5 Sites, One Payroll: Batch
Construction Timesheets, No Manual Merge

Why Batch Timesheet Processing Is Not Just Single Processing Repeated

Extracting data from one timesheet is an accuracy problem — can the tool read the handwriting and place each value in the right column? Extracting data from 30 timesheets spread across five job sites is a consistency, merge, and validation problem. Once you cross into batch territory, three structural challenges emerge that single-sheet workflows never encounter.

First, cost codes diverge by site. A concrete crew on the civic-center job is billing against CSI MasterFormat Division 03, while the framing crew on the medical-office build is under Division 06. The same worker — a carpenter — might appear on both sites in the same week, and the timesheet from Site A codes his hours to 03 30 00 (Cast-in-Place Concrete) while Site B codes them to 06 11 00 (Wood Framing). Merge those two rows without the job-site prefix, and your job cost report silently misallocates labor. The Construction Financial Management Association (CFMA) identifies real-time labor cost tracking as a best practice for project profitability precisely because delayed or misallocated hours corrupt the historical data that future estimates are built on.

Second, wage rates vary by classification and sometimes by site. A laborer making $22/hour on a private commercial job might be entitled to $31.75/hour on a federally funded project under Davis-Bacon prevailing wage schedules. When timesheets from both sites land on the same payroll clerk's desk Friday afternoon, the merge step doesn't just add numbers — it multiplies each hour total by the correct rate for that worker on that site under that classification. One transposed digit and you've overpaid by a margin that catches a DOL auditor's attention.

Third, the merge itself has no native home in most construction accounting stacks. Sage 300 CRE expects one clean import file. Viewpoint Vista expects hours coded to specific job phases and cost types. Foundation and HCSS both want worker classifications aligned to their internal rate tables before the payroll batch processes. None of these systems offer a native "combine 30 handwritten sheets from five foremen into one import" button. That consolidation step lives in a spreadsheet that someone builds manually — and EY's 2022 HR Processing Risk & Cost Survey found that the average organization makes 15 corrections per payroll period, with inconsistent field mapping as the root cause of most of them.

The Certified Payroll Dimension: When Consolidation Errors Become Compliance Violations

On federally funded construction projects, the weekly payoff for getting the merge right isn't just clean books — it's staying on the right side of the Davis-Bacon Act. Contractors must submit Form WH-347 certified payroll reports that break down every worker's hours by labor classification. If a carpenter split his week between two job sites with different prevailing wage determinations, the WH-347 requires two separate line entries — and the DOL instructions are explicit: if you cannot produce that breakdown, you pay at the highest applicable rate for all hours.

That's not a penalty. It's the DOL's default position when documentation falls short. The Copeland Act (40 U.S.C. § 3145) mandates weekly reporting, and FAR 31.205-6 requires timekeeping systems that identify labor by intermediate or final cost objectives — meaning you need to show not just that a worker logged 40 hours, but which cost objective each of those hours served. When timesheets from multiple sites feed into one payroll report, the classification split between sites is the first thing an auditor checks.

The math is straightforward but unforgiving: if a laborer spends 20 hours at the private rate ($22/hour) and 20 hours at the prevailing wage rate ($31.75/hour), an assignment error that codes 30 hours to prevailing wage costs you an extra $292.50 in a single week for one worker. Across a crew of 15, that's over $4,300 — and on a multi-year public-works contract, the accumulated variance from classification misallocation can trigger a full wage-and-hour audit.

This is why multi-site consolidation isn't a convenience — it's a compliance requirement. The merge step has to carry each worker's hours, classification, site assignment, and cost code into the report without losing or misaligning any of those dimensions. When that merge happens manually in a spreadsheet, the reconciliation burden falls on the person who knows the least about what happened in the field: the payroll clerk.

How AI Batch Extraction Handles the Multi-Site Merge in One Pass

The limitation with template-based OCR systems is structural: they require you to define where on the page each field lives, which means a timesheet from Foreman A (who writes hours in a neat Monday–Sunday grid) needs a different template than one from Foreman B (who scrawls "Mon 8 Tue 7 Wed 8.5" in a single line). With a dozen timesheets across five job sites, you're not running extraction — you're building templates.

