What AI Transformation Actually Looks Like

On one of our most-visited pages, we had two solid pieces of content competing for the same real estate. The question came up: should we run personalized content for a specific audience - show one thing to a defined group, something else to everyone else?

Before we could answer, we needed to know if the audience existed. It did - but it was old. Not kept up to date. Only a small fraction was actually showing up on the page when personalization fired.

Not a crisis. Just an outdated audience that hadn't kept pace with newer tooling. The fix was to rebuild it and keep it current. No biggie.

But to get there, we touched three teams: merchandising to ask the right questions and own the outcome, analysts to read the data, and the audience team to understand what we had. We weighed the opportunity cost - what message were we bumping? We thought through approval windows, testing requirements, the overhead of maintaining a second version. Legal may have needed to weigh in.

One decision. Ten minutes. An easy use case.

Dozens like it happen every day.

This is what modern marketing actually looks like

Managing a large digital platform isn't just managing real estate. It's managing resources - people, technology, content - all of them moving, all of them constrained, most of them connected. A lot is automated. A lot has guardrails. But underneath it all is a layer of coordination that doesn't go away even when the machinery gets smarter.

That coordination involves judgment. Relationships. Sorting out competing priorities no system has been built to handle yet. And the environment keeps changing - products, markets, regulations. The way you work today is never quite the way you'll need to work tomorrow.

Add tight resources. Add years of technical debt. Add enough moving parts that dependencies pile up in ways nobody planned for. Some of the most important decisions happen not in systems but in a Slack thread or a five-minute call between the right two people. You still need people in that loop - their knowledge, their relationships, their judgment. Especially when something new comes up and there's no playbook yet.

Now imagine we do it better

In the version we're building toward, that same scenario looks different.

Start with the audience question. Today, that means someone going to find the right team, asking if an audience exists, waiting for them to check, and learning it's stale. In the version we're building toward, there's a single system - connected by APIs to every channel and data source we use - that any team member can query in real time. Not a spreadsheet someone exported last quarter. A live, maintained list of audience segments, built and curated by people who specialize in exactly this. You ask, you get an answer, you move.

The data behind those audiences - the analytics that tell you whether a segment is actually worth targeting - comes through clean, reliable connections rather than manual pulls and reformatting. AI assists in keeping that data accurate and current, flagging anomalies, filling gaps. What used to require a scheduled meeting with an analyst can show up in a dashboard query.

Then there's context. Right now, knowing whether a content decision conflicts with something else in flight - another campaign, another team's deployment, a business priority - requires knowing the right people to ask. In a transformed workflow, that context comes from integrated project and portfolio management systems. The system knows what's live, what's scheduled, what's on hold, and flags conflicts automatically.

Requirements - the formal documentation of what needs to be built or changed to execute a decision - are among the most time-consuming parts of any digital workflow. A lot of that work is pattern-matching: this type of request needs these fields, these approvals, these dependencies. AI can draft those requirements automatically based on the request type, leaving a human to review and confirm rather than build from scratch.

And then implementation: the tasks that are repeatable, rule-based, and don't require judgment get automated. Redundant steps get eliminated. What's left - the edge cases, the conflicts, the calls that require a human with context and accountability - gets flagged for the right person.

The conversation still happens. The legwork compresses. The judgment stays with the people whose job is judgment.

This is what my team and I have been working toward - building the tooling, the skills, the process, the speed, and the stability to transform the pieces and together the whole. Workflow by workflow. Door by door.

And that was an easy use case.

So what does it take to transform an entire platform?

That plan I just walked you through - upgrading one small merchandising workflow - only works because I know that workflow. I know why each step exists. I know what it looked like before the current version, and what problem the current version solved. I know where it's solid and where it's fragile. I know which parts would hold up under a new use case and which would crack. I know the risks inside it and how to manage them. That knowledge is what makes transformation possible - not the technology. The technology is the second thing. The first thing is knowing the thing you're transforming, deeply enough to know what to keep, what to change, and what breaks if you get it wrong.

To do this well, you need all three: deep knowledge of the work, command of the technology, and a clear picture of what the work looks like on the other side.

Without that, even well-resourced transformation efforts could end up solving for the wrong problems.

Here's what the work actually requires.

A real baseline. What are the actual problems and how do you measure them? How do your workflows function today, step by step? Who's in the loop because they need to be - compliance, governance, judgment - and who's doing work that could be handled differently, freeing them up for the things there's never been time for.

Know what to change and what to keep. Not every human touchpoint is inefficiency. Some of it is there for good reason. Know the difference before you start.

Start with technical debt. Make it worth building on before you build on top of it. AI can help work through the backlog - and it gets your technical teams doing real AI work early, which matters.

Document where you are - with AI's help. What used to take six to twelve months and go stale immediately can now be done fast, kept current, and fed directly into planning. When it's time to redesign how things work, you have something AI can actually participate in.

Bring every team along at the same time. QA, design, analytics - each function has its own version of this work. It doesn't have to sequence. It can all move at once.

Keep it modular. Individual pieces can be improved one at a time, by the people who know them best. You don't need to move everything at once - and trying to usually causes problems.

Be straight with your people. Transformation only works if the people who know how things work are still engaged. Give them a clear picture of where things are going and a real role in getting there.

One more thing

This work is new. Not conceptually new - people have talked about AI transformation for years. Practically new. The kind of agentic, workflow-level AI that actually changes how a platform operates has only really broken through in the last couple of quarters. The best practices are being written as we go. The tooling is still maturing. The people who've genuinely done it at scale are few.

That's not a reason to wait. It's a reason to be honest about what you know and what you don't.

Real transformation here isn't a technology swap. The workflows, the data, the skills, the org model - they all have to move together, piece by piece, with people who understand both the technology and the work it's meant to improve.

Anyone who tells you they have it fully figured out probably hasn't gotten far enough in to know what they don't know yet. The mark of someone who's actually done this work isn't just what they can articulate. It's how clearly they can tell you where the map ends.