ACC TRUST INSIGHTS | Claims Strategy & Digitalisation
AI can now assemble construction claims from correspondence, progress reports, programme updates and contractual provisions, producing a coherent narrative and an apparently persuasive case for additional time or money. For a team facing a deadline, that is attractive. It also creates a dangerous illusion: that because the claim is well written, it is well evidenced.
A claim is not established by language. It depends on what happened, when, what the contract required, the consequences and whether they can be demonstrated. Technology may accelerate that examination. It cannot change the history.
THE MISSING RECORD DOES NOT REAPPEAR AT SUBMISSION
Consider an instruction issued during construction that changes the sequence of works. The site team adapts, resources are moved and an activity finishes later than planned. Months afterwards, the contractor seeks an extension of time and prolongation costs. An AI system can search thousands of documents for the instruction, assemble a timeline and identify references to disruption. What it cannot do is establish the actual effect on the critical path if the contemporaneous programmes were not properly updated, the relevant constraints were never recorded, or the resource information needed to substantiate the cost was not retained.
An email reporting delay is not, by itself, a delay analysis. A spreadsheet showing additional labour hours does not establish that those hours arose from a compensable event. A chronology may have a neat beginning, middle and end. The project itself may have been considerably less tidy.
Fluent drafting can conceal gaps between an event and its alleged consequences. If each material assertion cannot be traced to a reliable source, the narrative is more polished without being more robust.
A CLAIM IS A CHAIN OF PROPOSITIONS, NOT A DOCUMENT
Effective claims management begins long before a claim is drafted. An event must be recognised, its contractual implications considered and any required notice addressed within the applicable framework. The developing consequences then need to be monitored and recorded. Entitlement, causation, delay and quantum each require their own analysis.
An event is not a notice. A notice is not a substantiated claim. A claim submitted is not a claim assessed or approved. These distinctions remain fundamental when software is introduced. A system that generates a convincing report but fails to preserve the status of notices, instructions, records and assessments may make the process look more complete than it is.
The first edition of the ICCP Claims Management Protocol (2026) addresses this directly. Section 12 treats AI as an assistive tool only: it may help analyse large volumes of records and draft narratives, but cannot replace professional judgement on cause, effect, entitlement or quantum. Its output must be verified against the contract, the law and the contemporaneous records, and AI must never be used to create or embellish evidence. Material use should leave an audit trail, and users should check confidentiality and data handling before entering project information into any system.
The Protocol is professional guidance, not an amendment to the contract. And retrieving a contractual clause is not the same as proving entitlement under it: the submission must show why the documented circumstances satisfy its conditions.
THE REAL OPPORTUNITY IS EARLIER IN THE PROJECT
The most useful question may not be how quickly AI can prepare a claim at the end of an event. It may be how technology can help the project team preserve a reliable account while the event is unfolding.
Imagine a structured project record in which an instruction is linked to the affected scope, the contemporaneous programme, the relevant correspondence, a notice register, progress measurements and subsequent commercial assessments. Documents retain their dates, versions, authorship and approval status. Changes in position are visible rather than silently overwritten. The purpose is not to predetermine entitlement, but to preserve the information from which entitlement and consequences can later be examined.
Here AI becomes more useful. It can identify inconsistencies between progress reports and payment applications, flag missing documents, highlight approaching contractual deadlines or question a programme update. But a flag is not a finding, correlation is not causation, and a proposed action is not an authorised contractual decision.
Retrieving evidence from fragmented archives at the end of a project is expensive and uncertain. Connecting events and records as work progresses demands discipline but reduces the need to reconstruct the past. That is a more enduring efficiency gain than drafting the final document faster. (See also: Why Claims Begin With Documentation.)
TRANSPARENCY MUST TRAVEL WITH THE ANALYSIS
The risks are not limited to invented references or inaccurate summaries. They include the quieter risk of selective completeness: a system may produce a plausible account based on the documents it was given, while omitting contradictory records that were never uploaded, were misclassified or fell outside the search.
Material AI-assisted findings should therefore remain traceable to identifiable source records, so a reviewer can inspect the original correspondence, programme update or cost record rather than an extract detached from context. Source references, version control, documented assumptions and a record of human review are part of making the analysis dependable. Human review must test the contractual basis, evidence and causal reasoning, not merely approve a finished report.
Confidentiality deserves the same attention. Claims often involve commercially sensitive prices, negotiating positions, personal data and privileged material. The convenience of a tool does not establish authority to disclose information to it.
THE SAME STANDARDS APPLY ON BOTH SIDES
These principles are not confined to contractors. Employers, Engineers and other contract administrators also face large volumes of submissions. AI can help organise material and identify disputed issues, but it cannot justify a rejection simply because it has found a clause that appears to support one. Nor should a polished assessment obscure evidence that favours the claimant. As we argued in The Engineer’s Determination Is Not a Defence of the Employer, a response should engage with the case actually presented and explain how the evidence informs the assessment.
The more demanding question is not “Can we draft claims faster?” but “Are we creating records that make a fair assessment possible?” Otherwise, faster drafting may simply accelerate uncertainty.
THE PRINCIPLE
AI can help us find the evidence, test connections, expose gaps and communicate the result. It cannot retrospectively issue a notice, conduct a missed inspection, recreate an accurate programme update or establish costs that were never properly recorded.
The future of claims management will not be decided by which system writes the most persuasive narrative. It will depend on whether organisations use technology to protect the integrity of the project record from the beginning.
AI can draft the claim. Only the project can create the record.
HOW ACC TRUST CAN SUPPORT
At ACC TRUST, we see digitalisation as a means of strengthening contractual continuity and commercial visibility, not replacing professional judgement. We support contractors, employers and project teams in structuring notice and claims registers, preserving contemporaneous records and setting practical rules for the use of digital tools in claims work. A better claim may be one outcome. A better-governed project is the more important objective.
Explore ACC TRUST services: https://acctrust.ro/services
Discuss a specific project: office@acctrust.ro
Reference: Institute of Construction Claims Practitioners, Claims Management Protocol, First Edition 2026, Section 12. Professional commentary only; the Protocol is good-practice guidance and does not replace contract terms, governing law or legal advice.