AI Transformation

The Pattern Every Executive Recognises

Here's a scenario you may already know firsthand. Company buys an AI tool. Announcement goes out. Some teams experiment. Six months later, the tool works perfectly — and 90% of the company has gone back to their usual way of working.

It's the same story as every enterprise software rollout that didn't stick. The technology isn't the problem. The operating model is. Installing technology without changing how you work is just expense. Think of it like a treadmill nobody runs on — your legs work fine, the machine works fine, the missing piece is the decision to change the routine. No amount of upgrading the treadmill fixes that.

This is the single biggest reason AI transformations fail. Not bad technology. Not bad people. Bad operating models — the habits, rules, and decisions that govern how work actually gets done.

The Shopify Story: From Chaos to AI-Native in 18 Months

In 2023, Shopify's CEO Tobi Lutke looked across his company and saw something familiar to every executive reading this: AI experiments everywhere, results nowhere.

Teams were scattered across the building doing their own thing with ChatGPT. No shared tools. No shared measurements. No way to tell what was working. It was like an orchestra where every musician picked their own song — lots of noise, no music.

Lutke did something radical. He didn't buy better AI. He didn't hire a fleet of data scientists. He made three decisions that changed how the company operates:

Decision 1: "Justify AI before you justify a new hire." Every team had to explain why AI couldn't do the job before requesting headcount. Overnight, AI went from "nice-to-have side project" to "the first option we consider."

Decision 2: "No AI ships without automated quality tests." Before any AI feature reached a customer, it had to pass an eval gate — think of it like a driving test before you get the keys. This killed the "move fast and break things" mentality that sinks AI projects.

Decision 3: "Every product spec includes AI metrics from day one." Not "we'll measure later." Day one.

(These specific policy details are illustrative of Shopify's reported AI-first approach — only the headcount mandate has been publicly confirmed via Lutke's internal memo; verify other details against Shopify's published materials.)

The results: Shopify's CEO mandated AI consideration before any new hires, and (illustrative figure — verify against Shopify's published 2024 earnings materials) over 80% of Shopify engineers adopted AI coding assistants. AI features appeared on every major product surface. Support ticket deflection improved materially. All within roughly 18 months of those three decisions.

Here's the punchline: every competitor had access to the same AI models. Shopify's edge came from the mandate, not the model.

The Four Stages of AI Transformation

Think of AI transformation like learning to cook. Stage 1 is microwaving leftovers. Stage 4 is running a restaurant. You can't skip stages — the chef who never learned to boil water can't run a kitchen.

Notice something? Every bottleneck on this roadmap sits at an exec decision node. The technology at each stage largely exists. Companies stall because the mandate above it never arrived.

Here's the full breakdown:

A stalled organisation almost certainly lacks a mandate, not a model. Each stage builds on the previous one's infrastructure — there is no shortcut to AI-native.

Why "Just Installing AI" Doesn't Work

Let's make this vivid.

Imagine a hospital that buys a world-class MRI machine. They wheel it into a supply closet. Doctors still diagnose by gut feel. Nurses don't know how to read the scans. Nobody changed the patient intake form to route cases to the MRI. The billing system can't code MRI procedures.

The MRI works perfectly. The hospital doesn't use it.

That's what "installing technology without changing how you work" looks like. And it happens at company after company with AI:

• The CRM team buys an AI lead-scoring tool. Sales reps ignore the scores because their commission structure still rewards cold-calling volume, not AI-prioritised leads.
• The support team deploys an AI chatbot. But the escalation policy wasn't updated, so every bot conversation gets escalated to a human anyway — creating more work, not less.
• The finance team gets AI-powered forecasting. The CFO still trusts the spreadsheet she's used for 10 years. The AI forecast sits in a dashboard nobody opens.

In every case, the technology was fine. The operating model — the rules, incentives, and habits that govern how work actually happens — never changed.

The Transformation Playbook: What Execs Actually Control

Here's the uncomfortable truth: every time you leave AI adoption to the CTO alone, it stalls. Technology leaders can build the tools, but cultural change requires the CEO to visibly mandate it.

Think of it like a school principal. A math teacher can be brilliant, but if the principal doesn't put math on the schedule, doesn't buy textbooks, and doesn't make grades count — nobody learns math. The CTO is the math teacher. The CEO is the principal.

What execs actually control that matters:

You can accelerate every stage by deciding these four things upfront. Technology without that process change is just expense.

Building Capability: Start Internal, Then Go External

You can build organisational AI capability fastest by starting with internal tools. Why? Because internal tools let teams learn to work with AI before it reaches customers — where the stakes are higher.

Think of it like a pilot learning to fly. You don't start with a transatlantic flight full of passengers. You start in a simulator. Then a small plane. Then bigger planes. Then passengers.

Internal tools are your flight simulator. If the AI makes a mistake on an internal report, someone catches it and fixes it. If the AI makes a mistake on a customer's insurance claim, you're in the newspaper.

The progression looks like this:

1. Internal productivity tools (low risk, fast learning) — AI that helps employees draft emails, summarise meetings, extract data from documents
2. Internal decision-support tools (medium risk, building trust) — AI that recommends actions but a human always decides
3. Customer-facing features with human oversight (higher risk, proven capability) — AI that drafts responses for customers, reviewed by a human before sending
4. Autonomous customer-facing AI (highest risk, earned trust) — AI that acts independently because you've built the eval infrastructure to trust it

Now Apply It: The Northstar Insurance Challenge

You've learned the stages, the exec unlocks, and the "start internal" principle. Now put it all together on a real scenario.

Back to the Company With the Treadmill

Remember that company from the opening — the one where 90% of employees went back to their old way of working after the AI tool arrived? They didn't have a technology problem. They had an operating model problem. The treadmill worked perfectly. Nobody ran on it.

Shopify had access to the same models as every competitor. What Shopify had that the treadmill company didn't: three decisions that changed how work gets approved, tested, and specified. Not better AI — a different operating model.

The treadmill doesn't make you fit. The decision to change your routine does.

Key Takeaways

• Treadmill rule: Installing technology without changing how you work is just expense. The operating model — not the software — determines whether AI hits the P&L or stays in the pilot budget.
• Mandate, not model: Every time you leave AI adoption to the CTO alone, it stalls. Cultural change requires the CEO to visibly mandate it. Shopify's edge came from three process decisions, not better AI.
• Start internal: Build organisational AI capability fastest by starting with internal tools, so teams learn to work with AI before it reaches customers where the stakes are higher.
• Four decisions: Who approves AI features, how they're measured, what "done" means, and when AI gets considered. Decide these upfront and every stage accelerates.