Fictional plumber reviews a work note and a plumbing part in a workshop
AI-generated image; illustrative scene.

The part is ordered: can you confirm the return visit yet?

Organise a plumbing return visit by separating the order, part receipt and confirmed appointment, with human oversight.

Adaris Tech Team7 min

Ordered does not mean received or appointment confirmed

A plumber finishes an initial visit and records that a part is needed. The office orders it, but the supplier has not confirmed delivery. If the job is marked ready immediately, someone might propose a date without knowing whether the part will be available. Software adapted to the business could bring the visit note, order and appointment status together for review; AI could help locate details and questions without closing the job or booking on its own. For a related guide, see “A supervised AI proposal for extra plumbing work”.

The technician decides which part is appropriate and what work to perform. The office checks receipt and availability against its own records and agrees the appointment with the customer. A generated note cannot replace those checks.

A fictional example: two jobs and a partial delivery

Imagine two plumbing jobs. For A-14, the visit note says “part R-21 needed, reference to be confirmed”; purchasing records an order for one R-21 and the supplier gives an estimated date. For A-15, two R-35 parts are on order. A delivery note for one R-35 arrives, with the second still pending according to the order; an unnumbered note says “the part has arrived”. Every reference and document in this example is fictional.

A person prepared the reference output below; it is not a tested AI response. The R-35 delivery note does not establish that R-21 arrived and should not be assigned to A-14 simply because both jobs are open. An estimated date is not receipt.

What to retain before arranging the return
JobRecorded informationNext check
A-14R-21 requested; technical reference unconfirmed; estimated supplier date.Technician confirms reference; office checks receipt and then agrees a visit.
A-15Two R-35 parts ordered; delivery note for one, second pending according to the order.Office checks the delivery note against the order and job, then availability.
UnassignedNote: “the part has arrived”, without an identifier.Ask the author; do not assign by proximity.

From visit note to follow-up: a possible custom workflow

First, a reviewed identifier links the job, visit note, part reference and order. Each document keeps its source and date. Assisted reading could extract statements and flag missing or conflicting references. The responsible person checks the draft against the originals before saving it. If the match is unclear, the note stays unassigned.

The system could then show “requested”, “ordered”, “receipt checked” and “appointment agreed” separately, with the person and date for each change. A partial delivery does not make the whole order received. Receipt does not prove compatibility or reservation for that job. After checking material and the diary, the office prepares a message for a person to review and send; it records the appointment only after confirmation.

ServiceVisit documents separate operations in its own product for reserving stock, recording use and releasing unused reservations. Microsoft describes the relationship between work orders, inventory and purchasing in Dynamics 365 Field Service. These are examples of record keeping, not existing Adaris connections to those products.

When an agent could help and what access it needs

If information is spread across visit notes and orders, Adaris could assess an application that reads authorised sources and prepares a list of pending jobs for review. Any order or diary connection depends on the actual systems and their capabilities. An agent could consult authorised tools to gather information; for a small workload a shared record without an agent may suffice.

OpenAI documents how an application can connect models to functions and that application code executes those functions. In a proposed design, initial access would be read-only and limited to relevant jobs. Reserving stock, changing status, sending messages or changing a diary would need separate permissions and human approval. MCP could be considered if several tools need a shared interface; it is not essential to this workflow and does not imply an existing connection.

AI can make mistakes even when answers sound convincing. It must not select parts, treat a photo as compatibility evidence, promise a date or declare a repair complete. Before using real personal data or photos, agree authorisation, access and processing terms with the relevant providers.

How to evaluate a first pilot

Scope a pilot with fictional jobs or authorised documents and a responsible person. Include unconfirmed orders, partial deliveries, similar references and unidentified notes. Prepare a manual reference before testing; compare the draft against it and count omissions, wrong assignments and appointments marked confirmed without evidence. Also record review and correction time. These are proposed measures, not results.

The team can use those findings to decide whether to improve the current form, connect sources or commission software. Agree how to correct errors, who can see each record and how to export the history. Adaris can assess a scope for the process; no existing solution or integration is assumed. Commercial enquiries are handled in Spanish.

Related services

Sources

Are return visits waiting for parts?

Adaris’s AI-integration specialists for plumbing dispatch can summarise the requested part and unresolved reference while purchasing staff check supply before booking a return visit. Describe how your team tracks jobs without sharing private data and we can assess a first custom scope. Commercial enquiries in Spanish.

Discuss my process

Back to all insights