Here's a number that should stop you cold: 42% of companies abandoned most of their AI initiatives in 2025 — up from 17% the year before, according to S&P Global Market Intelligence. The failure rate more than doubled in twelve months. And yet budgets kept growing, vendors kept selling, and boards kept asking for updates. Something isn't adding up.
Meanwhile, a small group — roughly 5% — are pulling further ahead every quarter. The gap between them isn't the technology they chose or the size of their AI budget. It's whether they understood their processes before they started. Stay with me, because I know that sounds like the boring answer. It is. That's the point.
I've written about the failure pattern before — how the hype cycle keeps repeating itself across technology waves, and how the mid-market actually has a structural speed advantage over large enterprises if they move with intention rather than urgency. (If you missed those, "They Did Everything Right" and "Groundhog Day?" cover the ground.) This piece is about what to actually do.
Process intelligence is not what vendors are selling
There's a category of enterprise software built around this problem. Process mining platforms promise a data-driven picture of how your operations actually function — pulling event logs from your ERP, CRM, and core systems and surfacing where things break down. The pitch is compelling. The reality is that the leading platforms run north of $280,000 a year at the enterprise level, with first-year total costs — software, implementation, integration, and the internal resources to sustain it — routinely landing between $470,000 and $1.5 million.
And even then, you get an incomplete picture. Those platforms see what your systems record. They miss everything that happens in spreadsheets, email threads, and the judgment calls your most experienced people make without logging a thing. The workarounds. The exceptions. The "we've always done it this way because" moments that define how work actually flows in most organizations.
That gap is where most AI deployments fail. In "The $700 Billion Smokescreen," I made the case that AI ROI doesn't come from what you spend on infrastructure — it comes from whether the operational foundation underneath it is solid. Process intelligence is that foundation. And you don't need a seven-figure platform to build it.
The performance gap is widening. BCG's study of more than 1,250 firms found that only 5% of companies are achieving AI value at scale — while 60% are generating minimal value despite substantial investment. The companies in that top 5% generate 1.7x more revenue growth and 1.6x higher EBIT margins than laggards. The common thread: they treat workflow redesign as a prerequisite, not an afterthought.
What you actually need
McKinsey found that AI high performers are nearly three times more likely to fundamentally redesign their workflows — and that workflow redesign has one of the strongest contributions to meaningful business impact of all the factors they tested. Not model selection. Not data infrastructure. Workflow redesign. Which means you need to understand your process before you touch the technology.
The most underrated AI readiness move available to any mid-market company right now is a well-facilitated process walkthrough. Bring your process owners and the people who actually execute the work into the same room. Walk the process from end to end — inputs, handoffs, decision points, exceptions, and the unofficial fixes that keep things moving. Document what you hear in a way that's structured and usable, not just a wall of sticky notes.
Tools like Miro make this faster and more collaborative than anything available five years ago. A session that used to take weeks of manual diagramming can now be captured, structured, and turned into a working process model in days. AI-assisted transcription and summarization compress it further. The methodology hasn't changed much. The time it takes has.
The output isn't just documentation. It's a targeting system. It tells you exactly where in your process AI can move a number — and where dropping a tool into a broken handoff will just automate the mess you already have.
The companies getting this right treat it as infrastructure
The difference between the 5% pulling away and the 60% going nowhere isn't that the winners ran one good process mapping session. It's that they treat process knowledge as a living asset. There's a standard for how processes get assessed, documented, and kept current. Not a million-dollar Center of Excellence. A discipline — a repeatable way of staying honest about how work flows and where AI can change the outcome.
Most organizations don't have this. They have documentation that reflects how processes were designed, not how they're actually run. They have tribal knowledge that lives in the heads of their longest-tenured people. And they have AI pilots that couldn't scale because nobody could agree on what the process was actually supposed to do.
The companies winning with AI fixed that first. Not because it was glamorous. Because it worked.
A mid-market client came to us ready to automate a specific part of their operations. They had a use case, a budget, and a vendor shortlist. What they didn't have was a complete picture of the process they were about to touch.
When we mapped it end to end — pulling in stakeholders from departments they hadn't originally included — a different story emerged. The step they wanted to automate wasn't actually the problem. It was a symptom. The real friction lived two handoffs upstream, in a gap between teams that nobody owned and everyone worked around.
Automating the original target would have made one small piece faster while leaving the actual bottleneck untouched. The broader mapping exercise took more time and required more stakeholders than the client initially wanted to invest. It was worth every hour. The decision they made after it was fundamentally different — and so was the ROI.
Where to start
Pick one process — ideally one with high volume, high manual effort, or a pain point your team has been working around for years. Map it the way it actually runs, not the way it was designed. Get the people who do the work in the room, not just the people who manage it. Document the handoffs, the exceptions, and the workarounds. Then ask, with fresh eyes, where AI could change a specific outcome — cycle time, error rate, decision quality — in a way you can measure.
That's the starting point most organizations skip on their way to purchasing a platform.
None of this sounds exciting. It doesn't belong on a conference keynote slide. It won't win a digital innovation award.
But the data is unambiguous: organizations that take the time to genuinely understand their processes before deploying AI dramatically improve their odds of realizing the return they set out to achieve. Not marginally. Dramatically.
You can do this. It takes time, not millions. It takes your people, not a platform. And it's the single highest-leverage thing most mid-market companies aren't doing right now.
Be boring. Understand your process. The ROI follows.
Few tokens were harmed in this writing. ✌️