Every construction software vendor on earth will tell you their product has "powerful analytics." Most of the time what they mean is a pie chart. You get a dashboard with a spinning gauge, a couple of trend lines, and a big number that goes up and to the right, and none of it tells you the one thing you actually walked into the trailer to find out: is Thursday going to happen or not?
The honest truth after two decades of running jobs is that data doesn't fix a schedule. People fix schedules. But the right numbers, looked at the right way, will tell you where to point those people before the problem becomes a change order. This is a walk through the analytics that earn their keep on a real project, what each one is really measuring, and the traps that make a good-looking number lie to you.
Start with the one metric that predicts everything: PPC
If you only track a single number off your weekly work plan, make it Percent Plan Complete. PPC is dead simple: of the tasks a crew committed to finishing this week, what fraction actually got done? Twenty commitments, sixteen completed, that's 80 percent. It comes straight out of the Last Planner approach, and it's the closest thing our industry has to a leading indicator.
Here's why it matters more than your Gantt chart. A master schedule tells you where you're supposed to be. PPC tells you whether the crews can be believed when they say they'll be somewhere. A job humming along at 85 to 90 percent PPC is a job where the look-ahead is real and the sequence is holding. A job sitting at 50 percent isn't behind because people are lazy — it's behind because the plan is disconnected from the field, and every promise you make to the owner is built on sand.
The number that most crews chase early is a trap, though: don't reward a high PPC by itself. A foreman can hit 100 percent every week by only committing to work he already knows is bulletproof — quietly sandbagging. If PPC is perfect but the job is slipping, your crews are under-committing to protect their score. You want PPC high and the look-ahead full of meaningful, sequence-driving work. One without the other is theater.
The most valuable data on the whole job: reasons for variance
PPC tells you that something didn't get done. The reason code tells you why — and that's where the money is. Every time a committed task doesn't finish, somebody should log a short reason: prerequisite work not complete, material not on site, missing information/RFI open, weather, manpower, rework, changed priorities, permit or inspection.
One missed task is noise. But run those reasons for six weeks and stack them up, and a job's real disease becomes impossible to ignore. I've watched a project where PPC was mediocre and everyone blamed the mechanical sub — until the variance log showed 40 percent of misses tagged "prerequisites not complete." The mechanical crew wasn't slow; they kept showing up to walls that weren't ready. The problem was three trades upstream, and no amount of yelling at the pipefitters would have touched it.
That's the whole point of variance analytics: it moves the conversation from blame to cause. When you can put up a simple bar chart at the Monday meeting showing that "waiting on RFI answers" is your number one killer, you've just turned a personnel argument into a process fix. Good look-ahead tools — LookAheadWall included — let crews attach a reason the moment a task slips, so the pattern builds itself instead of getting reconstructed from memory a month later.
Trend lines only matter if you know what's normal
A single week's number is almost useless. Weather, a holiday, one bad delivery — any of it can wreck a week that means nothing about the trajectory. The value is in the slope. Plot PPC over eight or ten weeks and the story writes itself: a line climbing from 60 toward 85 means your planning discipline is taking hold; a line drifting down means something is quietly coming apart, usually two or three weeks before it shows up as visible schedule slip.
The rule of thumb I use: never react to one data point, always react to three in a row moving the same direction. Three straight weeks of declining PPC is a real signal worth a hard look at the six-week look-ahead. One bad week is Tuesday.
Manpower and productivity: watch the ratio, not the headcount
Raw headcount analytics are the easiest to fool yourself with. Forty guys on site feels like progress. What you actually want is output per unit of labor — linear feet of wall framed per crew-day, fixtures set per plumber-day, square feet closed per week. Track the installed quantity against the hours it took, and you get a productivity trend that tells you whether crews are getting more efficient as they learn the building or bogging down.
