Publication week in the research office
The quarterly survey is closed and cleaned. Now the bulletin. The analyst opens last quarter's Excel workbook, pastes in the new figures, and checks every chart range has picked them up. Some have not. Two charts need a new data point added by hand. The tables are rebuilt and then copied into the Word template, or sent to the designer to set in InDesign. The commentary, 'output rose for the third consecutive quarter, led by...', is rewritten around the new numbers, and every number in the text is checked against the tables.
Then the web version: a separate page, with images of the charts exported one by one, and a download of the data tables that has to be kept in step with the PDF. And the press release, which quotes figures that must match both.
By the time it is out, the analyst has spent a week producing a document that, structurally, is identical to last quarter's.
Why the bulletin is rebuilt every quarter
- Charts are Excel objects tied to cell ranges that must be extended by hand.
- The layout lives in Word or InDesign, separate from the data, so every figure is pasted in.
- Commentary numbers are typed, so each has to be checked against the tables.
- The PDF, web page, data download and press release are separate outputs with separate copies of each number.
- Revisions to earlier quarters mean hunting down every place an old figure appears.
None of this is analysis. The analysis is deciding what the figures mean for the sector, and that is the part that gets squeezed.
What the production week costs
The research team's time, above all, spent on formatting rather than insight. Publication later than it could be, which matters when journalists and officials want the figures while they are news. And a steady risk of a mismatch, such as a number in the commentary that no longer matches a revised table, which undermines trust in the statistics when someone notices.
A bulletin pipeline from dataset to publication
We build a pipeline that treats the bulletin as output from the data, not a document to be edited.
- The approved quarterly dataset, after your confidentiality checks, is the single input.
- Each chart and table is defined once in code (Python or R), styled to your house style, and regenerated from the dataset every quarter, including revisions to back data.
- Key figures (changes on the quarter and year, highs, lows) are calculated and held as named values, so every place a number appears uses the same value.
- An AI model drafts factual commentary from those named values, limited to describing what the numbers show; the analyst rewrites it with the sector interpretation only your team can give.
- The pipeline outputs a PDF from a template, a web page with interactive or static charts, and downloadable tables, all from the same run.
- A check step lists every number in the commentary and confirms it matches the calculated values before release.
| Output | Before | After |
|---|---|---|
| Charts | Excel ranges extended by hand | Regenerated from the dataset |
| Tables | Copied into Word or InDesign | Generated in house style |
| Commentary figures | Typed and checked by eye | Pulled from calculated values |
| Web version | Built separately with exported images | Produced in the same run |
| Revisions | Hunted through every output | Rerun the pipeline |
If your designer needs to keep control of the print layout, the pipeline can produce the charts and tables as files for InDesign instead of a finished PDF.
Bulletin week, afterwards
The analyst runs the pipeline on the approved dataset and gets a complete draft bulletin, web page and data download. The time goes into reading the figures, talking to the policy team and writing what the numbers mean. Revisions are a rerun. The press release pulls its figures from the same values, so it cannot disagree with the bulletin.
Is your bulletin built by hand?
- Chart ranges are extended manually each quarter.
- Figures are pasted from Excel into Word or InDesign.
- Numbers in the commentary are checked against tables by eye.
- The web version is produced separately from the PDF.
- Revisions mean editing several documents.