The Morning After Sign-Off
"The AI rollout went smoothly. But who's watching it from here?"
Last month, in the executive boardroom of a manufacturer. A director stood in front of the whiteboard, arms folded, and put the question out there. The PoC had wrapped successfully, the sign-off had come through, the contract was closed. And yet, not a single hand went up in the room.
This six-part series looks at the people-and-roles problems organizations actually hit after adopting AI. The first installment is about the silence in that room — the question of who owns AI.
IT Runs the Rollout, No One Runs Operations
At most companies, AI adoption is led by IT or an outside partner. The trouble starts the moment the project ends. IT moves on to the next initiative; the outside partner steps back as the contract closes. What's left is the tool, and an organization that agreed to use it.
A 2024 RAND study reported that 70–80% of AI projects fall short of their expected business outcomes — roughly double the failure rate of standard IT projects. Across the companies we've worked with recently, the biggest share of that failure doesn't trace back to model accuracy. It traces back to the absence of an operational owner. The tool exists; no one is pushing it forward.
Build a New Role, or Bolt It Onto an Existing One?
There are broadly two ways to fill the seat. One is standing up a dedicated function. In IBM's 2026 survey, 76% of participating organizations reported having a CAIO (Chief AI Officer) — a sharp jump from 26% a year earlier. That said, most respondents were global enterprises, which puts the number some distance from what mid-market and smaller companies actually look like.
The other approach is to bolt the responsibility onto someone already in place — most often as "the IT lead handles it on the side" or "strategy owns it." It's pragmatic, but as long as the double-hatter has their own KPIs, AI operations always get pushed down the queue.
Realistic Thresholds by Company Size
Watching a range of companies, here's roughly where the thresholds land. Under 50 headcount, a standalone team is a luxury. One executive wears the "head of AI" hat, and the work runs on about 0.3 FTE of an IT double-hatter. At this stage, whether that executive puts the topic on the weekly agenda — even for just 30 minutes — is what decides success or failure.
Between 50 and 300 people, one dedicated person is the minimum threshold. The title — "AI PM," "DX manager," take your pick — doesn't much matter. What matters is whether at least 60% of that person's weekly hours are actually allocated to AI operations. A 30% side-of-desk role is closer to self-comfort than a plan. Because no one supervises the other 70%, the original job eventually eats all the time.
Above 300, you need a team of at least two or three, and an executive sponsor who keeps the approval chain short. With even three approval steps in the way, experiment velocity drops from the company's speed to the speed of the approval cycle.
What the Seat Actually Needs
Creating a title and having the seat function are two different problems. Three things are required: time, authority, and budget autonomy.
Time is the 60% floor mentioned above. Authority is executive sponsorship strong enough to reshape cross-functional processes. Without a sponsor, an owner burns out within six months. Budget autonomy means the room to sign a few-thousand-dollar tool contract or run a pilot without escalating every time.
Honestly, few companies meet all three from the start. Most begin the first year missing one or two, and only fill in the gaps after the owner burns out and leaves. It's a pattern we at 5years+ have watched play out repeatedly working with Korean and Japanese companies.
What Happens When the Seat Is Empty
Take a retailer we know. They rolled out a chatbot, and six months went by without anyone being assigned to run it. In that time, the product catalog was reorganized twice and the promotion rules changed. Chatbot answers drifted further and further off, and the CS team started fielding daily complaints that "the AI is wrong." Eventually the CS lead went to the CEO and asked, "Can we just turn it off?" The tool wasn't the problem. There was simply no seat looking after it.
The One Question to Ask Before You Sign
If you're about to sign off on adoption, hold one question in reserve: "The first week of the month after this project wraps — who is watching this tool's performance, cost, and usage?" If the answer doesn't come back immediately, designing that seat comes before the rollout itself.
In the next installment, we'll look at the situation where the seat exists but the people on the floor still don't use the tool. Far more often, this is a workflow-friction problem, not a resistance problem. If you're thinking through how to design this seat right now, feel free to reach out for a conversation.