AI EMAIL REVIEW AGENT

The sentence that quietly
becomes a commitment.

Construction disputes are rarely built on contracts alone. They are built on ordinary emails: a date promised without conditions, a duration read as a guarantee, a claim keyword answered casually. The agent reviews outbound correspondence before it sends.

  • Reads: Draft, thread, role, scenario
  • Detects: Commitments + claim triggers
  • Produces: Verdict, findings, rewrites
  • You confirm: Every email that leaves

Talk to us

How this job is done now,
and what it costs later.

  • Coordinators write under time pressure: A helpful, friendly update is written between site visits, with no legal review and no time for one.
  • Dates become promises: "Scheduled for delivery before Wednesday" reads as a commitment the supplier never gave you.
  • Durations become guarantees: "About two weeks" is an estimate to the writer and an entitlement to the reader.
  • Claim language slips past: Words like variation, delay or formal notice carry procedural consequences that a casual reply can forfeit.
  • The wrong recipient: A tenant is not a contracting party. Sending direct can breach the notice path the contract sets out.

Trigger → assess → flag → your confirmation.

  1. Trigger: A coordinator drafts an outbound email and submits it with the prior thread, their role and the scenario.
  2. Assess: The agent runs the rule library over the draft: commitments, conditions, claim keywords, recipient boundary and approval requirements.
  3. Flag: It returns a verdict, an itemised issue list with rule codes, and a conditioned rewrite for each risky sentence.
  4. Confirm: The sender edits and sends, or routes to management and legal where the verdict requires it. The agent never sends.

What the reviewer sees.

The agent does not return a score. It returns the sentence it objected to, the rule that caught it, the reason in commercial terms, and a version that says the same thing without creating the exposure.

Outbound email review inside DDDI Group's management system. The interface, the rule codes and the limitations panel are the real ones; the draft is ordinary site correspondence with nothing identifying in it.

Three severities.
One of them stops the send.

The verdict routes the email. It never sends one, and it never edits one silently.

VerdictWhat it meansWhat happens next
BLOCKA claim trigger, a procedural obligation or an unapproved commitmentThe email is held; management or legal approval is required
REVIEWA commitment that needs conditioning before it goes outThe sender applies the rewrite, or accepts the risk knowingly
MINORWording that could be clearer but carries no exposureOptional; the sender decides

It also says what it could not check.

The review limitations panel is not a disclaimer bolted on at the end. It is part of the output, because a reviewer who cannot see the contract, the programme or the approved scope should say so rather than imply the draft was verified against them.

  • Without project facts: Dates, durations and scope in the draft cannot be checked against the programme, contract or start notice. The agent says which checks it therefore could not run.
  • Without the thread: Fact-checking, the commitment ledger and escalation detection all degrade, and the interface warns the sender before they submit.
  • Recipient authority: Whether direct contact is permitted, and in what form, is flagged as unconfirmed rather than assumed.

Built on the document intelligence layer,
running inside a live group.

The agent extends the same parsing layer that reads drawings and contracts, applied to correspondence. It runs today inside DDDI Group's management system, where the coordinators sending these emails work.

Ships the way every Cyberate agent ships: shadow first, then assisted, then routine, with human confirmation designed in from day one.

Document intelligence · Variation & claims agent

Common questions.

What is the Cyberate Email Review Agent?

The Cyberate Email Review Agent is a construction AI agent that reviews outbound correspondence before it is sent, flagging unconditioned commitments, claim and notice triggers, and recipient-boundary risk, and offering a conditioned rewrite for each finding.

Does it send or edit emails?

No. It returns a verdict, findings and suggested wording. The sender edits and sends, and a BLOCK verdict routes the draft to management or legal approval first.

Is this legal advice?

No. It applies a configured rule library to wording and process, and escalates to your people for the legal and commercial judgement.

What if it flags something that is fine?

Then the sender sends it anyway. REVIEW findings are advisory; only BLOCK holds the email, and the rule library is yours to tune.

Does it read our whole mailbox?

No. It reviews the draft and the thread a sender submits, inside the agreed data boundary.

Say the helpful thing
without saying the costly one.

Discuss your correspondence risk · How we deploy AI