Agentic Project Cost Accounting: Catching Work-in-Progress Overruns Before They Distort Percentage-of-Completion Revenue

An agent with access to the job-cost ledger can track each cost category's burn rate against the physical percentage of that scope actually completed, and flag when the trend implies the existing EAC no longer holds — a labor category that's consumed 70% of its budget while the schedule shows the work at 50% complete is a mechanical signal, not a judgment call, that the estimate needs revisiting. This doesn't replace the estimator or project manager who has to decide what the new number actually is; it replaces the silent default of nobody looking until close, by turning "the EAC might be stale" from something someone has to remember to check into something that surfaces on its own.

A subcontractor's purchase order for $80,000 of scope doesn't show up as $80,000 of cost the day it's issued — it shows up in pieces, as invoices post against it, often weeks or months later. A job-cost report built only from posted actuals looks artificially healthy right up until a backlog of invoices lands at once. An agent pulling open commitments (issued POs, signed subcontracts, change orders not yet invoiced) alongside posted actuals gives a truer picture of where the job's total cost is actually headed, which matters because the EAC recompute above is only as good as the cost data feeding it — committed cost is real cost the job has already incurred an obligation for, even if the invoice hasn't posted.

Cost-to-cost isn't the only way to measure percent complete — units delivered, milestones achieved, or an engineer's physical-progress assessment are all independent signals that should roughly agree with what the cost ratio implies. When they diverge — the cost ratio says 60% complete but the milestone schedule says 40% — the discrepancy itself is the useful output, because it usually means either costs are running over on work that isn't actually further along, or a chunk of cost was miscoded to the wrong job or cost category entirely. An agent flagging that gap doesn't resolve which explanation is true, but it turns a silent, slow-forming distortion into a specific, dated exception someone can chase down while the underlying documents are still fresh.

Every job on percentage-of-completion carries a billings-vs-earned-revenue position: over-billed (billed ahead of revenue earned, a liability) or under-billed (earned more than billed, an asset). A single period's position is normal and expected on almost every job at almost every point in its life. What's worth flagging is a job whose position is moving the wrong direction period over period — under-billing that keeps growing, for instance, often means either billing is falling behind actual progress or the job's margin is fading and nobody's caught it yet. An agent tracking the trend across the whole active-job portfolio, not just each job in isolation, can surface which jobs are drifting before the drift becomes the kind of number that shows up as a surprise at year-end audit.

Individual EAC revisions are judgment calls, but judgment has patterns. A PM whose remaining-cost estimates come in low relative to what the job actually consumes, consistently, across multiple jobs, isn't necessarily doing anything wrong on any single job — normal estimating has variance — but the pattern itself is a real signal about how much confidence to place in that PM's current forecasts. An agent that tracks estimate-to-actual accuracy per PM over their book of jobs can't tell you why the bias exists, but it can tell finance leadership which forecasts deserve a second look before they're taken at face value in this period's revenue recognition.