Recurring reports and documents

Every organisation produces a family of documents on a fixed cycle and in nearly the same way each time: the monthly report for management or a client, the quotation built from the same sections, the minutes that follow the same running order, the periodic return for a public body. The content changes, the drudgery does not: data is gathered from three systems, pasted into a template, and the text around the numbers is brought up to date. That is hours of work which a well-built system reduces to minutes of review.

What the system does

• It gathers data at the source. From the management software, the databases, the shared spreadsheets — each figure from its authoritative source, defined once at design time. The most fragile step of all disappears: copy and paste.
• It assembles the document in your template. Tables, charts and text take their places in the layout you already use. The prose around the numbers — movement against last month, variances that need a comment — is proposed by the language model from the data, in the register of your previous documents.
• It checks before it assembles. Incomplete data, values out of range, periods that do not line up: anomalies stop the run and get reported, because a punctual but wrong report is worse than a late one.
• It delivers a draft, not a fait accompli. The document reaches whoever signs it with a record of what was generated and from which data. Review is quick because it starts from work already done — but it remains a review.

Typical cases

• Periodic reporting: project progress, status statements, accounts to clients or public bodies, monthly dashboards in document form.
• Quotations and offers: assembly from price lists and specifications, with the descriptive sections proposed from similar offers already made.
• Minutes and standard letters: from the recording or the notes of a meeting to a draft in the usual running order; repetitive correspondence personalised from the recipient's data.
• Mandatory documentation: documents required at fixed intervals by regulation or contract, where the structure is prescribed and the data sits in your systems.

Where the boundary lies

The system is most useful where the document is determined by the data. Where the content is original analysis — a strategic assessment, an opinion, a technical report on a unique case — the language model can prepare materials and first drafts, but the document remains an author's work. That distinction is drawn at the analysis stage, document by document: automating the mechanical part of a serious document is excellent practice; passing judgement off as automatable is not.

Frequently asked questions

Does the document come out ready to sign?

It comes out ready to read. The draft is complete — data, tables, text, layout — but signing off stays with people: on documents that leave the organisation, human review is part of the design, not an option. The time is saved on assembly, which is the long part, not on checking, which is the responsible part.

Does it keep our format and layout?

Yes: the system writes inside your templates — headers, styles, section structure — not in a format of its own. A generated document should be indistinguishable, in form, from one prepared by hand; if your current template is untidy, tidying the template is the first thing we do together.

Where does it take the data from?

From the systems where it already lives: management software, databases, spreadsheets, files exported from other programs and, where needed, from the documents themselves via the extraction described on the documents and data entry page. The project defines the authoritative source for each figure, so the report does not become a place where versions diverge.

What if one month the data is incomplete or odd?

The system notices before it writes: every run starts with completeness and plausibility checks on the data, and an anomaly stops the draft and reports what is missing, instead of producing an elegant, wrong document. That is the difference between automating a report and automating a mistake.