The Report Said Non-Billable. I Asked Why.

The overlooked question

The report said the hours were non-billable. That was being treated as the end of the analysis. It should have been the beginning.

A system label cannot know whether a project meeting was out of scope, whether the wrong task was selected, or whether someone misunderstood the billing rule. The person leading the project often can. So before billing, I started asking a simple question: Should this actually be non-billable?

A synthetic non-billable system label becomes a question for the project leader, leading to a decision to keep the classification or correct it manually. — FIG. 01 A label starts the review; business context resolves it

The human control loop

I reviewed non-billable time on client projects, surfaced the edge cases, and asked the relevant project leader for the missing context. If the classification was right, it stayed. If it was wrong, I corrected the verified entry manually in the source system and checked the result.

The point was not to force every hour to become billable. It was to stop accepting a default classification without the business judgment needed to defend it.

Five-step human control loop: detect a questionable classification, ask the project leader, decide, manually correct a verified error, and verify the source record before repeating the process. — FIG. 02 The repeatable mechanism remains human-approved and manually executed

Taking operating ownership

I took ownership beginning with the October 2025 billing cycle. The review became one control inside a larger manual operating cadence: pre-checks, invoice generation, PDF preparation, edge-case review, Outlook drafting, batch delivery, and closeout reconciliation.

Faster execution mattered, but so did repeatability. I retrained project leaders on timely, accurate timesheet approvals and made the same questions part of each monthly pass. The process moved from roughly three weeks to under one week.

Chronology from the August 2025 baseline through October 2025 billing ownership and the January to May 2026 measured period, followed by the later owned data layer and June 2026 Billable Lens. — FIG. 03 October ownership precedes the measured post period and the later tooling

Six-stage monthly billing cadence from pre-check and generation through delivery and closeout, with cycle execution moving from about three weeks to under one week. — FIG. 04 A sanitized reconstruction of the owned monthly operating cadence

A surviving sent-output trail confirms 50 sent items within 124 seconds during the intended November 2025 run. That supports the batch-delivery step only. It does not establish total cycle duration or prove that a surviving script drafted or sent the messages.

The result—and the limit

Average monthly non-billable hours logged by delivery staff on client projects were 63% lower, from 509.0 in the August 2025 baseline month to 186.5 across the January–May 2026 post period—approximately 320 fewer hours per month. The share of logged time fell separately from 4.6% to 1.8%.

That is the claim the evidence supports. It does not support saying the difference was recovered, invoiced, converted to capacity, or converted to revenue.

The boundary matters for two reasons. First, 46.8% of the observed difference occurred on lump-sum projects, whose time entries do not generate invoice quantities. Second, the current database strongly associates T&M billable hours with invoice lines, but it does not retain the pre-correction values needed to identify which entries changed.

The 322.5-hour observed monthly difference splits into 171.7 hours on time-and-materials projects and 150.8 hours on lump-sum projects. — FIG. 05 Nearly half of the observed difference sits outside quantity-based invoicing

Current T-and-M billable-hour linkage is strong, but missing entry-change history prevents tracing manually corrected hours to invoice lines. — FIG. 06 Current association cannot reconstruct corrected-hour lineage

Six evidence limitations covering asymmetric comparison windows, current-state classifications, missing change history, missing invoice-creation timestamps, mixed billing types, and the observed-difference-only conclusion. — FIG. 07 The limitations travel with the result

The claim that failed

Every aggregate reproduced. The sentence did not.

The first compound headline joined 4.6% to 1.8% with −63%. The rate fell 61%, while average monthly non-billable hours fell 63%. The sentence had attached the right number to the wrong basis.

I kept the failure. The corrected claim attaches 63% to the hours comparison and reports the rate separately. A useful control should survive scrutiny; so should the story told about it.

FIG. 08 The hours result and rate result use different denominators

Later controls, separated from causality

Later, I built read-only tooling to make the control easier to inspect and the metric easier to reproduce. The owned data layer preserves the source fields and re-runs the aggregate checks. Billable Lens, added in June 2026, restores a missing billable-status indicator in the ERP interface.

Those tools are later iterations. They did not create the October 2025 result, and they did not replace the project leader’s judgment or the manual source correction.