The AI Mandate That Skipped the Foundation

When Shopify's CEO told the entire company last year that using AI well was no longer optional, the message was clear. AI proficiency became a baseline expectation for every employee, factored into performance reviews and headcount requests. It has become one of the more visible examples of a pattern moving through large organizations right now: set the expectation, put it in writing, and trust the organization to follow.

The organization does not always follow. A survey of 2,400 knowledge workers this spring found that 29% had taken deliberate action against their own company's AI push, from ignoring a mandated tool to quietly turning in weaker work. Among Gen Z respondents, the number was 44%. A separate global survey of 3,750 executives and employees found something similar: 54% had quietly reverted to doing the work themselves, with the AI tools sitting unused. The mandates were clear. The compliance underneath them was not.

The foundation a mandate skips

No leadership team would attempt a merger or a restructuring this way. Before execution starts, they align at the top on the approach, build a case for change that people believe in, and work through resistance before it hardens. That discipline is treated as non-negotiable everywhere except here. Almost no one skips it on purpose, which is exactly why it goes unexamined. With AI, the foundation becomes a formality to revisit later, if at all.

That foundation is made of four things, and a mandate substitutes for none of them. 

1. Alignment. Ask the CFO, the COO, and a business unit leader what the initiative is supposed to deliver, and each will give a different, perfectly defensible answer. None of them are wrong. But when the mandate goes out under one banner and gets executed against three different definitions of success, the first budget review after launch turns into a debate about whose number counts, and the initiative loses a quarter to a disagreement that should have been settled up front.

2. A why that resonates. The version leadership hears is about competitive position and the cost of moving too slowly. The version a functional leader running a team actually needs is different: whether their judgment still counts, what they're being measured on now, and whether the tool is there to help them or to replace the headcount line they're on. Write only the first version, and the second gets filled in by whatever people are already telling each other.

3. Real buy-in. Training completion and login counts look strong in the first quarter, which is exactly why they get reported up the chain as proof that adoption is working. But neither number says anything about whether people actually changed how they work. Six months later, the productivity and quality the initiative was meant to deliver haven’t moved, because people finished the training and then quietly went back to the process they already trusted, and no one asked what would make the new way worth trusting. 

4. Decision ownership. In most organizations, nobody can say who is accountable when an AI-assisted recommendation turns out to be wrong. It hasn't come up yet, because the tools are new enough that nothing has gone wrong in a way that mattered. The first time something does, a flawed forecast, a bad risk call, a decision a regulator wants explained, whoever has to explain what happened runs into the same gap: no one was ever assigned to own it.

Each gap makes the next one worse. Skip alignment, and the why splinters into as many versions as there are leaders repeating it. A why written only for the top keeps buy-in stuck at the surface. Weak buy-in means the operating model gets built around what leadership assumed people would accept, not what they'll actually use. A mandate can produce activity at every stage. It doesn't produce the organization that was ready before the tools arrived.

What the leaders who get this right do differently

In the transformations we support, the leaders who get real value from AI do this work up front.

• Settle the trade-offs in one room, once. Before anything goes out, get every leader in the room and agree on what productivity, quality, and value creation the initiative is meant to produce, specifically enough that two leaders describing it in different rooms sound like they read the same script. Skip that step, and the first team to hit a real tradeoff, speed against accuracy or headcount against oversight, resolves it on their own. 

• Write the version for the people whose work is actually changing. An executive audience needs competitive risk and the cost of delay. The people being asked to work differently need something else: what happens to the judgment they've built over years, and what they'll be measured on the day the new tool arrives. Leaders who write only the executive version, assuming it translates downward, are the ones surprised months later that adoption never got past the early enthusiasts.

• Earn buy-in while it can still change the rollout. Compliance is what you get when people have no real path to object. Buy-in is what you get when their objections arrive early enough to matter. The leaders who do this well bring the people into the design before it's final, treat pushback as information, and visibly adjust the plan in response. That's the same pattern in the survey data above: workers who never got a real say find their own ways to say no.

• Decide who owns the outcome before the tools arrive. Before the first AI-assisted recommendation reaches a real decision, someone needs to already own what happens next: who signs off, where a person's judgment still overrides the model, and how the new workflow actually retires the old one.

The bottom line 

Shopify's memo, and the mandates that have followed elsewhere, reflect something real: leadership is finally taking AI seriously enough to demand results. That's the right instinct aimed at the wrong lever. A mandate can set an expectation. It can't align a leadership team, translate a message for the people it's asking to change, earn buy-in instead of extracting compliance, or decide who owns the outcome before the tools show up. Those are the four things every other kind of transformation treats as the actual work, and the four things AI mandates keep skipping. The leaders who get real value from AI ask for it before they ask for compliance.

Andrea Schnepf

P.S.: Everything above supports the argument for building the operating model before the mandate ever goes out. We're exploring exactly how to do that on August 19. Join us: Making the AI Vision Real: Adapting Your Operating Model to Scale AI.