Episode 3, the short cut. Watch the full episode on YouTube.
Bringing in an independent set of eyes before you sanction a turnaround is about as settled as practice gets in this industry. It is in every major operator's work process and nobody argues about it. Ask ten turnaround managers what they got out of their last one, though, and you will get ten different answers. Some got a report that changed how they ran the event. Some got a binder and a bill.
I have run these reviews for a long time, and I hit the same ceiling everyone hits. So this is not a complaint about anybody else. It is a description of what an owner should expect from an assurance review, why the results vary as much as they do, and what to ask for before the next one.
Five marks of a review worth buying
It is timed to your gates. Four reviews, one per phase, each just ahead of the gate it feeds: alignment in concept, challenge right after scope freeze, final ahead of sanction, and closeout about a month after execution while the roster is still intact. Readiness measured at a defined countdown point is among the strongest controllable predictors of what an event costs and how long it takes, which is the argument for measuring it early and more than once.
It is independent. Nobody on the review team has held line accountability for what they are reviewing. A self-graded checklist does not count.
Its findings rest on sampled evidence. Packages against the register, the estimate against the scope, materials against need dates. Not on what people said in a room.
It is scored against a documented standard that does not move. Same criteria, same scale, every event, every gate. That is what turns three reviews into a trend instead of three opinions, and it is the mark most often missed, because there is frequently nothing written down to be consistent with. Ask what standard your last review was measured against. If the answer is not something you can read, the yardstick lived in somebody's head.
It recommends, it never decides, and every recommendation gets dispositioned. Accepted or declined, with a name and a date. Declining is legitimate. Silence is not. The steering team owns the gate.
A review that does all five is a real assurance review. One that misses two is a document.
Why the results are so uneven
If the marks are that well understood, the variance has to come from somewhere. In my experience it comes from three places, and none of them is bad intent.
Coverage gets rationed by a calendar. A week on site, a dozen discipline sessions, six or eight people in a room, and the senior voice answers while the room nods. How many people got asked was set by my calendar, not by who had something worth hearing.
Interviewing is a skill, and experience does not hand it to you. I have watched very good turnaround people run very bad interviews, and I have done it myself: twenty minutes of war stories, and the one question that mattered never got asked.
The schedule gets a look instead of an analysis. Somebody scrolls the critical path in a conference room and forms a view, on tens of thousands of activities.
Those are capacity problems and preparation problems. They are not judgment problems, and that distinction is the whole design brief.
Do the preparation before anyone travels
A Turnaround Assurance Review (STO·AR) is built so that every layer sets the agenda for the next one.
It starts weeks out with your coordinator, who gives us the roster, every name and role we need to hear from, and loads the documents against a request scaled to the review level. A gap report shows what is in and what is missing. Anything you gave us last review, you validate or update, you do not resubmit. The steering team answers a short poll on its own leadership behaviors ahead of the review.
Then the documents get read, all of them, against each other. Where two disagree, say a committed duration of thirty five days in one and thirty six in another, that does not get quietly resolved. It becomes an interview question, routed to the seats it implicates and framed as the records disagreeing, never as somebody getting it wrong.
The schedule does not get a look. It runs against a schedule quality standard of twenty five elements graded against the DCMA 14-point assessment, the GAO Schedule Assessment Guide and AACE recommended practice, and it returns two answers kept separate: how well was it built, and is it ready to execute. A schedule can be soundly built and still have no computable critical path. Hearing that six months out is worth more than most reports.
On day one the whole team polls live in the room, on their own phones, and the results reveal to everyone at once. That gives two numbers: the level, which is planning maturity, and the spread, which is alignment. A team can average a healthy-looking 3.8 while the answers scatter from one to five, and the scatter is the more important signal.
Every voice, not a sample
By the time the first interview starts, the interviewer already knows the roster, where the team thinks it is, where it disagrees with itself, where the documents are silent or in conflict, and what the schedule says. Then every named stakeholder gets their own interview. One person, one conversation, voice or text, on their own time, in parallel. The week that used to carry a dozen group sessions has the capacity for forty or fifty individual ones. We wrote about that change in an earlier article.
The interviews are AI-led, and each one works from a brief for that seat. Every check the seat owns is marked one of three ways. Probe: the documents are silent, so this conversation is the evidence. Verify: a document suggests an answer, so test it. Recorded: already established, so do not walk it again. It never asks what the pre-read already answered. It asks about your tie-in, on your unit, with your date, and it asks for the document while you are still in the conversation.
After twenty five years of running these interviews I sat one myself, from the other side of the table, to see how it held up. It was better than me. Not faster, better. It never drifted, it never let a vague answer pass, and it was as sharp on the fortieth conversation as on the first, which I have never been able to say about myself.
What the AI does not do
It does not decide. It has no judgment, and it has never stood in a unit at three in the morning with a startup slipping and a call to make in ten minutes. The interviews gather and the analysis proposes. Every proposed score lands in front of a consultant with its evidence attached: one voice, corroborated by a second person, backed by a document, or in conflict with one. A consultant decides what is a finding and what goes in the report, every time.
What the AI has is capacity and patience, and it reads everything. That buys the expert the one thing an expert never had enough of: every voice instead of a sample, all the evidence instead of what fit in the week, and the schedule analyzed to the element before the first conversation starts. The judgment is the same. It is finally applied to the whole picture.
Four lenses, and the gap between them
That leaves four independent lenses on one event: the steering team's view of itself, the team's view of its own readiness, the independent evidence-based read, and the schedule's structural score. When four lenses agree, you have evidence. When they do not, the gap is the finding, and it is your next conversation while there is still time to have it.
What lands on your desk is not a grade. It is a prioritized, risk-ranked set of actions, each traceable to the words or the document it came from, with a readout on site before we leave.
Five questions for your next review
Whoever runs it, these are worth asking before you sign. What documented standard will it be scored against, and will the next review use the same one? How many of my people will be heard individually, and who decided that number? Will the schedule be analyzed or looked at? Who decides what is a finding? And what happens to each recommendation after the readout?
The episode above walks one review end to end. All figures and screens shown are illustrative.