McKinsey wrote an AI transformation manifesto for companies that can still get lost in their own org charts. I read it the way I read most things written for giants: carefully, gratefully, and with a pencil ready to translate.
The tools they describe are the same tools available to a $15 million manufacturer, an $80 million services firm, a $250 million regional player. The themes still matter. What changes is the how. Fewer layers. The CEO closer to the work. Smaller teams. Less platform theater. More leverage per person. Partner versus build decisions that have to be honest. Data that lives in QuickBooks, an ERP, a CRM, and a handful of spreadsheets that everyone pretends are temporary. Trust that is still non-negotiable. And speed — real speed — that mid-market companies can keep if they refuse to copy enterprise bureaucracy as a status symbol.
I spend part of my work sitting with leaders in that band of companies, helping them put AI into practice without drowning in theater. What follows is not a checklist stolen from a consulting deck. It is the spirit of twelve themes McKinsey extracted from its Rewired work, reimagined for organizations where the conference room still fits most of the people who can change the P&L before lunch.
Capabilities beat toys
The early winners at AI, McKinsey notes, are often the same companies that were already good at absorbing technology. Their advantage is not a proprietary model. It is a muscle: the ability to turn a new tool into a repeatable way of working.
In a mid-market company, that muscle does not look like a center of excellence with a logo and a lobby. It looks like three things you can name out loud. A shared way of framing problems so the next one does not start from zero. A small library of workflows that actually run every week, not a graveyard of pilots named after last year’s vendor summit. And a habit of documenting what worked so the capability survives vacation, turnover, and the next shiny demo that arrives dressed as destiny.
A chatbot bolted onto a broken process is a toy. A capability is what remains when the demo ends and Monday begins — when the same exception is handled the same careful way by whoever is on shift, because the method outlived the excitement.
Pick the few places the money actually lives
Most companies have a list of AI use cases long enough to wallpaper a hallway. Successful ones, in McKinsey’s telling, concentrate on economic leverage points — the few places where improvement changes the physics of the business. Freeport-McMoRan on yield and throughput. Toyota on supply chain integration. Depth over breadth.
For a mid-market CEO, the translation is almost kinder. You do not have forty domains. You have a handful of places where cash, margin, and customers actually move: quote-to-cash, on-time delivery, claims and exceptions, customer retention, labor productivity on the floor or in the field, the thin margin between a won bid and a polite loss. Choose one to three. Go deep enough that the work changes, not just the slide.
I have watched teams spend a quarter generating forty ideas and none of them strong enough to rearrange a Tuesday. A use-case catalog is a comfort object. A leverage point is a decision with a name, an owner, and a number you already check.
If it does not move the P&L, it is a hobby
McKinsey studied roughly twenty AI-leading companies — large enterprises — and reported averages that still stop people in meetings: about a 20 percent EBITDA uplift, breakeven in one to two years, roughly $3 of incremental EBITDA for every $1 invested, concentrated on one to three domains. Those are large-company outcomes from their study. I will not pretend a $40 million firm magically inherits the same multiples by buying the same software.
What travels is the logic. Value that does not show up in cash, margin, cycle time, working capital, or customer retention is vanity wearing a lab coat. Stage-gate the investment. Tie it to a business KPI someone already owns — days sales outstanding, scrap rate, first-call resolution, win rate on competitive bids. Kill the pilot that impresses the board and never touches operations.
Wonder is allowed. Self-deception is not. The sand that thinks is still a tool. Tools earn their keep in the ledger, or they become expensive stories we tell ourselves about being modern.
The CEO and two or three people have to own it
McKinsey is blunt: they do not have a single success story where senior business leaders were not in the driver’s seat. In the enterprises they study, those leaders often sit one to three levels below the CEO, combining domain depth with enough technology fluency to reimagine how work gets done.
In the mid-market, that distance collapses. Often it is the CEO, the COO, and one operator who knows where the bodies are buried in the ERP. That proximity is an advantage if you use it. It is a liability if you delegate AI to “the IT person,” fund a pilot from leftover budget, and wait for a miracle to climb the stairs to the executive suite.