Semantic extraction takes the opposite approach. Instead of telling the tool where to look on the page, you tell it what to look for — column names like "Worker Name," "Date," "Regular Hours," "OT Hours," "Cost Code," "Job Site," "Classification." The AI reads each timesheet independently, locates each value anywhere on the page by understanding what it means rather than where it sits, and places the result in the correct column of a unified output table. When you upload 30 timesheets together, all 30 get processed in the same batch, producing one spreadsheet where every row is already aligned to the same column structure.

This is where "custom column extraction" — a term worth defining since it underpins the entire workflow — changes the consolidation math. Unlike template-based tools that require you to draw rectangles around each field on every new form, a custom column system lets you type the field names you need ("Worker Name," "Cost Code," "Regular Hours," "Job Site") and the AI locates each value anywhere on the page by reading the document semantically. For construction payroll, that means you define one set of columns — including cost code, classification, and job-site fields — and every uploaded timesheet, regardless of which foreman filled it out or what format was used, populates the same structured table. The merge step that used to consume hours of manual reconciliation now happens at extraction time.

A Five-Site Weekly Payroll Run: Step by Step

Here is what a concrete multi-site batch workflow looks like, using a mid-size general contractor running five active job sites as the scenario. The foreman at each site photographed the crew's weekly timesheet — paper forms, filled in by hand, with cost codes written in the margin on three of them. Total: 28 individual timesheets representing 4–6 workers per site, each sheet carrying between 5 and 7 daily hour entries.

The step that normally consumes two-thirds of a payroll clerk's Friday — the manual merge — doesn't exist in this workflow. The AI does the merge at extraction time because it's working from a single shared column definition across all 28 documents. The consistency that manual consolidation struggles to achieve is the default.

What Changes When the Merge Is Automated

The clearest signal that batch extraction changes the payroll workflow isn't the time saved — it's the questions that stop coming. When every timesheet row already carries its cost code, job site, and classification, payroll doesn't need to call the foreman Tuesday morning to ask which project a particular carpenter's Friday hours belong to. That back-and-forth, documented vividly by construction payroll professionals on forums, is where the real cost of manual consolidation hides: not in the data entry, but in the reconciliation loop that follows it.

For contractors running certified payroll, the change is even sharper. The WH-347 requires every worker's hours classified by trade and project. When the merge step preserves those classifications across sites automatically, the certified payroll report becomes a byproduct of the extraction — a view of the same data, not a separate document requiring separate data entry. The payroll clerk who used to spend one to three days processing a weekly payroll run — a figure reported by contractors who have since switched from paper — can complete the same run in under an hour.

There's a subtle but important distinction here: automated consolidation doesn't eliminate the need for review. A foreman's note that "Carlos covered for Miguel on Thursday" is context the AI can't infer from the timesheet alone. What it eliminates is the mechanical work of typing 140 daily hour entries into a spreadsheet while mentally cross-referencing five cost-code lists — the part of the process where transposition errors accumulate and where most of the payroll-period corrections documented in the EY survey originate.

Handling Cost Codes, Classifications, and Rate Differences Across Sites

In construction, labor is never just labor — it's labor assigned to a specific phase of a specific project, performed by a worker in a specific classification, paid at a rate that may differ by site. A concrete finisher's hours on the parking-garage project belong to CSI Division 03 at a private rate. The same finisher's hours on the county courthouse addition belong to CSI Division 03 at a prevailing wage rate $14 higher. If the payroll merge collapses both into a single row under "Concrete Finisher — 40 hours," the job cost report is wrong, the certified payroll report is wrong, and the estimate you build for the next parking-garage bid will be wrong because it's based on cost data that was wrong.

