The structure is automatable. The judgement is not. You stay the author.
Last updated
Zepth Core module
Construction documents are formulaic in structure and unforgiving in content. A method statement has a shape. An inspection and test plan has a shape. An extension-of-time claim has a shape, and a very particular one, and everybody in the industry knows what it is.
That shape is the part you spend your evenings on — and it is the part a machine can assemble. The Studio drafts the structure and populates it from your project’s actual records. Your engineer, your QS, your safety manager reads it, corrects it, and signs it. The first draft is the machine’s. The document is yours.
Nobody in construction is short of things to say. They are short of the eleven hours it takes to say them in the format the consultant will accept.
An ITP has to enumerate the hold points, the witness points, the acceptance criteria and the references — and every one of those is already in the specification, which somebody is now reading with a highlighter at nine in the evening. A delay narrative has to walk an assessor through the events in sequence, with the records cited — and every one of those records is already in the daily reports, the instructions and the correspondence. The information exists. The document does not, because writing it is a fortnight nobody has.
So the work gets done badly, or late, or by the most expensive person available. The claim goes in thin. The ITP gets copied from the last project and quietly fails to match this specification. And the method statement is a template with the project name changed, which is exactly what the safety officer suspects when he reads it.
A drafting tool that writes from nothing produces plausible construction prose, which is the most dangerous document type there is: fluent, confident, formatted correctly, and untethered from your project.
The Studio drafts from the connected record. The delay narrative comes from the diaries and the programme. The claim analysis comes from the instructions, the notices and the contemporaneous records. The ITP comes from the specification sections that actually apply. And every draft cites what it used — so the first thing your engineer does is not “is this true?” but “is this the right record?”, which is a question a human can answer in seconds.
That is the line between drafting and inventing, and it is not a subtle one. A document assembled from your records, with the records named, is auditable. A document generated from a prompt is a well-written guess.
Quality and safety. Inspection and test plans from the specification. Method statements and risk assessments for the actual activity, not a template with the project name changed. Quality plans, HSE plans, and toolbox-talk content for the hazards this week actually contains.
Commercial. Claim analyses and delay narratives assembled from instructions, notices, diaries and the programme. Variation drafts. Formal letters, built from the correspondence register and the records a position actually rests on — rather than from somebody’s recollection of them at eight in the evening.
Procurement. Bills of quantities, tender documents, and bid comparisons — the documents whose structure is rigid, whose content is project-specific, and whose preparation is the reason procurement teams work weekends before a float.
Reporting. Progress reports, mobilisation plans, manpower histograms. The recurring documents that are mostly assembly, and that consume a disproportionate share of the time of people who should be doing something else.
Nothing issues without sign-off. Every draft carries its sources, so the review starts from the record rather than from the prose — and it flags where it had to infer rather than find, because the gaps in a draft are the most important thing in it.
The engineer is the author. That is not a legal disclaimer; it is the design. A document that goes to a client with your company’s name on it was written by your company, and the Studio’s job is to make sure the person who signs it spent their time on the judgement rather than on the formatting.
And where it pays first is exactly where the pressure is: the EOT claim skeleton, assembled from the records in minutes instead of a fortnight — which changes not just the cost of making a claim but whether a thin, under-evidenced one gets made at all.
The evening work — the formatting, the assembly, the enumeration — stops being human work.
Every draft cites the records it was built from, so review is verification rather than authorship from scratch.
The EOT claim skeleton is assembled in minutes, which changes whether a well-evidenced claim gets made at all.
The person signing the document spent their time on the judgement, which is the only part that needed them.
ITPs from the specification, method statements and RAMS for the actual activity, quality and HSE plans, toolbox content for this week’s hazards.
Claim analyses and delay narratives from the instructions, notices and diaries. Variations. Formal letters built from the register rather than from memory.
BOQs, tender documents and bid comparisons — rigid structure, project-specific content, and the reason procurement works weekends.
Progress reports, mobilisation plans, manpower histograms. Mostly assembly, and disproportionately expensive in people’s time.
Each draft names the records it used, and flags where it inferred rather than found. The gaps are the most important part.
Nothing leaves without a human author. Then it enters the document register and the transmittal machinery like any other document.
ITP, method statement, delay narrative, formal letter, BOQ, progress report. The shape is known; that is the point.
From the specification, the diaries, the instructions, the correspondence — and it asks for what it cannot find rather than inventing it.
Every claim in the document points at the record it came from. The first review question is “is this the right record?”, which takes seconds — not “is this true?”, which takes an evening.
The engineer, the QS, the safety manager. They are the author. Nothing issues without them.
Into the document register, the transmittal, the correspondence register. A drafted document is still a document, and it is governed like one.
From the specification, the diaries, the programme, the instructions, the correspondence. Not from a prompt and a general knowledge of construction — which produces fluent, confident, plausible prose about somebody else’s project.
Every draft names the records it used. Which turns the review question from “is this true?” — a long evening — into “is this the right record?”, which a human answers in seconds.
Where the record was silent and the draft had to infer. This is genuinely the most valuable output: an unrecorded fact is a hole in your project record, and you would rather find it now than in cross-examination.
The EOT narrative assembled from what actually happened, in sequence, with the records cited — so the question becomes whether the claim is good rather than whether anybody has a fortnight to find out.
The engineer’s judgment stays in charge; the AI removes the latency and the blind spots.
Documents drafted, by type and by project, with the records each one drew on. The gaps the Studio could not fill from the record — which is a report about your project record rather than about your documents, and it is the more useful of the two. Review and sign-off status. And the drafted documents flowing into the register and the transmittal machinery, governed like every other document, because that is what they are.
The formulaic ones, which is most of them: inspection and test plans, method statements and risk assessments, quality and HSE plans, toolbox content; claim analyses, delay narratives, variations and formal letters; bills of quantities, tender documents and bid comparisons; progress reports, mobilisation plans and manpower histograms. The common thread is a rigid structure and project-specific content — the structure is assembly, and assembly is what you should not be doing at nine in the evening.
Read the full answerFrom your project’s connected records — the specification, the daily diaries, the programme, the instructions, the correspondence, the registers. Not from a general model of what construction documents usually say, which produces fluent and confident prose about somebody else’s project. Every draft cites the records it used, and that citation is the whole difference between drafting and inventing.
Read the full answerIt can assemble the skeleton — the events in sequence, the records cited, the narrative structured the way an assessor expects to read it — from the diaries, instructions, notices and programme, in minutes rather than a fortnight. What it cannot do, and must not, is form the legal and delay-analysis judgement. That is your planner and your commercial lead. What changes is that they now spend their time on the argument instead of on finding the records.
Read the full answerNo. Nothing issues without sign-off, and the engineer, QS or safety manager is the author — not the reviewer. This is design, not a disclaimer: a document with your company’s name on it was written by your company, and the person who signs it owns every word in it.
Read the full answerEvery draft names the records it drew on, and flags where it had to infer rather than find. Read the citations before you read the prose — a document that reads beautifully and cites the wrong records is considerably more dangerous than one that reads badly, because nobody checks the beautiful one.
Read the full answerThen it says so, rather than filling the gap. And that is arguably the most valuable thing it does: a gap the Studio could not fill from your records is a hole in your project record, and it was going to hurt you in a dispute whether or not you ever drafted the document. Better to discover it on a Tuesday than in cross-examination.
Related modules
Related answers
Zepth is the construction project delivery platform — it runs construction, procurement and asset management on one record, and does the work: reading the drawings, reviewing the submittals, matching the invoices and flagging the risks, with a human sign-off on anything consequential.
A short, tailored walkthrough on your real workflow — no generic demo.