140 Conversations on Mid-Market AI Adoption

Pattern 1: Data Scattered Across 3-7 Tools (And Nobody Knows Where the Truth Is)

A $14B wealth advisory firm runs 90% of their business through email. Outlook is the system of record.

A construction CFO uses spreadsheets for budgets, WhatsApp for site updates, and paper forms for change orders. When I asked where the actual cost data lives, she paused. "Everywhere. Nowhere. I don't actually know."

A packaging manufacturer operating in 4 countries has product specs in CorelDRAW (design software), pricing in a legacy system, and customer relationships scattered between email and no structured CRM.

A wellness brand was running on Shopify, Systeme.io, WhatsApp, Calendly, Stripe, and three other tools at once. No single place to see which customers existed, what they'd bought, or how to reach them.

This is not a technology problem. This is the default state of growing a company manually. The fix usually starts with data consolidation before AI.

Pattern 2: CEOs Get Visibility by Asking People, Not Looking at Data

"I don't know if we're making or losing money until the project is done."

That's a construction CFO talking about a three-month engagement. The numbers exist somewhere, invoices, timesheets, purchase orders, change orders. But they live in different places, entered by different people, updated on different schedules. The CFO has to call the project manager. The project manager has to check with the site supervisor. The site supervisor has to check with the crew leads.

Three weeks later, she has an answer.

This happens in every mid-market company I've worked with. Data exists. Visibility doesn't. So leadership makes decisions on 30-day-old information, or gut feel, or whatever the loudest person said in the last meeting.

Pattern 3: Someone Bought an AI Tool That Nobody Uses

The Microsoft Work Trend Index found that 78% of employees bring personal AI tools to work. That's because the company's official AI or productivity tool doesn't solve their actual problem.

In one conversation, a legal firm had subscribed to an advanced legal research tool. Expensive. Barely used. The lawyers were using ChatGPT anyway because it was simpler and they already had an account.

In another, a construction company bought a project management platform with "AI insights." Nobody logged into it. The crews stayed on their existing system. The insights never got made.

Buying the tool is easy. Making it the default is hard. Making it matter is harder.

Pattern 4: "We Don't Know Where to Start", The #1 Trigger for Moving Forward

I've heard this phrase in six separate conversations, verbatim. Sometimes it's phrased differently. "We need someone to tell us what to do first," or "We know AI is important but we're stuck," or "Everyone's doing it and we're not and it's weird", but the underlying anxiety is the same.

And here's what's interesting: this is the moment people actually move forward. Because "we don't know where to start" is different from "we're not interested." It's a gap between intent and action.

The companies that move forward from here are the ones that get clarity first. An audit. A conversation. A roadmap. Something that says, "Here's the first thing you do, and here's why." This is also why mid-market companies fail at AI: they skip the clarity step and buy tools before they know what to fix.

Pattern 5: Phase-Based Buying Works. Big Programs Don't.

Not a single "big transformation program" has closed. Zero.

Every deal that actually happened started small.

A construction company started with AI-powered document search (Capataz) to find cost data faster. First month: problem solved. Second month: the team asked if they could track costs in real-time. Third month: full platform. Fourth month: integration with their other systems. Still ongoing, still adding value, still owned by the client.

A wealth advisory firm started with an audit of their data environment. Small project, low risk. Findings justified a content and visibility strategy. That led to an ongoing retainer. Now they're on the roadmap for custom AI tools.

A nonprofit consultant started with a Claude workspace setup and document organization. A small fixed-fee engagement. Two weeks later they asked about workshops for their networks.

The phase-based pattern is: start small, prove value, let the next phase sell itself.

Across all 140+ conversations, the ones who actually decided to act had four things in common.

They start with data, not tools. They asked, "Where does our information live?" before asking, "Which AI tool should we buy?" This takes 2-4 weeks to map out as a fixed-fee engagement. And it's the difference between implementation that fails and implementation that sticks.

They pick one problem with measurable ROI. Not "AI transformation across the organization." One problem. One metric that matters. A construction company measured time saved finding cost data. A law firm measured response time to client questions. A manufacturer measured days from order to production clarity. One problem with a number attached means you can actually prove AI worked.

