Manpower forecast by cost code

Labor overruns rarely show up on the day they happen. They show up when the weekly cost report lands, and by then the hours are paid and the crew has moved on.

The problem

Labor overruns rarely show up on the day they happen. They show up when the weekly cost report lands, and by then the hours are paid and the crew has moved on.

On a commercial solar job, the two-week lookahead says what each crew is doing. It does not say what that does to the labor budget. Six electricians on DC wire pulling and two on module install looks fine on the board. Against the budget, it may be too much labor on a code that was supposed to be nearly done and too little on the one that is behind.

Someone could sit down Thursday afternoon, map every crew line to a cost code, total the hours, and compare against the budget. Almost nobody does, because Thursday afternoon is already spoken for. So the check happens after payroll, when all you can do is explain it.

How the agent workflow works

The agent reads three inputs: the two-week lookahead, the cost code budget with hours earned to date, and the crew roster with each person's usual role and rate class.

Each week, once the superintendent updates the lookahead, the agent does the following:

1. Pulls next week's activities and the crews assigned to each.
2. Maps each activity to a cost code using a lookup table you maintain. "Pull string wire, Block C" maps to the DC wiring code, for example.
3. Estimates planned labor hours per code from headcount, shift length, and working days.
4. Compares the planned hours against the budget shape: how many hours remain on each code, and how much of that remainder next week would burn.
5. Flags codes where the plan runs well ahead of or behind the pace the remaining budget and remaining work imply.

The output is a one-page draft: a table of planned hours by code, a short list of flagged codes, and a plain sentence on each flag. For example: "Planned hours on trenching next week exceed what is left on that code, and the lookahead shows trenching running two more weeks."

The human checkpoint comes before anything moves. The superintendent reads the draft, corrects any bad mappings, and decides what to do about each flag: reshuffle the crew, request a budget transfer, or accept the overage knowingly. The agent never edits the schedule, never reassigns anyone, and never sends anything to payroll or the owner.

What you need to set it up

• An AI model that can read documents and spreadsheets, with a scheduled run or a simple trigger when the lookahead file changes.
• The lookahead in a consistent, readable format. A spreadsheet or exported table works. A photo of a whiteboard does not.
• A cost code budget export with hours budgeted, hours spent, and percent complete by code.
• A mapping table from schedule activities to cost codes. Building this is the main setup work.
• A crew roster listing who is on which crew.

Plan on several hours to build the mapping table and test the first few runs against weeks you already know the outcome of. Run it beside your manual process for a couple of weeks before trusting it.

Where it breaks

The mapping table is the weak point. Activities get renamed, split, or added mid-job, and the agent will either guess or drop them. A dropped activity makes a code look underloaded when it is not. Review unmapped items every week.

Planned hours are not worked hours. Weather, late deliveries, and a failed inspection will change the week by Tuesday. The forecast is a Thursday view, not a promise.

Percent complete is only as good as whoever entered it. If progress is overstated, the agent will say the budget has room when it does not. It checks the numbers it is given and cannot tell they are optimistic.

Shared crews are hard. A foreman who splits a crew between two areas, or a laborer who floats, will be allocated by rough assumption unless the lookahead says otherwise. Add split notes to the lookahead, or accept some error.

Rate differences matter if you budget in dollars rather than hours. This workflow is built around hours. Dollar forecasting needs rate data you may not want in the loop.

Finally, a flag is not a diagnosis. Heavy planned hours might mean the crew is overstaffed, or the budget was thin from the start. The agent cannot tell which.

The bottom line

This workflow moves the labor check from after payroll to before the week starts. It does not make the decisions. It does the mapping and arithmetic that nobody has time for and puts the exceptions in front of the superintendent while there is still time to change the plan.

Expect a rough first month while the mapping table settles. After that, the Thursday review is a short read instead of a skipped task.

Related use cases

• Daily logs that write themselves from delivery tickets Reporting
• Daily production reports, built by an agent Reporting
• Two-week lookahead, built not typed Management