How Quantity Surveyors Can Use AI to Identify Unrecovered Preliminaries in Final Accounts

Final accounts are where money walks out the door. You’ve got a stack of site diaries, a disputed EOT, three versions of the preliminaries schedule, and a contractor pushing for sign-off before Christmas. Somewhere in that pile is $40,000 in unrecovered prelim costs that your team has missed. It happens on almost every project. AI for preliminaries recovery construction is changing that — not by replacing QS judgement, but by doing the grunt work of pattern-matching across hundreds of documents your team simply doesn’t have time to read twice.

⬢ Workflow Diagram

flowchart TD
    A["Final Account Submitted"] --> B["Gather Preliminaries Data"]
    B --> C["AI Analyzes Cost Records"]
    C --> D{Unrecovered Prelims Found?}
    D -->|Yes| E["QS Prepares Recovery Claim"]
    D -->|No| F["Approve Final Account"]
    E --> G["Contractor Negotiation"]
    G --> H["Settled & Signed"]
    H --> F

Why Prelims Always Get Underrecovered — and What AI for QS Workflows Actually Fixes

At 4:30pm on a Thursday when your final account submission is due Friday morning, nobody is methodically cross-referencing the site establishment costs against the extended programme. That’s the reality. Prelims are the easiest cost bucket to leave on the table because the evidence is spread across dozens of document types — daily reports, RFIs, superintendent instructions, variation logs, wet weather records, and subcontractor delay notices — none of which talk to each other.

The specific categories QS teams routinely miss include extended site establishment during EOTs, additional crane and hoist usage triggered by design changes, increased site management costs from scope creep, and temporary works associated with unresolved RFIs. These aren’t obscure line items. They’re in your documents. The problem is retrieval under deadline pressure.

AI document analysis tools fix this by ingesting your project corpus and running structured queries against it. Instead of a junior QS manually grepping through 200 site diaries looking for references to the tower crane, you run a prompt in two minutes and get a consolidated list of every date the crane was used beyond the contracted programme, with source references.

how to set up a QS document management system for AI analysis

Tools worth knowing:

ChatGPT with file uploads (free tier available; GPT-4o from $20/month) — Best for QS teams starting out who want to interrogate PDFs without a technical setup.
Klippa DocHorizon (custom pricing, free demo available) — Best for larger contractors needing automated extraction from structured cost documents and site records at scale.
Draftwise (from $49/month) — Best for contract-specific clause analysis alongside cost recovery work.

How to Run a Prelims Recovery Audit Using Construction Final Account AI

preliminaries_recovery_analyzer.py

# AI Preliminaries Recovery System v2.1
# Analyzing Final Account Claims for Unrecovered Costs

from preliminaries_analyzer import PreliminaryClaimDetector
from contract_terms_engine import ContractConditionsParser
from cost_variance_module import UnrecoveredCostIdentifier
from ai_claim_validator import EvidenceMatchingEngine
from document_classifier import PreliminaryItemCategorizer
from timeline_correlator import DelayImpactAnalyzer



# Scanning final account documentation and contract schedules...

✓ Contract preliminaries loaded: 47 line items identified
! Missing supporting documentation for Site Establishment Phase (3 items flagged)
✓ Delay periods cross-referenced with preliminary cost allocations
! Unrecovered mobilization costs detected: £18,500 (requires further validation)
✓ Cost variance analysis complete: 12 preliminary claims vs. actual expenditure matched
✗ 2 items lack sufficient daily report evidence - recommend manual QS review

During your next final account close-out — ideally in the week before the reconciliation meeting with the superintendent — this is the process that surfaces the gaps.

Step 1: Assemble your document pack — Gather the signed contract preliminaries schedule, the as-built programme, all superintendent instructions, the EOT determination (if issued), and at least 80% of your daily site reports. Gaps in the site diary record will create gaps in your recovery analysis, so push the site supervisor for anything missing before you start.

Step 2: Upload to your AI tool and establish context — Load documents into ChatGPT (GPT-4o with file uploads) or a similar tool. Open with a context-setting prompt that identifies the project, contract type (e.g. AS4000, NEC3), and the specific prelims categories you’re targeting.

Step 3: Run targeted extraction queries — Ask the AI to identify every reference to specific prelim items — tower crane, site manager, traffic management, temporary fencing — and log the date, document source, and any associated cost or programme impact.

Step 4: Cross-reference against your contracted rates — Compare AI-extracted usage data against your preliminaries schedule line items and identify dates of operation that fall outside the original contract period or are linked to client-instructed variations.

Step 5: Build the recovery narrative — Use the AI to draft a summary that connects each unrecovered cost to its contractual basis — extended prelims under clause 34, or a specific superintendent instruction number.

Step 6: Sense-check with the cost plan — Run the recovered amounts against your cost-to-complete and flag anything that looks anomalous before it goes into the claim.

Using AI to Extract EOT-Linked Prelim Costs from Site Records

When the 9am EOT meeting wraps up and you’ve got an agreed extension of 6 weeks, the clock starts on your delay damages and extended preliminaries claim. Most QS teams calculate the headline daily prelim rate and multiply it by 42 days. That’s the floor, not the ceiling.

