The mid-year trap: your program looks green, your outcome hasn't moved, and nobody in the room has noticed yet.

If your steering committee left the last update satisfied, this issue is worth reading before the next one. This week I am looking at two programs that hit the same mid-year wall — one team asked the right question and turned 15% adoption into 80% in two months, one team didn't ask it and quietly scrapped a million-dollar AI deployment after peak season. The difference wasn't the technology. It was what each team chose to measure.

Inside: the pattern that separates both programs, a diagnostic prompt to run before you build your next steering committee deck, and the one question worth asking in the room before anyone declares progress.

The adoption report landed on a Thursday.

Fifteen percent. Four months into a knowledge retrieval system that leadership had called a "strategic investment" in January, only fifteen percent of employees were actually using it. The program director told me she sat with that number for a moment before she walked into the steering committee.

She had two options. The first was the easy one: frame it as a change management problem. People needed more training, more internal communications, more nudging. The technology was sound; the humans just hadn't caught up yet.

She chose the harder one. She called it a workflow design problem.

"The tool was technically correct," she said. "It was also completely invisible in everyone's actual workday. To use it, you had to leave what you were doing, open a separate interface, remember it existed, and then come back. We had built a perfectly good AI that lived in a place nobody went."

The mid-year pivot wasn't glamorous. No new model. No bigger budget. A UI overhaul and a Slack integration — moving the tool to where people already spent their time. By June, adoption was at 80%. New employee onboarding time dropped by half.

The number that changed wasn't the one on the AI dashboard. It was the one the program was originally funded to move.

That is what a mid-year review is supposed to do. Not confirm that the deployment happened. Ask whether the deployment landed where it needed to.

Somewhere around the four-month mark, most AI programs develop a reporting problem that nobody names. The metrics available to measure — deployment rates, training completions, license activations — are metrics of build, not metrics of use. They tell you whether the tool was installed. They don't tell you whether the tool changed anything.

McKinsey's latest AI survey found that while 88% of organizations report regular AI use in at least one function, only about one in three have managed to scale across the enterprise. BCG is starker: 60% of companies generate no material value from their AI investments. Five percent create substantial value at scale. A Writer study found 54% of C-suite executives say the internal cultural pressure to adopt AI is actively tearing their organizations apart — not because AI doesn't work, but because it's being bolted onto workflows that were never redesigned to receive it.

That gap is not a technology gap. It is a workflow gap and a data gap, and mid-year reviews are almost never designed to surface either one.

A major global retailer learned the data version of this the expensive way. They invested heavily in a generative AI shopping assistant ahead of peak season. By mid-year it was quietly scrapped — not because the model failed, but because the model had nothing reliable to work with. Product databases were fragmented across legacy systems. The AI was impressive and completely uninformed. Nobody had audited the data pipeline before they audited the model.

The financial services firm and the retailer had the same mid-year moment. One team asked the workflow question in time. One team hadn't thought to ask the data question until it was too late.

Your mid-year review will find whatever it is designed to find. Design it to find the right things.

Run this before building your next mid-year steering committee deck.

Paste this prompt into Claude, ChatGPT, or Gemini after filling in the brackets:

"I am preparing a mid-year AI program review for [program name]. Our deployment metric shows [X%] of users have access to the tool. Our active adoption rate is [Y%]. Help me diagnose the gap between access and use by asking me five questions — about how the tool is integrated into daily workflows, what the data pipeline looks like underneath it, and where employees have to leave their current tools to use it."

If the follow-up questions surface things you haven't presented to your steering committee yet — that is your agenda for the next meeting.

When your steering committee reviews your mid-year numbers, which question are they actually answering —

"Did we build what we said we would build?"

or

"Did what we built change the work?"

And if it's the first one, who in the room is responsible for asking the second?