Operating Model Fit

Most leaders look at AI adoption and ask "where can we add AI?" That's the wrong question. The right one is: "where is the work already going well, and how could AI make it faster or scale further?"

The instinct to start with broken processes is understandable. Broken things hurt the most, and AI feels like a fix. But AI rarely fixes what's broken. It accelerates whatever is already happening. If a process is broken, AI breaks it faster.

The workflows that are actually good candidates for AI are the ones that already work. They have a clear input, a clear output, and a known way to tell when something has gone wrong. Those are the ones where AI can take a load off humans without introducing new risk. The ones that look like good candidates but aren't are the ones where success quietly depends on human judgment, and removing that judgment doesn't show up as a failure until much later.

It rarely looks like "we picked the wrong process." Usually it looks like one of these:

• A team automated something that looked simple, but the simple part was actually being held together by an individual quietly making good calls.
• AI was deployed on a process that doesn't have a feedback loop, so when it goes wrong, nobody catches it for weeks.
• The org targeted the noisiest pain point instead of the cleanest opportunity, and now the noise is louder and faster.
• Adoption stalled because the workflows AI was supposed to help with weren't actually documented well enough to hand off to a tool in the first place.

If any of these are familiar, your score here is lower than it looks.

Look for what's already working.

Identify two or three workflows that produce reliable outcomes today, end to end, without heroics. Those are your best AI candidates. The work has already been figured out. AI just makes it faster or extends its reach.

Make the workflow visible before you try to augment it.

If a process isn't documented well enough for a new hire to follow, it's not documented well enough for AI to extend. Map the steps, the inputs, the outputs, and the moments where someone has to make a call. The moments of judgment are where humans need to stay.

Pressure-test where human judgment is doing invisible work.

Pick a workflow you're considering automating and ask: "what's the worst that happens if AI gets this wrong, and how would we know?" If the answer is "we wouldn't notice for weeks" or "it could harm someone or damage a relationship," that's a workflow where judgment is doing more work than people realize.

• You can describe the workflow end to end without using the phrase "and then [person] just knows what to do."
• There's a clear, fast feedback loop so that when AI makes a mistake, you find out within days, not months.
• The workflows being augmented were already running well before AI was added. AI is making good things faster, not propping up broken ones.
• People closest to the work were involved in deciding what to automate, and they agreed.

• Workflows that look simple but aren't. Customer support tagging, social media posting, and performance summaries all look like clean AI candidates. They're not, because the hidden cost of getting it wrong is huge and the failure is invisible at first.
• Skipping the documentation step. If you can't write down the workflow clearly, AI can't extend it cleanly. The temptation is to skip ahead. Don't.
• Forgetting the failure mode question. Every workflow being augmented needs a clear answer to "how do we know if this goes wrong?" If the answer is vague, the workflow isn't ready.

Operating Model Fit depends on Data & Infrastructure Readiness being in decent shape, because workflows are only as good as the data flowing through them. And it sets up Education & Enablement, because the people closest to the augmented workflows are the ones who need to understand what AI is doing and where they still need to step in.

If Operating Model Fit is weak, you'll end up with AI bolted onto processes that don't have room for it, and the result is faster confusion instead of better outcomes.