Agents in the Trades
Finance

The unbilled job hiding in your own board

The job wrapped three weeks ago. The invoice was never sent. Sunday night, an owner scrolls his board — jobs marked complete, quotes approved and gone quiet — and asks the assistant connected to his Jobber account one question: where is it leaking?

Source: Jobber, "Jobber Launches MCP, Bringing Home and Commercial Service Data and Workflows Directly Into ChatGPT and Claude," PR Newswire, Sep 25, 2026 — https://www.morningstar.com/news/pr-newswire/20260925ca57084/jobber-launches-mcp-bringing-home-and-commercial-service-data-and-workflows-directly-into-chatgpt-and-claude

Watch the 53-second version — Agents in the Trades, E004.

The problem

Every shop has the column nobody owns. Jobs marked complete. Quotes approved and gone quiet. Work that got done — and never got billed, because the board is the business, and whatever falls off it falls out of the bank account.

Nothing about the leak looks like a leak. A completed job sits in the same board view as a billed one. An approved quote waits politely for a scheduling step nobody flagged. There is no alarm for revenue that simply never got asked for, so it surfaces months later, if at all — usually as a vague sense that the shop was busier than the bank account says it should have been.

How the agent workflow works

  1. Connect the assistant to the system of record. Jobber's MCP connection lets an assistant like ChatGPT or Claude read the shop's real data — quotes, jobs, invoices, clients — instead of a spreadsheet export that was stale the day it was made.
  2. Ask the leak question in plain language. The owner asks one question — where is it leaking? — and the agent audits the board against itself: every quote in an approved state with no job scheduled behind it, and every job marked complete with no invoice raised.
  3. Get a list ranked by dollars, not a data dump. Each finding comes back as an action item — the client, the job, the amount, how long it has been sitting — ordered by the money at stake, so Monday morning starts with the biggest recovery first.
  4. Work the list; the agent does the clerical half. The owner makes the calls and the judgment decisions. The agent drafts the invoice from the completed job's own record and the follow-up message from the quote, so acting on a finding takes minutes instead of a paperwork session.
  5. Close the intake side too. When a new enquiry lands in email, the agent turns it into a client, a property, and a request in Jobber — nothing re-typed, nothing parked in an inbox to become next quarter's leak.

What you need to set it up

Where it breaks

It can only audit what the board knows. Work that was never entered — the cash job, the favor for a neighbor, the change order agreed on a handshake — is invisible to the audit. The list is the board's leaks, not the business's.

Status drift creates false alarms. A job invoiced outside the system, or a quote the customer declined by phone, still reads as a leak until someone updates it. Expect the first run to be part recovery list, part cleanup list — that cleanup is what makes the second run trustworthy.

Recovery is still a phone call. The agent ranks and drafts; it cannot make the awkward call to a customer about a three-week-old invoice feel routine. The owner's voice is the step that turns a finding back into money.

Access cuts both ways. An assistant connected to live business data can answer questions the owner never thought to ask — including ones asked by anyone with access to that account. Connect it to the owner's login, not a shared one, and review what it can touch.

The bottom line

Run this if your shop is busy and the bank account disagrees. The audit costs one plain-language question against data you already have, and the first list it returns is usually money the shop already earned. After that, it is a cadence: ask on schedule, work the list, and let the board stop being the place finished work goes to be forgotten.

Related use cases

← All use cases

One working agent workflow, every day.

Real AI agent use cases from solar, construction, and field operations — what they do, what they cost, and where they break. No hype, no theory.