The Contoso Convergence, Episode 5: One Hour
An Advanced Dungeons & Dynamics 365 session. Episode 5 of 14.
Previously on: Episode 4, Page Seventeen, the party found the missing integration from the missing SOW page, an undocumented patch quietly feeding the shadow ledger, failing silently for three weeks with nobody watching.
Every implementation eventually produces a moment like this one: a real decision, a hard deadline, and not enough information to make the decision comfortably. This episode is that moment.
The Session
CFO: One hour. Fix it or freeze it. I have a board call.
Marge: Vex, can you patch the retry logic in an hour?
Vex: I can patch it. I can’t test it in an hour. Not against three weeks of backlog.
Thorne: Then we freeze it and post the backlog manually. Clean, auditable, slow.
Sable: Slow is fine. Wrong is not fine. I vote freeze.
Marge: I vote fix. If we freeze it now, it becomes tribal knowledge that “the integration is off” and nobody turns it back on. I’ve seen that before.
Thorne: Two and two.
Ai. Cassiopeia: I was not asked to vote. I will offer the information anyway. The backlog is not three weeks of data. It is three weeks of data plus a duplicate batch from a retry storm on day one. The real backlog is roughly half what Vex estimated.
Vex: …that changes the math. I can test that in an hour.
Marge: Then we fix it. Vex, go. Sable, you’re validating every posting before it touches the real ledger.
The room splits into motion. Vex pulls up the retry logic, hands moving fast. Sable stands over his shoulder, watching every line scroll past like she’s guarding a gate. Thorne quietly starts drafting a manual reconciliation plan anyway, just in case the fix doesn’t hold.
Thorne: (not looking up) Belt and suspenders.
Marge: Good instinct. Keep going.
Twenty minutes in, Vex hits something.
Vex: Found the actual bug. It’s not the retry logic itself. It’s a timeout value that’s too short for the freight audit feed on high volume days. The retries were never the problem. They were a symptom of a timeout nobody ever tuned.
Sable: So the real fix is smaller than we thought.
Vex: Much smaller. One config value.
Ai. Cassiopeia: I recommend documenting this timeout setting explicitly going forward. It has apparently been a default value since implementation, never revisited.
Forty one minutes later, Vex’s hands are still on the keyboard when the CFO walks back in early.
CFO: Well?
Why Cassiopeia’s vote mattered more than a vote
The most interesting moment in this scene isn’t the fix. It’s the tie. Two votes for freeze, two for fix, and the deciding information came from the one party member who doesn’t get a formal vote at all. That’s not a coincidence, and it’s not there to make a point about AI having opinions. It’s there because Ai. Cassiopeia had access to something the humans in the room didn’t have time to check under pressure: the actual shape of the backlog, not the estimated shape of it.
Under a real deadline, teams default to their gut. Freeze feels safer because it’s reversible. Fix feels riskier because it isn’t. Both instincts are reasonable, and both were wrong in the same way, because both were built on an estimate nobody had time to verify. The team wasn’t voting on values. They were voting on incomplete information, and neither side knew it.
This is the actual case for a grey collar workforce member in a room like this one. Not to replace judgment, Marge and Thorne still made the call, but to remove the guesswork underneath the judgment before the decision gets made instead of after. And notice what Vex found once the pressure came off enough to actually look: the retry logic everyone assumed was broken wasn’t the real problem at all. It was a five minute config fix hiding behind three weeks of assumed complexity.
That’s usually how it goes. The scary problem and the real problem are rarely the same problem.
What happens next
The fix holds, and landed cost starts posting clean for the first time in weeks. But validating every posting before it touches the real ledger means someone has to look closely at everything already in the system, and that closer look surfaces something nobody in the room was looking for: a second, older shadow process that’s been quietly running in the background of Contoso’s books for a lot longer than three weeks.
Next episode: Episode 6, The Ghost in AP (coming soon)
If your team has ever voted on a fix under a deadline without actually knowing the real size of the problem, you’re in good company. Gamifying the Enterprise: Game Mechanics for Continuous Proficiency is available now on Amazon: https://www.amazon.com/dp/B0GY3VWLVX
And if you want a framework for surfacing the real problem before the deadline forces a guess, start here: adnd365.com/start
A Tinkerers Notebook 