The gotcha here is stacking. When productivity per man drops even though headcount went up, you're almost always looking at trade stacking — too many crews crammed into the same area, tripping over each other. The analytics won't say "stacking" in plain English; you have to read it. Density up, output-per-man down, in the same location, is the fingerprint. This is exactly the kind of thing a location-based weekly work plan surfaces early, because you can literally see two trades assigned to the same zone in the same week before they collide on site.
Trade partner reliability: the number that should drive who you hire
Once you're logging PPC by trade, you're sitting on the most useful subcontractor scorecard in the business: reliable promising. Which subs hit their commitments, and which ones tell you what you want to hear on Monday and vanish by Wednesday? Roll that up across a few jobs and you've got hard data for prequalification that has nothing to do with the low bid.
A word of caution so you don't punish the wrong people: a sub with low completion rates might be perfectly reliable and just chronically starved of ready work areas by the trades ahead of them. Always read reliability next to the variance reasons. If a sub misses because their own manpower or coordination fell short, that's on them. If they miss because they keep getting handed unready work, that's a sequencing failure that belongs to the plan, and your scorecard needs to say so or you'll blacklist your best subcontractor.
Quality and safety data: hunt for the pattern, not the incident
Punch and deficiency logs feel like a records-keeping chore, but analyzed in aggregate they're a heat map of your real problems. Sort deficiencies by type, by location, and by responsible trade. When the same defect keeps showing up — say, the same detail failing at every exterior corner — you don't have twenty problems, you have one systemic problem repeated twenty times. Fix the detail or the crew once and the whole cluster disappears.
Safety analytics work the same way, and here the leading indicator is gold. Don't wait on the lagging number — recordables, lost-time incidents — to tell you a job is dangerous, because by then someone's already hurt. Track near-misses and safety observations instead. A spike in observations in a particular area or activity is the building warning you before it bites. Treat a cluster of near-misses the way you'd treat a cluster of deficiencies: as a pattern demanding a process change, not a stack of paperwork to file.
Forecasting: useful, but keep both hands on the wheel
The newest crop of tools will project a completion date or forecast resource needs off your historical pace. Used honestly, that's genuinely valuable — a data-driven finish projection off actual PPC and productivity is worth ten optimistic guesses in a status meeting, and it gives you leverage to raise the alarm early instead of confessing a slip at the end.
But treat every forecast as a conversation starter, not gospel. A projection built on the last ten weeks assumes the next ten look like the last ten, and construction rarely obliges — weather turns, a long-lead item lands, a sub demobilizes. The forecast's job is to make you ask "what would have to be true for this to be wrong?" A number that shuts down thinking is worse than no number at all.
How to actually put this to work
You don't need a data science department. You need a short, disciplined loop that turns field reality into a handful of numbers you look at every week. Here's the practical version:
- Track PPC weekly, per crew and per trade. Make committing to the weekly work plan a real act, and hold the number up in the Monday meeting.
- Log a reason on every miss, every time. This is the discipline that pays for itself. No reason, no learning.
- Review trends monthly, not weekly. Look for three-point moves. Ignore single-week noise.
- Read every metric against its neighbors. PPC next to variance reasons. Productivity next to crew density. Sub reliability next to who's feeding them work.
- Feed what you learn back into the six-week look-ahead. Analytics that don't change next week's plan are just a scrapbook.
That last point is the whole game. The reason short-interval scheduling and look-ahead planning produce good analytics in the first place is that they generate honest, structured data as a byproduct of how crews already work — commitments made, commitments kept, reasons when they aren't. Software like LookAheadWall is worth its keep here mostly because it captures that data at the moment of truth, out in the field, instead of forcing someone to reconstruct the week from a clipboard on Friday afternoon.
The best superintendents I've worked alongside weren't the ones with the fanciest dashboards. They were the ones who picked three or four numbers that actually predicted trouble, looked at them religiously, and — this is the part nobody markets — acted on them while there was still time to act. The analytics don't run the job. You do. They just make sure you're pointed at the right problem before it costs you the schedule.