In brief A recurring report pack is mostly a set of numbers from known sources, which is why it automates well. Pull every input on a schedule into one staging sheet, fill the fixed tables by rule, draft the commentary from the numbers for a person to edit, and flag only what moved more than usual. The reviewer approves a finished pack instead of spending days building one.
-
Inventory the pack List every table, chart and paragraph in the current pack and where each number comes from. A report pack is really a list of sources, and most of the build is writing that list down.
-
Automate the gathering Each source is pulled on a schedule, whether that's an export, an API or a spreadsheet. The numbers land in one staging sheet with timestamps, so there's never a question about how current they are.
-
Template the fixed, draft the variable Tables and charts fill from the staging sheet by fixed rules. Commentary is drafted from the numbers, with changes since last period called out, for a person to edit rather than accept as written.
-
Flag exceptions, not everything Thresholds mark anything that moved more than usual, so the reviewer reads three flags instead of forty pages. Those flags are where the reviewer's time is worth most.
-
Approve and distribute One reviewer approves, then the pack is produced in its final format and sent where it needs to go, with a log. Putting the pack together takes about as long as reviewing it.
Recurring report packs aren’t really written each month. They’re put together from
the same sources in the same order, and putting things together on a schedule is what
automation does well. The steps above turn a monthly slog into a review.
| The build at a glance | |
|---|
| Best first candidate when | A recurring pack eats days each cycle |
| Human gate | One reviewer approves the assembled pack |
| Never through a model | The tables: they fill deterministically |
| Measured by | Assembly days becoming a review hour |
| Typical shape | Fixed-price Kick-starter |
Start with the inventory
The inventory is most of the project. Firms tend to underestimate this step, then
discover that the pack’s forty numbers come from eleven places, and two of those are
someone’s memory. Writing that map down is most of the work, and it’s where you learn
how the pack really gets made.
Rules for the tables, drafts for the words
Tables never go through a language model. They fill from the staging sheet by fixed
rules, so the numbers are always the numbers. The written commentary is drafted by AI
and clearly marked for editing. Keeping those two apart is what keeps the pack
trustworthy, and it’s the same split the
reconciliation pattern uses.
Flags turn reading into reviewing
Exception flags are where senior time comes back. The pack shows what moved, and the
person reviewing it says what that means.
