vendor demo ever shows this part because it's different for every company... and it's where the real value gets created.
Next $300K: Tool selection and implementation. Now you pick the platform. Except this time, you know exactly what you need because you've already mapped the workflows, prepared the data, and built the integration layer. The tool fits into a system that's ready for it... not the other way around.
Remaining $200K: Iteration and expansion. Measure what's working. Double down on the use cases that deliver. Kill the ones that don't. This is where real ROI compounds... not in month one, but in the ongoing refinement that turns a good implementation into a competitive advantage.
Same $800K. Completely different outcome. Because you're spending it in the right order.
The Question Nobody Asks Before They Buy
Before your next AI purchase, ask this: "Do we have the people and processes to actually use this thing?"
If the answer is no... and for most mid-market companies it is... that's not a reason to skip AI. It's a reason to start with the people and processes first.
The tool is the easy part. The transformation is the hard part. And the companies that figure that out before they sign the contract are the ones who don't end up with $800K paperweights.