Lumen Technologies and Air Canada deployed AI in the same era, using the same underlying technology. One created $50 million in annual value. The other handed a grieving customer a liability and ended up in a tribunal. The difference wasn’t the AI. It was the playbook. Or the lack of one.
Lumen had a problem every mid-market company would recognize: their sales team was spending four hours researching each potential client — jumping between Salesforce, ServiceNow, Gainsight, and internal documentation systems. Multiplied across 3,000 sellers, that fragmentation was costing them millions. Their fix wasn’t a better model or a bigger budget. They gave their AI clear instructions for handling different sales scenarios, connected it to their actual customer data across systems, and gave it the ability to generate presentations and automate reports. That four-hour research process now takes 15 minutes.
Now consider Air Canada. They deployed a chatbot to help customers with common questions. A man whose grandmother had just died asked about bereavement fares. The chatbot confidently told him he could book at full price and claim a discount retroactively within 90 days. He did exactly that. When he submitted his claim, Air Canada denied it — their actual policy explicitly prohibits retroactive bereavement applications. The chatbot had never been given access to the real policy. A Canadian tribunal held the airline liable.
The Real Problem
Here’s what the data tells us: adoption isn’t the issue. RSM’s 2025 middle market survey found that 91% of mid-market companies have adopted generative AI. But 92% of those companies encountered significant challenges during rollout, and more than half reported feeling only “somewhat prepared” to implement it.
- 91% of mid-market companies have adopted generative AI
- 5% of AI pilots are delivering measurable P&L impact (MIT NANDA)
- 56% of companies are getting zero measurable return from AI investments (PwC)
These aren’t technology failures. They’re strategy failures. And the available frameworks for fixing them assume you’re a Fortune 500 company with a dedicated AI team, a multi-million-dollar transformation budget, and a 12-month roadmap.
If you’re running a $150 million manufacturing company or managing a portfolio of mid-market businesses, those frameworks don’t fit. You don’t have 50 data scientists. You probably don’t have one. And here’s the thing: you might not need one.
The same MIT report found something that should give mid-market leaders real confidence: when mid-market companies do have the right approach, they convert pilots to full implementation in about 90 days. Enterprises take nine months or longer. The mid-market’s advantage is speed and decisiveness. What’s been missing is a method built for that advantage.
Three Questions That Separate the 5% from the Rest
The companies that succeed with AI — whether they’re Lumen or a 200-person regional bank — answer three questions in sequence. Skip one, or answer them out of order, and you end up in the 95%. Each deserves a deep dive (and will get one later in this series), but here’s the overview.
Question 01: Where is AI already creating value in your company?
Most companies with more than 50 employees already have people using AI tools for work — many without telling anyone. MIT’s research found that while only 40% of companies have purchased an official AI subscription, workers at over 90% of companies reported regular use of personal AI tools. That gap between official adoption and actual usage isn’t a governance crisis. It’s the best free market research a CEO will ever get. Your employees have already been running experiments. The first step isn’t to buy anything. It’s to find out what’s already working.
Question 02: Which of those opportunities are actually worth pursuing?
Not every task benefits from AI, and the difference between companies that waste money and those that save millions often comes down to whether they aimed for 80% or 100%. At Areté, we recently worked with a contract manufacturer struggling with excess inventory. We helped them cut their average balance from roughly $35 million to under $25 million in under a year — not with a new ERP or a demand forecasting overhaul, but with an AI system that identified finished goods missing only one, two, or three raw ingredients, then prioritized purchase orders for those ingredients. The scoring principle is straightforward: tasks that are repetitive, data-intensive, and easy to verify are strong AI candidates. Tasks that require high emotional intelligence are not.
Question 03: Does the AI have what it needs to actually succeed?
This is where Lumen got it right and Air Canada got it wrong. Every AI deployment needs three things: clear instructions (what to do and how), the right background information (company policies, customer data, historical context), and the ability to complete the workflow — not just analyze, but act. Lumen gave their AI all three. Air Canada’s chatbot had instructions but no access to the actual bereavement policy, and no ability to verify its own answer against the source document. Most AI failures trace back to a missing element here. Fix the gap before investing in a better model.
What to Do This Week
If you’re a CEO or operating partner evaluating AI across your business, here are two things you can do in the next five days.
1. Run a 10-person discovery conversation
Ask people across different functions a simple question: “Where are you already using AI tools — officially or not — and what’s working?” Don’t make it a compliance exercise. Make it a strategy conversation. The answers will tell you more about where AI fits your business than any vendor demo.
2. Audit one active AI initiative against the three elements
Pick whatever AI project is currently underway — or recently stalled — and ask: Does it have clear, specific instructions? Does it have access to the information it needs? Can it complete the full workflow, or does a human have to finish the job? If you can identify the missing piece, you’ve found your bottleneck. And it’s almost certainly cheaper to fix than to replace the whole system.
The Bottom Line
You don’t need a Chief AI Officer. You don’t need a 12-month transformation roadmap. You need a method that starts with what you already have, focuses your investment on the opportunities with the highest return, and deploys with the fundamentals in place. The mid-market’s advantage has always been the ability to move fast and adapt. That advantage applies to AI too — if you have the right framework.
This is Part 1 of 5. Next week: your employees already know where AI works in your company — and it’s time to start listening.