Proof Pack

The magic is inspectable.

A prepared morning brief should feel effortless to the team. Underneath it, the source layer needs to be visible: what arrived, what failed, what was caveated, and what a human still approved.

What a pilot has to prove

Reliability should have artifacts, not vibes.

A reporting automation pilot should prove more than "a script can make a file." It should prove the workflow knows its inputs, names its caveats, refuses unsafe guesses, and can be handed to an operator without mystery.

01

Source inventory

The exact reports, systems, owners, timing, required status, and business-date rules before anything is built.

02

Run log

A readable record of each run: files processed, warnings, missing inputs, output links, and approval state.

03

Exception behavior

Rules for late files, duplicate files, changed columns, wrong dates, stale data, and unsafe AI summaries.

04

Sample brief

The morning output itself: metrics, exceptions, caveats, and first moves in a format the team will actually read.

05

QA matrix

Concrete tests for happy path, missing required file, missing optional file, duplicate conflict, wrong date, and AI caveat behavior.

06

Handoff README

Operating notes covering what the workflow does, what it does not do, how to rerun it, and when to escalate.

Source inventory

The first drawer is the list of inputs.

Most hotel reporting workflows are described too loosely. The inventory turns "we pull the PMS report and a few spreadsheets" into explicit operating inputs.

PMS on-the-books Expected 7:00 AM. Required for occupancy, ADR, revenue, RevPAR, and pickup sections. Required
Rate shop export Expected 7:15 AM. Supports rate-position notes and gets caveated when late. Optional
Forecast workbook Expected 7:20 AM. Required for forecast variance and pace context. Required
Budget context Static or weekly file used for budget gaps, month-to-date commentary, and owner-facing context. Context

Run log

The run log is the trust layer.

Required PMS export received.

Business date matched the control cell. Original file archived before cleanup.

Pickup report processed.

Column names matched expected schema. Seven-day pickup section cleared.

Rate shop late.

Optional section allowed to ship with caveat. Rerun path recorded for the team.

Brief composed with caveat.

AI summary held to source-backed statements and warning language carried into the output.

Tests before handoff

The expected failures are part of the build.

A pilot is not production-like until the ugly morning scenarios are tested. The goal is not to hide bad inputs. The goal is to make them impossible to miss.

01

Missing required PMS report

Output blocks or blanks dependent KPIs. Prior-day values are not silently reused.

02

Missing optional rate shop

Brief can still ship if safe, but rate-position notes are visibly caveated.

03

Duplicate conflicting file

Both raw files are archived and the workflow flags conflict instead of overwriting quietly.

04

Wrong business date

Date mismatch blocks or caveats affected sections before interpretation runs.

05

Column renamed upstream

Required field mapping fails visibly. Blank or zero is not treated as a real number.

06

AI summary with missing source

The summary must cite missing inputs and avoid unsupported recommendations.

Next drawer

Bring the actual reports.

A workflow review is useful when we can inspect the real source reports, timing, owners, output, and failure points.