Review duplicate HVAC service requests without deleting evidence

Compare requests rather than customer identities

An office may receive a web form, a phone note and an email about apparently similar work. Start with approved request metadata: request ID, property reference, time received, customer wording, equipment reference when verified, active owner and recorded status. AI can flag candidate clusters for review without merging or closing anything.

This is different from cleaning duplicate client records. The customer may be correctly identified while multiple requests represent separate issues. Preserve each incoming report and its channel so the team can reconstruct what was received. Keep the company’s urgent-call procedure outside any waiting queue created for duplicate review.

Use a fictional same-property counterexample

At 10:00 a facility contact reports poor cooling in the east office. At 10:15 another contact reports water in a west utility room. Both belong to the same property, but those facts do not establish one issue. Keep them separate and route their relationship to the service owner. Do not infer the technical cause.

Now suppose a third report explicitly references the first request ID and says the caller is checking its status. The cluster table can record an explicit status-follow-up link with its source. That still does not authorize deleting the third contact or ignoring new information it contains. Use relation types such as explicit reference, possible same issue, distinct reported issue and unresolved.

Choose a coordinating record through human review

The service owner verifies the relationship and identifies the coordinating record under existing procedures. Record which incoming requests remain linked, who owns the next step and what information must be carried forward. If the organization closes a duplicate, the authorized person follows its normal retention and audit process; the assistant does not perform the closure.

Before creating a new dispatch action, check the verified active-work reference and let the dispatcher decide whether another appointment is needed. Do not promise that clustering prevents duplicate truck rolls. Pilot on fictional mixed examples and measure false grouping, lost details and unresolved ownership. A smaller request count is not evidence of a cleaner service queue if distinct problems disappeared.

Release checklist

• Keep every source request and channel reference.
• Separate explicit links from inferred similarity.
• Protect distinct reported issues at one property.
• Require human selection of coordinating records and next actions.

Source context and limits

Jobber documents request records and their workflow. Our candidate-cluster method is an internal review proposal, not a built-in duplicate-service detector or evidence that repeated contacts should be discarded.