Matt Hire

Closing the gap between AI hype and real results. The Human Side of AI newsletter ↓

[Matt Hire @MJHire]

Every enterprise I work with has a shadow AI problem.

Last month I ran an informal audit for a financial services client. 14,000 employees. Official AI tools deployed: 2. A document summarizer and a chatbot for HR questions.

Then we looked at browser extension data and expense reports.

340 employees had personal ChatGPT Plus subscriptions. 87 were using Claude. A dozen had Midjourney accounts. Two people in compliance were running sensitive customer data through a tool nobody in IT had ever heard of.

The CIO's reaction: "We need to shut this down."

My reaction: you need to understand why 400 people went around your official tools. They're telling you something. Your approved solutions are too slow, too limited, or too hard to access. They found something better and didn't wait for permission.

You can write a policy to stop it. Or you can treat it like the market research it actually is.

The companies getting this right are doing both. Light guardrails on data handling. And a fast track to evaluate whatever their people are already using.

Because if 400 employees independently decided your AI strategy wasn't good enough, maybe they're right.

[Matt Hire @MJHire]

A client automated their entire invoice processing workflow last year. Beautiful system. Cut processing time from 4 hours to 20 minutes.

Sixty days later, half the finance team was still doing it manually.

When I asked why, one of the senior processors told me: "The new system is faster, but when something goes wrong, I don't know how to fix it. So I run both."

She wasn't resisting change. She was managing risk the only way she knew how.

The implementation team had optimized for speed. Nobody had optimized for confidence. Nobody sat with the team and said "here's what to do when the system flags something weird."

Automation without enablement is just expensive shelfware.

[Matt Hire @MJHire]

Conversation I had with a CTO last Thursday. Paraphrasing but this is close.

Him: "We need an AI strategy."
Me: "What problem are you trying to solve?"
Him: "We need to figure out where AI fits."
Me: "Where is your team spending time on work that's repetitive and predictable?"
Him: "I'd have to ask my directors."
Me: "Have you asked them?"
Him: "Not yet. We wanted to get the strategy done first."

This is the loop I see in maybe 60% of the companies I talk to. They want a strategy for a technology before they've inventoried the problems the technology might solve.

It's like hiring an architect before you know how many people live in your house.

AI strategy doesn't start with AI. It starts with a Tuesday afternoon watching your operations team do their jobs. It starts with asking the person who's been doing the same task for 6 years what they'd automate if they could.

The strategy writes itself once you've done that work. Most companies skip it because it feels too simple.

[Matt Hire @MJHire]

The gap between "we're doing AI" and actually doing AI is almost always one person.

Someone who translates between what the model can do and what the business needs done. Who sits in the room with engineers and then walks down the hall and explains it to operations in plain language.

Every successful AI deployment I've seen in the last 2 years had this person. Most of them don't have "AI" anywhere in their title.

[Matt Hire @MJHire]

Something I got wrong early in my consulting career that took me years to unlearn.

I used to walk into AI engagements looking for the biggest, highest-impact process to automate. The logic seemed obvious: start with the process that touches the most revenue, the most people, the most transactions.

Three times in a row, those projects stalled. Org politics, unclear ownership, dependencies on legacy systems nobody wanted to touch.

The fourth time, I picked the smallest, most boring process I could find. Expense report categorization. Maybe 3 hours of someone's week.

It worked in 9 days. The finance team loved it. Word spread. Two months later I had budget approval for 4 more automations, including the big one I originally wanted.

I stopped selling transformation and started selling evidence. Small wins create believers. Believers create budgets. Budgets create transformation.

Every AI engagement I run now starts with the same question: what's the smallest thing we can finish in two weeks?

[Matt Hire @MJHire]

5 questions I ask in the first 30 minutes of every AI engagement:

1. Who owns this after the consultant leaves?
2. What did your team stop doing to make time for this?
3. How will you know this worked 90 days from now?
4. Has anyone asked the people doing the actual work what they need?
5. What happens if this succeeds and you need to scale it?

The answers tell me more than any technical assessment ever has.

[Matt Hire @MJHire]

Myth: you need clean data before you can do anything with AI.

I hear this from executives every single week. "We're not ready. Our data is a mess."

Here's what I've learned after 15+ engagements where "dirty data" was the stated blocker:

In 11 of those 15, the data was fine for the use case they actually needed. They were measuring their CRM data against some imagined standard of perfection that had nothing to do with the problem.

In 3 of them, the data issue was real but fixable in under 2 weeks with a focused cleanup sprint.

In 1 case, the data was genuinely unusable. One out of fifteen.

"Our data isn't ready" has become the socially acceptable way to say "we don't know where to start." And it lets leadership delay AI adoption indefinitely without ever looking like they're against it.

If someone on your team keeps saying the data isn't ready, ask them: ready for what, specifically? That question usually changes the conversation.

[Matt Hire @MJHire]

The scariest meeting in AI transformation is the one where the CFO asks "what did we get for this?"

Most teams can't answer it. They have usage dashboards. Login counts. Tokens consumed.

None of that tells you whether the $400K you spent changed a single business outcome.

The companies that survive that meeting are the ones who tied their AI work to a metric the CFO already cared about before AI entered the conversation. Revenue per rep. Days to close. Cost per claim.

If you can't connect AI spend to a number that was already on the P&L, you're going to lose that budget.

[Matt Hire @MJHire]

Two manufacturing clients. Same industry. Similar revenue. Both started AI projects in Q1 2025.

Company A hired a consulting firm, built a 90 day roadmap, evaluated 6 platforms, and launched a predictive maintenance pilot on their most complex production line.