The solution isn't more careful manual entry — it's a column structure that captures the dimensions that matter before the merge happens. Instead of extracting only hours, add columns that capture classification, cost code, and job site as first-class fields. When the AI processes each timesheet, it populates those fields alongside the hour totals. The result is a table where every row is self-contained: you don't need external context to know that Row 14 represents a carpenter's 6 hours on the medical-office site, coded to CSI 06 11 00, at the Davis-Bacon carpenter rate.

For teams already allocating hours by cost code and job phase, batch extraction dovetails with existing workflows. The same column structure that handles a single-sheet allocation scales to 30 sheets without modification — because the AI applies the same column rules to every document in the batch, producing a merged output that's structurally identical to what you'd get from processing sheets one at a time, just without the manual step of copying rows from separate spreadsheets into a master sheet.

Where it gets genuinely interesting for multi-site contractors is with inferred columns. If every timesheet from the county courthouse project should be tagged with a specific prevailing wage determination number, you can define a column like "Wage Determination (infer from job site)" and the AI will apply the tag automatically when it identifies the job site on the timesheet. This isn't template matching — it's the AI reading the document context and applying a rule, much like a payroll clerk would, but in milliseconds rather than minutes per sheet. The same approach works for project numbers, contract codes, and any other field that can be inferred from data already present on the sheet.

For contractors who also receive handwritten timesheets in mixed formats, the same batch-processing approach applies — the AI's semantic understanding handles format variety without requiring separate templates per foreman, which is what makes multi-site consolidation achievable in a single pass rather than a multi-day reconciliation exercise.

Frequently Asked Questions

Can batch extraction handle different timesheet formats from different foremen?

Yes — and this is the core advantage over template-based OCR. Because the AI locates values by understanding what they mean rather than matching pre-defined positions, a timesheet with hours written in a grid and another with hours written in a single line both resolve to the same output columns. The only requirement is that the field values are legible. If a foreman consistently omits the cost code because "everyone knows which project Site D is," that column will show as blank in the output — the AI won't guess — but the other columns will populate correctly.

What happens if some timesheets are missing fields — no OT hours, no cost code written?

The AI leaves those cells blank in the output. It does not fabricate values to fill gaps. For missing overtime hours, blank is the correct default — assuming zero OT could underpay a worker. For missing cost codes, the blank cell flags itself for manual review in the preview table before export. You can spot and correct gaps while the data is still in a single consolidated view, rather than discovering them after import when the payroll system rejects the file.

Will the output integrate with Sage 300 CRE or Viewpoint Vista?

The export is a standard XLSX file with consistent column headers — the format that Sage 300, Viewpoint, Foundation, HCSS, and virtually every construction ERP accept as an import source. The key is that the column structure in the output matches what your ERP expects. If Sage expects "Job Code" and your extraction column is named "Cost Code," rename the column header in the output before import. The data is structured; the naming convention is yours to control.

Does this work for certified payroll / WH-347 reporting?

Yes, with the caveat that the AI extracts the data — it does not certify it. The WH-347 requires the contractor's signature on a Statement of Compliance attesting to the accuracy of the reported wages. What batch extraction provides is the structured data the form requires: worker name, classification, hours by day, rate of pay, and project identification — all merged across sites into a single output. The certification step remains the contractor's legal responsibility.

How does this compare to digital time-tracking apps like hh2 or WorkMax?

Digital time-tracking apps solve the problem by eliminating paper entirely — workers clock in on mobile devices, and hours flow directly to the ERP. Batch extraction solves a different version of the same problem: the paper already exists, and the organization isn't positioned to switch every crew to an app this week. The two approaches are complementary, not competing. A general contractor with an internal crew using a digital app may still receive paper timesheets from subcontractors whose workers aren't on that system. Batch extraction processes both digital-native and paper-source data into the same payroll report.

What's the practical limit on batch size?

Processing time scales roughly linearly with document count — 30 timesheets take approximately 30× the processing time of a single sheet. For a typical weekly payroll run of 20–50 timesheets, the batch completes in minutes. For runs exceeding 100 sheets, splitting into sub-batches of 30–50 speeds processing without changing the output structure, since each sub-batch exports to the same column layout and can be merged trivially in Excel.