They insist on ownership. No lock-in. No vendor dependency. The companies that move fastest are the ones that say, "We own what you build, and we can maintain it ourselves or with someone else." This removes the fear of getting stuck.

They start small and let results justify the next phase. Nobody I know who said "let's do everything" actually did everything. The ones who said "let's do this first phase, see what happens" are the ones who are six months in, happy with the results, and planning the next wave.

This is the question that comes up in every conversation eventually.

AI Ops Audit: Fixed fee, 2-4 weeks. You get a map of where your data lives, what's working, what's stuck, and a prioritized roadmap. This is the starting point that works.

Foundation Build: Fixed-fee, 4-16 weeks depending on complexity. This is the implementation. You're building dashboards, automating processes, connecting tools, training people. You own the code. You own the data. You keep it.

Ongoing AI Operations (Retainer): Fixed-fee monthly retainer. This is fractional Head of AI. New tools get evaluated, new processes get designed, your team gets trained, emergencies get handled. Ongoing strategy plus execution.

Here's the math: a Chief AI Officer is expensive once you factor in salary, benefits, and recruiting. The retainer model is a fraction of that, and the retainer only happens after you've seen results in the first two phases. The full breakdown is in what AI consulting actually costs for a 20-200 employee company.

One more thing. If you build custom tools or automate processes, your company likely qualifies for R&D tax credits. IRC Section 41. 65% of contractor payments qualify as Qualified Research Expenses. On a $50K AI build, that typically translates to roughly $4,550 in federal credit (Year 3+ ASC rate) plus Section 174A deduction value, together offsetting 18-23% of the build cost. The credit also works retroactively: you can amend returns for the prior 3 tax years. Worth asking an R&D tax specialist about before you build. Not tax advice, verify with your CPA.

The patterns hold across industries, but the specifics vary.

Construction companies operate on data that lives in email, spreadsheets, and job site notebooks. The first win is always cost visibility, knowing real-time spend, not finding out when the project is done. One company caught a six-figure cost overrun in month two because the system surfaced data they already had but couldn't see. Second win is always document intelligence, architects, estimators, and crew leads ask questions about specs and compliance in real-time instead of asking people. Third wave is full project visibility: labor, materials, schedule, risk, all in one place.

Legal firms operate on client relationships and case timelines. The first win is research speed, answers to procedural questions in seconds instead of hours. Second is client communication, systematic responses to common questions instead of email thread chaos. Third is capacity, senior partners doing client strategy instead of document review.

Financial services (wealth advisory, asset management) operate on client information and portfolio strategy. The first win is data consolidation, one system that knows what they own instead of asking advisors. Second is reporting, custom client narratives instead of templates. Third is planning, running scenarios and showing impact, not just historical returns.

Manufacturing and distribution operate on product information, pricing, and inventory. The first win is always product knowledge, customers and sales teams asking what's in stock, what's compatible, what's available in volume. Second is operations, orders flowing into production without manual data entry. Third is omnichannel, inventory, pricing, and customer data talking to each other.

The sequence matters. You can't automate what isn't visible. You can't optimize what isn't tracked. You can't scale what still needs human judgment for every decision.

I'm not saying AI is new. I'm saying the bottleneck has moved.

Two years ago, the barrier was "can we build this at all." Now the barrier is "how do we start without breaking everything."

Two years ago, companies didn't know which tools existed. Now they're overwhelmed by tools and need help connecting them.

Two years ago, AI was a competitive advantage. Now it's the default, and the companies without it are noticeably slower.

The companies moving forward are the ones who stopped waiting for the "right moment" and started with the right first step. An audit. A conversation. A small win that proves the larger thesis: your company has more information than you're using. Your people are spending time on things machines could handle. And once you see it, you can't unsee it.

That's the point of the report I put together. 140+ conversations. Real patterns. Real findings. Industry-specific breakdowns. ROI framework. And a self-assessment so you can ask yourself the same questions your peers are asking.