Change Management Is the Project, Not the Postscript

My free readiness assessment asks twenty questions. The one that predicts the most while asking the least is this: what ships alongside an AI deployment?

The lowest-scoring answer is "an announcement email." It is also, in my experience, the default. The tool launches; everything else stays the same.

I've written before that the winners do workflow redesign, not tool overlay — it kept showing up as a bullet point in other essays. It deserves its own argument, because it's the bullet point most programs skip.

The email is the tell

You know this rollout. Procurement closes. Licenses are provisioned. Someone drafts the email — "We're excited to announce..." — and books a lunch-and-learn. There's a Slack channel with a robot emoji. Everyone involved experiences real momentum, because everything on the checklist is genuinely getting done.

Six months later, the license count and the weekly-active count have never met. Usage spiked the week of the email, sagged by week four, and settled at the handful of people who would have found the tool anyway. Nobody's job changed. No process changed. The pilot quietly joins the 95%.

The email isn't the mistake. It's the symptom. Announcing a tool says you believe the work ended when the software arrived. The work started there.

The tool is the small part

This isn't a hunch; it's the most replicated finding in enterprise AI.

• BCG's rule of thumb is 10-20-70: 10% of the effort is algorithms, 20% is technology and data, 70% is people and processes. Most budgets are allocated in exactly the opposite proportions.
• McKinsey tested roughly 25 attributes in its State of AI survey and found that redesigning workflows had the biggest effect on whether gen AI showed up in EBIT. Not model choice. Not spend. Workflow redesign.
• Prosci's benchmarking research — 12 editions, 10,800+ contributors — finds initiatives with excellent change management are seven times more likely to meet objectives than those with poor change management: 88% versus 13%.
• Gartner predicted that at least 30% of GenAI projects would be abandoned after proof of concept by the end of 2025. Look at its reasons — unclear business value, poor data quality, inadequate controls, escalating costs. Organizational failures, every one.

Buy the tool and skip the 70%, and you haven't done a smaller version of the project. You've done a different project — one with a known outcome.

We ran this experiment a century ago

Stanford economist Paul David wrote the definitive paper on this in 1990, "The Dynamo and the Computer." Factories began electrifying in the 1880s. For roughly forty years, productivity barely moved — because managers swapped the steam engine for a big electric motor and kept the factory exactly as it was: one central power source, line shafts and belts overhead, machines arranged around the drivetrain instead of the work.

The gains arrived in the 1920s, when factories adopted "unit drive" — a small motor on every machine — and redesigned the whole building around the flow of materials instead of the flow of power. Electrification went on to account for about half of manufacturing productivity growth that decade.

Overlay: four decades of nothing. Redesign: half of all productivity growth. A copilot bolted onto an unchanged workflow is group drive. The announcement email is its press release.

What shipping the change looks like

"Change management as the project" sounds like a slogan until you write the plan. Concretely:

• The deployment plan is a workflow redesign plan. Map the process, name the steps the AI now owns, and reposition the humans to exceptions and judgment. If the process diagram is identical before and after, you're shipping an overlay.
• Roles change on paper, not by vibes. Who reviews exceptions? Who owns the queue? What does the reviewer's job description say now? Unwritten role changes revert within a quarter.
• Metrics and incentives move with the process. People measured on the old way of working will keep working the old way. Rationally.
• Adoption is tracked after go-live like a product metric. Weekly active use by the target team, against a target, reviewed by a named owner. Seat licenses provisioned is a procurement metric, not an adoption metric.
• Training is enablement, not an event. A session at go-live is "training at go-live" — a session, then silence. The habit forms in weeks three through ten, which is exactly when most programs go quiet.
• The budget passes a simple test. If the license line is bigger than the redesign-and-adoption line, the plan is inverted. BCG says the ratio should embarrass your software vendor.

None of this fights your workforce. PagerDuty found 66% of office professionals have used AI tools at work they believed weren't even permitted. The appetite exists. The overlay rollout wastes it.

Walking back up the rubric

The assessment question scores four answers, and they map cleanly to real programs. Zero is the announcement email: tool launches, nothing else moves. Two is training at go-live: a session, then silence. Three is training plus some redesign: process tweaks accompany the bigger launches. Five is a redesigned process with adjusted roles and adoption tracked after go-live.

Here's the part worth noticing: the five isn't dramatically more expensive than the zero. It's differently sequenced. The redesign work replaces the second and third pilots you'd otherwise fund to compensate for the first one nobody adopted. Prosci's 88%-versus-13% gap isn't a premium outcome for a premium price — it's what happens when the same money ships in a different order.

Where to start

Take your next deployment — or your biggest stalled one — and write the go-live plan with the tool deleted from it. What changes about the process, the roles, the metrics, the week-ten follow-through? If the page is empty except for a training session, you've found the announcement email wearing a project plan.

The change-management question is one of twenty in my free AI Readiness Score, across pilots, data, talent, and governance. Ten minutes gets you a score, a benchmark against enterprises your size, and your first three moves. And if your honest answer today is "an announcement email," you'll at least know — that one is fixed by a decision, not a purchase.