Serge BelovEvery few months a new client calls with the same story. They bought ChatGPT Enterprise. They rolled out Copilot to engineering. Maybe they subscribed to an AI writing tool for marketing. Leadership announced an AI initiative. People attended a lunch-and-learn. And now, six months later, nothing has changed.
The tools are there. Some people use them. Most don't. The ones who do mostly use them for the same things they'd use Google for. The promised productivity gains haven't shown up in any metric anyone tracks. The exec who championed the whole thing is starting to get awkward questions from the board.
I keep seeing this pattern. And the problem is never the tools.
The decoration problem
Most companies are decorating, not transforming.
What I mean: they take their existing workflows, same meetings, same approval chains, same handoffs, same reporting cadence, and they sprinkle AI on top. Someone uses ChatGPT to draft an email that still goes through three rounds of review. Someone uses Copilot to write code that still sits in a pull request queue for four days. One step gets faster. The overall process doesn't change.
When you put a faster engine on a broken machine, you get a faster broken machine. The bottleneck was never the speed of the individual step. It was the workflow design itself.
The question that actually matters isn't "where can we add AI?" It's "if we were building this process from scratch today, knowing what AI can do, what would it look like?"
What redesign actually looks like
Take customer onboarding. The traditional version in most service businesses goes something like this:
- Sales closes a deal and emails the operations team
- Someone manually creates accounts in three different systems
- Someone sends a welcome email with a PDF guide attached
- Someone schedules a kickoff call
- Someone follows up a week later to check if the client has questions
Every step involves a person doing something, waiting for someone else, and hoping nothing falls through the cracks. Things fall through the cracks constantly.
The redesigned version doesn't add AI to each of these steps. It replaces the workflow:
- Deal closes in the CRM, triggering an automated onboarding sequence
- Accounts are provisioned via API integrations
- A personalized onboarding guide gets generated based on the client's use case and sent immediately
- Scheduling happens through a self-service booking link
- An AI agent monitors the client's activity in the first two weeks and reaches out when engagement drops
The human involvement shifts from executing routine steps to handling exceptions and building relationships. Signed contract to productive client goes from two weeks to two days.
That gap is the difference between adoption and transformation.
The productivity gap is visible now
This used to be a theoretical argument. It's not anymore.
When a competitor's team of five is outproducing your team of fifty, boards notice. A recent survey found that half of all CEOs believe their job is on the line if AI initiatives don't produce results. Corporate AI spending is projected to double as a share of revenue this year. Boards aren't asking whether you need AI. They're asking why it hasn't worked yet.
The pressure is real. But the typical response, buying more tools, keeps missing the point.
Why the advantage compounds
Three things are making this hard to ignore.
First, the productivity difference is measurable now. AI-enabled teams are outperforming traditional ones in standard business metrics: revenue per employee, time to delivery, cost per acquisition. These aren't anecdotes from conference talks. They're numbers in quarterly reports.
Second, the advantage compounds. Companies that reorganize around AI get lower operating costs, which fund more investment, which creates more advantage. The uncomfortable part: once a competitor achieves that structural edge, you can't catch up by buying the same tools. The advantage is in the workflow design, not the subscription.
Third, AI agents are entering production operations right now. They're handling customer service tickets, processing documents, managing schedules. The companies deploying them aren't waiting for perfect governance frameworks. They're building while they fly. If you're waiting for everything to be figured out, you're already a cycle behind.
How to tell which side you're on
Ask yourself these:
Are your AI tools optional? If someone can ignore the tooling and do their job exactly the way they did before, that's decoration. In a transformed workflow, AI is part of the process, not a sidebar.
Did any workflow actually change? Not "we added a step where someone checks ChatGPT." Did the sequence change? Did handoffs get eliminated? Did approval chains get shorter? If the workflow diagram looks the same as a year ago, you've adopted a tool. You haven't transformed anything.
Can you measure the impact? Not "people say they like it." Actual numbers. Time saved per process. Error rates before and after. Revenue per employee. If you can't measure it, you have a subscription, not a transformation.
Do you track what AI gets wrong? Error rates, hallucination rates, customer satisfaction shifts. Without these, you're guessing.
Does someone own this? Not the IT department, not legal. Someone who understands the business and the technology. If nobody is accountable for what AI does in your organization, nobody is transforming anything.
Where to start
If this sounds familiar, here's what I'd do.
Figure out what's already happening. Most companies have no idea where AI is actually being used. Employees are already using ChatGPT on their phones to write emails, draft proposals, analyze data. You can't design a strategy without understanding the current state.
Then pick a few workflows close to revenue. Customer onboarding, sales enablement, content production, lead qualification, whatever is closest to money in your business. Don't try to transform everything. Start where the impact is obvious.
For each one, redesign instead of augmenting. Ask: if we built this from scratch today, what would it look like? Work backward from that to a realistic migration path.
Decide what still needs a human. Some workflows should run autonomously. Some need someone reviewing AI output. Some need a person making the final call with AI providing options. The answer is different for every workflow, and being deliberate about it matters more than most people realize.
And measure everything. What AI produces and what it gets wrong. Time saved, revenue impact, error rates, satisfaction scores. These numbers are how you know if you're transforming or just spending.
This gap gets wider every quarter
I'm skeptical of the "act now or die" framing that's popular in AI discourse. But the catch-up cost does go up every quarter. Companies that have redesigned their workflows have lower costs, and those savings fund further redesign. It compounds quietly.
You don't need to replace your tech stack tomorrow. You need to look at your workflows honestly, pick the ones that matter most, and rebuild them around what's now possible.
Every company has access to the same models and APIs. The difference is what you do with them. That's a workflow design question, not a procurement one.