AI tools let you go deeper. Feed in the site diaries for the extension period and ask the tool to identify any additional cost events beyond the baseline rate — extra security shifts during the extended defects period, concrete pump hire that ran longer because of the delayed handover, additional utility connections that were held pending the superintendent’s instruction.

On a recent commercial fit-out project (eight-storey CBD building, 14-month programme extending to 17 months), this approach surfaced $67,000 in extension costs that had been absorbed into the contingency rather than claimed. The evidence was in the site diaries and the subcontractor notices. It just hadn’t been connected to the EOT entitlement.

Try this prompt:

You are assisting a quantity surveyor with a preliminary cost recovery analysis for a construction project under AS4000. The project has an agreed EOT of [insert weeks]. I am uploading the following documents: daily site reports for the extension period, the approved preliminaries schedule, and the EOT determination. Please identify every reference to plant, labour, and site overhead costs that occurred during the extension period but are not captured in the standard daily prelim rate. For each item found, record the document name, date, cost reference or trade involved, and the likely preliminaries category it belongs to (e.g. site management, mechanical plant, temporary services). Present results as a table.

This prompt works in ChatGPT with file uploads, Claude (free tier available; Pro from $20/month — best for large document sets due to its 200K context window), and Gemini Advanced (from $19.99/month — best for Google Workspace-based QS teams).

Scanning Variation Logs and RFIs for Hidden Prelims Cost Recovery Construction Claims

At the end of a project review meeting, when the variation register is being reconciled, someone always asks: “Did we capture the prelims associated with that RFI?” The honest answer is usually no — or at least, not completely.

RFIs and superintendent instructions generate variation work, and that variation work has a prelims tail. A design change that required the structural steel sequence to be reordered didn’t just cost you the steel resequencing — it cost you two weeks of crane standing time, additional site management coordination, and revised traffic management because the delivery sequence changed.

AI tools are particularly strong here because they can trace a chain of events across document types. Give Claude or ChatGPT your full RFI log and ask it to identify every RFI that had a programme impact, then cross-reference those RFIs against the daily reports to find prelim costs that were incurred during the period the RFI was open and unresolved.

how to write a stronger variation claim using AI-assisted documentation

Use this template:

Review the attached RFI register and site diary records for [Project Name, e.g. Riverside Mixed-Use Stage 2]. Identify all RFIs with a status of ‘open for more than 10 days’ that coincide with programme-critical activities. For each identified RFI, list: RFI number, date opened, date closed, the trade or work area affected, and any references in the site diaries to plant, labour, or site overhead costs incurred during the period the RFI was open. Flag items where the variation associated with the RFI does not include a preliminaries component.

This workflow consistently surfaces 3–7% of contract value in underrecovered prelims on complex projects. Not all of it will be recoverable depending on your contract, but you need to know it exists before you sign off the final account.

Frequently Asked Questions

Can AI tools actually read and understand construction contract documents?

Yes — modern large language models handle PDF and Word documents well, including contracts written in AS4000, NEC3, FIDIC, and GC21 formats. They can identify clause references, cost categories, and programme impacts in context. The quality of output depends heavily on the quality of your prompts and the completeness of your document set. Start with well-structured documents and specific queries rather than uploading everything at once.

Is AI for preliminaries recovery construction reliable enough for a formal claim?

AI is a retrieval and analysis tool, not a signatory. The output needs QS review and professional judgement before it goes into a formal claim. Use it to surface evidence and draft structure, then apply your own contract interpretation. Think of it the same way you’d use a cost report generated by your project management software — useful data, human sign-off required.

What’s the biggest risk of using AI for final account analysis?

Incomplete document sets. If your site diaries have three weeks missing or the superintendent instructions weren’t saved to the project folder, the AI will work with what it has and won’t flag what’s absent. Garbage in, garbage out. Audit your document pack before you run any AI analysis — a quick index check takes 20 minutes and significantly improves output quality.

Which AI tool is best for QS teams working on large infrastructure projects?

Claude Pro (from $20/month) handles the largest document volumes due to its 200,000 token context window, which equates to roughly 150,000 words of document content in a single session. For infrastructure projects with multi-year site diary records and complex EOT histories, this matters. ChatGPT is more familiar for most users but has lower context limits on standard tiers.

Conclusion: Three Things to Do Before Your Next Final Account

First, audit your document pack before you run any analysis — AI is only as good as the documents you feed it. Second, use structured prompts that target specific prelim categories and reference your contract type — vague queries get vague results. Third, don’t stop at the daily prelim rate in EOT claims — the real recovery is in the event-specific costs buried in RFI logs, superintendent instructions, and site diaries that nobody had time to read at 4:30pm on a Friday.

Prelims recovery isn’t glamorous QS work, but on a $10M project it’s often worth $150,000 or more in legitimately claimable costs that get abandoned under deadline pressure. AI doesn’t replace your judgement on what’s recoverable — it just makes sure you’ve actually found everything before you decide.

explore more QS workflow guides and AI tools for construction cost management

If you want practical guides like this one delivered to your inbox — covering AI tools, QS workflows, and construction cost management — subscribe to the ConstructionHQ newsletter and stay ahead of how the industry is actually using these tools on real projects.

AI for Preliminaries Recovery Construction: QS Guide