Owning it does not mean coding. It means defining how the business will change, sitting in the design of the workflow, asking what stops when the model is wrong, and refusing to accept a dashboard as a substitute for a result. Tech can support. Business must steer. If the people who feel the P&L are not in the room when the workflow is drawn, you are decorating, not rewiring.
People transformation, without the enterprise costume
Every tech and AI transformation is a people transformation. McKinsey’s large-company guidance leans on dense in-house talent — their “30–70 shifts,” a preference for highly skilled builders over sprawling armies of lower-skilled staff, and business leaders who become solution owners accountable for outcomes.
For SMBs, copy the density, not the dogma. You may never staff a seventy-percent in-house AI army, and you should not pretend that is the moral high ground. What you need is ownership density: a few people who can build or direct build work, clear accountability for outcomes, and honest partner choices for what you will never staff well. Partner versus build is not a personality test. It is a portfolio decision. Small, skilled, accountable beats large, vague, and rented forever.
As agents take on more coordination and routine execution, human roles move up the stack — setting objectives, defining constraints, judging quality, making trade-offs. Engineers spend less time on the repetitive and more on architecture and controls. Operators spend less time ferrying information between systems and more time deciding what good looks like. Fewer people, higher leverage, clearer names on the door. That part scales down beautifully — if you let titles and job descriptions catch up to the work.
Metabolic rate is a mid-market superpower — until you smother it
Speed, in McKinsey’s language, is metabolic rate: how fast insight becomes decision, and decision becomes action. Enterprises fight latency with platforms, embedded talent, and funding tied to outcomes rather than projects. They win the innovation race not because they own better models — everyone can rent those — but because their operating model redeploys attention without drowning in dependencies.
Mid-market companies often start with less latency. The CEO can walk onto the floor. A decision that takes six committees elsewhere can take one conversation here, sometimes before the coffee cools. The danger is importing enterprise process as a badge of seriousness — governance that exists to feel grown-up rather than to reduce real risk, steering committees that meet to admire the problem.
Protect the metabolic rate. Shorten the path from “we saw it” to “we tried it” to “we kept it or killed it.” Embed whoever can build next to whoever owns the P&L. Fund results, not theater. If you are slower than a company ten times your size, something has gone wrong that no model will fix. Speed is not recklessness. It is the refusal to add floors between knowing and doing.
Platforms, right-sized
Platforms are strategic assets in the McKinsey frame: shared, safe capabilities that drive reuse, lower unit cost, get technology into the hands of people who need it, and let AI scale without reinventing the plumbing for every team.
In a $100 million company, a “platform program” that lasts six months and produces a slide about architecture is usually a tax dressed as strategy. A right-sized platform looks more like shared workflows, common prompts and skills with named owners, a thin layer of integration between the systems you already live in, access controls that match how you actually work, and rules so the next team does not start from a blank chat window every Monday. Invest in what gets reused. Skip the cathedral until you have a congregation.
Architecture still matters. Pretending you need a data lake before you can improve invoice exceptions does not. Understanding the latitude your stack gives you — what you can change quickly, what is glued shut — is now as much a CEO skill as knowing why last quarter’s margin moved.
Make the data edible
AI needs good data. Most mid-market data is not in a lake. It is in an ERP that only three people trust, a CRM with duplicate accounts, QuickBooks that is clean enough for the accountant, and spreadsheets that hold the truth everyone is afraid to put in the system of record. Sometimes the best customer history still lives in a sales manager’s head and a folder named Final_Final_v7.
Productizing data here does not mean a multi-year master-data saga. It means making the critical tables and files discoverable, trustworthy enough for a defined use, and easy to consume for the one or two domains you chose. Enrich what creates advantage — customer context, machine history, quote win/loss reasons, field notes that explain why a job ran long — instead of boiling the ocean because a white paper said “data platform.”
Ask a practical question: can a capable person get what they need in an afternoon, or are they still wrangling for a week? If it is the week, you do not have a model problem. You have a data meal that is still raw. Feed the work you chose. Leave the museum of unused fields for later.