Company B asked their maintenance team what broke most often. The answer was a conveyor belt bearing that failed every 11 weeks. They stuck a $200 vibration sensor on it and connected it to a basic anomaly detection model.

Company A is still in pilot. 14 months later. They've spent $380K and the model accuracy isn't where they want it.

Company B has saved $94K in unplanned downtime. They've expanded to 6 more failure points using the same approach. Total spend: $12K.

The difference between these two companies has nothing to do with technology. Company B started with a problem. Company A started with a platform.

[Matt Hire @MJHire]

Most AI strategies are really vendor strategies.

Someone from sales gave a great demo. Leadership got excited. A "strategy" formed around that product's capabilities.

Then 6 months later the org wonders why adoption is low.

You built your strategy around what a tool can do instead of what your team actually needs done. That is vendor strategy wearing an AI strategy costume.

I see this in about 70% of the companies I walk into.

[Matt Hire @MJHire]

The worst AI implementations I've seen had the best business cases.

127 page ROI analysis. Executive steering committee. Deloitte on speed dial.

Meanwhile down the hall, a claims processor named Dave built a macro that auto-categorized 400 daily emails using GPT. No business case. No vendor. Took him a weekend.

Dave's thing has been running for 14 months. The $1.2M "enterprise AI initiative" got shelved after the vendor missed their third integration deadline.

I see this pattern constantly. The projects with the most governance produce the least value. The ones with the least oversight produce something people actually use.

I'm not saying skip governance. I'm saying if your governance process takes longer than your implementation, something is backwards.

[Matt Hire @MJHire]

A client spent 6 months evaluating AI vendors. Sat through 23 demos. Built a 40 page comparison matrix.

Never ran a single pilot.

When I asked why, the VP of ops said "we wanted to make sure we picked the right one."

They were so afraid of choosing wrong that they chose nothing. Meanwhile their competitor automated invoice processing in week 2 with a tool that cost $200/month.

Perfect is the enemy of deployed.

[Matt Hire @MJHire]

The most underrated metric in AI transformation: time from insight to action.

A logistics company I worked with had an AI system flagging inventory anomalies 72 hours before stockouts.

The alerts went to an inbox that three people checked once a week.

72 hours of lead time. Compressed to zero by a process that nobody redesigned when the AI went live.

The model worked perfectly. The org chart made it irrelevant.

[Matt Hire @MJHire]

Sat in a leadership offsite last month where the CEO spent 45 minutes on the AI roadmap.

Powerful slides. Clear timelines. Ambitious targets.

Then a director raised her hand and asked: "Who's going to train my team?"

The room went quiet for about 8 seconds. Then someone said "we'll figure that out in phase 2."

Phase 2 is where every AI roadmap hides the hard part. The technology ships in phase 1. The behavior change gets deferred indefinitely.

[Matt Hire @MJHire]

I keep a running list of questions clients ask me about AI.

90% of them are about the technology.
10% are about their people.

The 10% are the only ones that matter.

"Will the model hallucinate?" matters less than "will your team know when it does?"
"Can it process 10,000 documents?" matters less than "does anyone know what to do with the output?"
"Is it secure?" matters less than "do your people trust it enough to actually use it?"

The technology questions have answers. The people questions have organizational design problems. And those take longer than any implementation.

[Matt Hire @MJHire]

The AI talent war is already over and most companies lost without realizing it.

They posted job listings for "AI engineers" and "ML specialists" while the actual skill gap was somewhere else entirely.

The person they needed was the operations manager who could look at a process, identify what data flows through it, and articulate what a better version would look like.

Every company I talk to has engineers who can build AI. Almost none have operators who can tell them what to build.

The highest leverage AI hire in 2026 is someone who has never written a line of code but has run a department long enough to know where the real waste lives.

[Matt Hire @MJHire]

A board member asked me last quarter what their "AI readiness score" was.

I told him the score was meaningless. Then I asked three questions:

How many decisions did your team make last month using AI generated analysis?
How many of those decisions turned out better than what they would have done without it?
How many people on your team could explain how the AI reached its recommendation?

He answered the first one. The other two got silence.

That silence is the gap between buying AI and actually being ready for it. Most companies can tell you what they deployed. Almost none can tell you whether it changed a single decision for the better.

[Matt Hire @MJHire]

Everyone is debating whether AI will replace jobs.

Meanwhile I'm sitting in rooms where AI created 3 new roles that didn't exist 18 months ago.

One client now has a "workflow translator" who sits between the AI tools and the operations team. Another hired two people just to manage the data pipelines feeding their models. A third created an entire internal consultancy for AI adoption support.

The real pattern: AI doesn't eliminate headcount. It reshuffles where the expertise needs to sit. And right now, most companies haven't figured out where that is.

[Matt Hire @MJHire]

Every company has someone who "gets AI."

The person who built the first prototype. Set up the pilot. Got the team excited.

Then they get promoted. Or leave. Or move to a different project.

6 months later, nobody remembers how the system works. The documentation is a Confluence page from March with 3 broken screenshots. The vendor contract auto-renewed but nobody knows who owns it.

AI adoption doesn't fail at launch. It fails at the second handoff, when the person who built it is no longer in the room.

[Matt Hire @MJHire]

A client called me last Tuesday. Their AI agent had been running for 3 weeks.

"It keeps giving wrong answers about our returns policy."

Turns out their returns policy lived in 4 different documents, none of them matching what the team actually does. The agent was reading the official doc. The team was following a process nobody ever wrote down.

The fix took 2 hours. One conversation with the returns team, one updated doc, one re-indexed knowledge base.

The AI was never broken. The context was.