Field notes

Nobody Builds to the Master Schedule

Every job I have worked on had two schedules: the CPM file in the contract, and the whiteboard the crews actually built to. The gap between them is where the money is, and it is a constraint problem long before it is a software problem.

Every heavy civil job I have worked on had two schedules. There was the one in the contract, a CPM file with a few thousand activities, updated monthly, printed at D size and pinned to the trailer wall. And there was the one on the whiteboard, three weeks wide, redrawn every Monday in marker, which is the one the crews actually built to.

Nobody says this out loud much, because it sounds like an admission that the real schedule is a mess. It is not. The two documents exist because they answer to different people. The master schedule answers to the owner, the consultant, and eventually a lawyer. It is a contract instrument and a claims record, and it has to be defensible six months after the fact. The whiteboard answers to a foreman deciding whether his crew shows up at six or at eleven. Asking one document to do both jobs is how you end up with a schedule that is neither defensible nor useful.

So keep both. What I push clients on is the traffic between them, because that is where the money is.

The lookahead is where the constraints live

A master schedule tells you the order of work. It rarely tells you why work is not happening, and on a live job that is the only question worth asking.

The three-week lookahead is where that shows up, if anyone bothers to write it down. The pour is not waiting on the crew, it is waiting on a rebar submittal that has been sitting with the consultant for nine days. The pile driving is not behind, it is waiting on a tide window that comes around twice a day for four hours and does not care about your schedule. The barge is booked and the diver is available, just not in the same week.

Those are constraints, and constraints are the actual schedule. A logic-driven CPM file assumes that when the predecessor finishes, the successor starts. On a real job the predecessor finished three weeks ago and the successor is waiting on a permit nobody has chased since June.

So the first thing I ask for is a constraint list attached to the lookahead. Not a form. A short list, one line each: what is blocking the work, who owns clearing it, and the date it has to clear by for the plan to hold. If you do only this and nothing else, you will still be ahead of most jobs, because naming an owner is what turns a complaint into a task.

Then measure whether you did what you said you would

My favourite number in construction came out of the Last Planner System, and it is almost insultingly simple. Of the tasks your crews committed to last week, what fraction actually got done? Percent plan complete. No weighting, no partial credit, no dollars.

The first honest count usually comes in lower than anyone guessed. That is normal and it is not an indictment of the crew. What matters is that this number predicts the master schedule far better than the master schedule does. A team that delivers half of what it promises week to week is not going to hit a milestone three months out, whatever the CPM file says on the trailer wall.

The other thing it gives you is a reason list. Every missed commitment has a cause, and after six weeks the causes sort themselves into a very short list, usually two or three recurring items. That list is your actual improvement plan, and the items on it are rarely what people expect. Laydown is a common one: material shows up with nowhere clear to put it, and half the morning goes into shuffling it around. That is not a scheduling problem, and no scheduling software is going to find it for you.

Where AI fits, and it is smaller than the pitch suggests

I build this kind of software now, so let me be specific about what is worth automating here.

Drafting next week's lookahead from this week's daily logs and the open constraint list is genuinely useful. It is exactly the tedious assembly work these tools handle well, and it turns a two-hour job into a review pass. Flagging constraints that have gone quiet, where the need-by date is inside two weeks and nothing has moved, is useful, because that is nagging, and people are bad at nagging consistently. Matching submittal approval dates against the activities that depend on them is useful, and nobody has ever enjoyed doing it by hand.

What is not worth automating is the commitment itself. The whole value of the Monday meeting is that a foreman looks at the work, looks at his crew, and says yes or no in front of his peers. A model can draft the candidate list. It cannot make a promise. A schedule built out of predictions rather than promises is the thing we already have and already do not trust.

There is also a hard floor on what any system knows. Nothing on that job site is going to tell you the barge is fogged in, the diver called in sick, or that the only operator who really knows the machine is on days off. That information exists for about twelve hours, in one person's head, and the only mechanism that reliably captures it is a short meeting where someone writes it down.

The modest version

If you take one thing from this, take the cheap one. For the next six weeks, write down what your crews commit to on Monday, count what got done on Friday, and keep the reasons for the misses. Do not buy anything yet.

If after six weeks the reasons all point at one broken process, that is the thing to fix, and I would be glad to look at it with you. But most of the value here is in the counting, and the counting is free.

Dealing with something like this?

Book a free 30-minute discovery call. No pitch, practical next steps either way.

Book a call