Design for the Tuesday after the demo
AI creates value only when it is adopted and scaled. McKinsey’s warning is concrete: a model may predict equipment failure days ahead, but if maintenance still follows the calendar, nothing happens. The prediction sits in a dashboard like a prophecy no one is paid to believe.
Design the surrounding work. Who gets the alert? What do they stop doing? What policy changes? Who owns the exception when the model is wrong? How does the night shift learn the new rhythm without a hero explaining it every time? Scaling across sites or teams needs modularity and a clear handshake between whoever builds and whoever receives — decided up front, including run costs, not as a retrofit after the pilot photo for the newsletter.
Heroics do not compound. Habits do. If adoption depends on one person’s memory and goodwill, you have a story, not a system.
No trust, no right to deploy
When AI fails, it does not fail politely. It bruises trust with customers, employees, partners, and regulators. Digital trust is the confidence that you protect data, practice real cybersecurity, ship products and processes that behave, and tell the truth about how AI is used — not only in the brochure, but in the fine print of a bad week.
Agentic systems raise the stakes. Excitement is getting ahead of many companies’ ability to test and control what semi-autonomous workflows can touch. Mid-market is not exempt because the logo is smaller. If anything, one bad deployment is a larger fraction of your reputation, and you do not have a press team to sand it smooth.
Would your current AI use survive a customer asking hard questions — or a reporter, or a regulator with patience? If the answer is a shrug, you do not yet have the right to scale. Guardrails are not anti-innovation. They are how innovation earns the next permission. At CoPilot Innovations we see this constantly: the companies that move fastest over a year are often the ones who refused to move carelessly in a month.
Agents as plumbing, not science fiction
Foundation models can now sustain longer stretches of autonomous work. Agentic workflows — systems that plan, act, check, and hand off across steps — are becoming a practical capability, especially where unstructured documents, tickets, and multi-step operations dominate. In software development the productivity gains have been striking; the same pattern is arriving in operations, service, and back office work for anyone willing to engineer it rather than mythologize it.
For SMB leaders, keep this grounded. Start where the work is repetitive, rules-heavy, and expensive when humans ferry information between systems: order exceptions, document extraction into the ERP, first-pass support triage, proposal assembly from approved content, reconciliation that eats Fridays. Automate the guardrails with the workflow, not after the incident. Codify what works into a playbook a new hire can follow on day three.
The question is not whether agents sound futuristic. It is whether they become your next quiet advantage — or the thing you scramble to understand after a competitor already rewired a painful process while you were still collecting use cases.
(Re)learn like the business depends on it
The half-life of skills is shortening. McKinsey’s closing theme is almost pastoral: organizations that learn, unlearn, and relearn fastest win, and the most important thing a CEO can do is take the leadership team on a learning journey until conviction is shared — until each leader knows their role in the transformation and the pathway is no longer abstract.
In the mid-market, that journey cannot be outsourced to a retreat and a binder. It is a CEO discipline. Read. Touch the tools with your own hands. Sit with a real workflow for an afternoon — not a demo, a Tuesday. Bring the COO and the skeptical operator into the same room until the strategic opportunity stops sounding like a keynote. Conviction is not a mood. It is the moment the top team stops treating AI as someone else’s project and starts treating it as how the company will compete.
I still feel wonder when sand-turned-silicon drafts a paragraph or spots a pattern in a week of exceptions. Wonder without hype. Caution without paralysis. The craft is still ours: what to build, what to refuse, and how to keep the company close enough to the work that rewiring is a decision, not a costume.
The giants will keep publishing for giants. The themes travel. The advantage, if you want it, is that you have fewer floors between the person who decides and the person who does.
---
Inspired by / responding to McKinsey’s “The AI transformation manifesto” (April 2026), from QuantumBlack, AI by McKinsey, and themes drawn from the Rewired capabilities. Reimagined here for mid-market leaders. McKinsey’s published figures on AI-leading companies are cited with attribution; they describe large-enterprise outcomes and are not claimed as typical SMB results. Copyright McKinsey & Company — themes paraphrased, not reproduced as manifesto text.