Two Monthly Business Reviews, both with an AI-generated forecast on the agenda. In the first, the model's number goes up on the screen, no one in the room has a reason to disagree, and the meeting moves on — the forecast got a seat at the table and quietly took over the whole conversation. In the second, the model's number goes up on the screen, and it becomes the starting point for the only fifteen minutes of the meeting that matter: why does this number look different from what the regional lead is hearing from her three biggest accounts, and which one should the plan actually follow. Same model, same forecast, completely different value to the business.
That is the real question in front of most operating reviews right now, and it is not "should we trust the model." It is narrower and more useful: which parts of this meeting should the model own, and which parts still require someone in the room who can be held accountable for what happens next. Get that split right, and the MBR gets sharper and faster. Get it wrong in either direction — treating the model as gospel, or ignoring what it's actually good at — and the review stays exactly as slow and as shallow as it always was.
Where AI Has Earned Its Seat at the Table
Building the base case, not defending it. The most wasted time in a traditional operating review is spent re-deriving a number everyone in the room could have agreed on from the data. A model that builds the starting forecast, the variance analysis, and the trend line frees the actual meeting time for the part a spreadsheet cannot do: deciding what to do about the gap between the base case and what the room actually believes is coming.
Flagging the exception before the room finds it. A well-built model does not wait for someone to notice a site is trending off-plan — it surfaces it before the meeting starts, ranked by how much it matters. That shifts the entire structure of the review from "walk through every line" to "spend the time on the three things that are actually at risk," which is the difference between a ninety-minute status meeting and a thirty-minute decision meeting covering the same ground.
Running the scenario math live. "What happens to service levels if we shift ten percent of volume to the backup site" used to be a question that got tabled until someone could run the numbers offline. A model that can answer it in the room changes what an operating review is capable of deciding in real time, instead of deferring every real trade-off to a follow-up meeting three weeks later.
Where It's Still Asked to Leave the Room
Trade-offs with asymmetric consequences. Deciding whether to protect a strained but strategically important client relationship at the cost of margin on that account is not a forecasting problem. It is a judgment call that weighs relationship history, reputational exposure, and a read on what the client will actually do — inputs a model was never built to hold, because it has never sat across the table from that client.
Situations with no real pattern to learn from. A model is only as good as the pattern it was trained to recognize. The genuinely new situation — a first-of-its-kind supplier failure, a regulatory shift with no precedent, an acquisition integration decision — is exactly where historical pattern-matching runs out, and exactly where the room needs someone willing to make a call without the comfort of a confidence interval.
The moment someone has to own the outcome. Every operating review eventually reaches a line item where a name has to go next to a decision. A model does not get asked back to the next MBR if the call was wrong. A person does. That accountability is not a technicality — it is the entire reason the review has authority in the first place, and it is not something that can be delegated to the system that built the forecast.
The Trap Most Operating Reviews Fall Into
The first trap is letting the model's output substitute for the debate it was supposed to sharpen. A forecast that goes unquestioned because it came from a model instead of a person is not a better forecast — it is the same blind trust operators used to place in whoever presented the most confident slide, just with better production values.
The second trap, just as common, is refusing to change how the meeting runs at all. Bolting a dashboard onto an MBR that was already built around status updates instead of decisions does not fix the meeting — it just adds a more sophisticated status update. The leverage only shows up when the agenda itself gets redesigned around what the model is actually good at, freeing the room's time for what it is not.
What a Well-Run AI-Enabled MBR Looks Like
- Send the base case and the flagged exceptions before the meeting, not during it. The room's time should start where the model's confidence runs out, not at zero.
- Spend the meeting on the two or three items the model flagged as most uncertain or highest-impact — not a walkthrough of everything it got right.
- Name an owner for every judgment call the model surfaces but cannot resolve. A flagged exception without an accountable owner is just a more accurate way of noticing a problem no one is fixing.
- Keep a standing agenda item for "what doesn't fit the pattern." The most important thing happening in the business is sometimes the thing the model has no history to recognize yet.
What I Am Seeing Now
The clearest version of this I have built firsthand was standardizing CRM and reporting automation inside a logistics operation — work that cut administrative time by 40% and pulled meaningful cost out of procurement, not because the system made the decisions, but because it gave the operating review a cleaner starting point and freed the room to spend its time on the calls that actually needed a person. The pattern holds everywhere I have seen it work: the model earns real estate on the agenda for exactly as much as it can do well, and the room stays in charge of everything that still requires a name attached to the outcome. That balance is not a permanent settlement — the model's seat at the table gets bigger every year. But the seats that are still asked to leave the room are the ones that were never about data in the first place.