The Eleven Days Between Finishing the Work and Sending the Bill

The most expensive automation problem in most businesses isn't glamorous. It's the gap between doing the work and invoicing for it, paid for in working capital every month.

Where the money actually sits

Ask a business owner where their cash-flow problem is and they will usually say customers pay late. Sometimes true. But before the customer is late, there is a stretch nobody counts: the days between the work being finished and the invoice being sent.

In a field-services business it is the engineer's job sheet sitting in a van until Friday. In manufacturing it is a delivery note waiting for someone to reconcile it against a purchase order. In professional services it is timesheets that get written up at the end of the month because that is when someone chases them. In each case the work is done, the money is earned, and the clock on getting paid has not started.

It is invisible because no system owns it. Your accounting package measures from invoice date. The delay happens before the invoice exists, so from the software's point of view it never happened.

What the delay costs

Take a business turning over $12M a year, invoicing on 30-day terms, with an average of eleven days between job completion and invoice going out.

Measure Before If the gap closes to one day
Effective days to cash 41 31
Revenue tied up in receivables ~ $1.35M ~ $1.02M
Working capital released - ~ $330,000, once
Annual cost of financing that gap at 10% ~ $33,000 a year recovered

A third of a million dollars is not new revenue. It is your own money, arriving sooner, and it lands as a one-off release plus an ongoing saving on whatever you were paying to bridge the gap. For a business using an overdraft or invoice financing, that saving is a direct line-item.

There is a second cost that is harder to put a number on and often larger. Invoices raised eleven days after the fact, from a job sheet somebody wrote in a van, are the invoices that get disputed. Every dispute is a phone call, a credit note, and another thirty days.

Why it happens

Almost never because anyone is lazy. It happens because the information needed to raise the invoice is scattered across the job management system, a photograph of a signed docket, a parts list in the inventory system, and somebody's memory of what was agreed on site. Assembling it is a genuine piece of work, and it is nobody's actual job.

So it gets batched. Batching feels efficient and is the single most expensive habit in the process, because it converts a variable delay into a guaranteed one.

What automating it looks like

The goal is not an AI that writes invoices. It is to make the invoice a by-product of finishing the work rather than a separate task.

  1. Capture at source. The engineer confirms parts and hours on a phone before leaving site. Not a form to fill in later; the thing that marks the job complete.
  2. Reconcile automatically. Job record against purchase order against delivery note against the contract rate. This is deterministic work and software is better at it than people.
  3. Draft and route by exception. Anything that reconciles cleanly drafts and sends. Anything that does not goes to a person, with the discrepancy already identified. In most businesses the clean majority is the large majority.
  4. Chase without anyone remembering to. Escalating reminders on a schedule, stopping the moment payment lands.

Note what is not in that list: replacing the finance system. This work almost always sits on top of Dynamics 365, Business Central, Salesforce, or whatever the business already runs. Ripping out the ERP to fix an invoicing delay is the most expensive possible answer to a cheap problem.

Where AI helps, and where it's the wrong tool

AI earns its place at exactly one point in this chain: turning messy human input into structured data. A photographed delivery note, a handwritten job sheet, a supplier invoice in a layout nobody has seen before, an email that says "same as last time but two more units." That is genuinely hard for conventional software and genuinely easy for a modern model.

It is the wrong tool for the arithmetic. Whether a line item matches a contracted rate is not a judgement call, and a system that is probabilistic about money will eventually be confidently wrong about an invoice. Extraction is probabilistic; reconciliation is deterministic; the boundary between them is the most important design decision in the build.

A practical consequence: give extraction a confidence threshold and route anything below it to a human. The number of documents a person touches falls a long way, which is the win, and the ones they do touch are the ones worth their attention.

Where to start

Measure the gap before you do anything else. For the last hundred jobs, take the completion date and the invoice date and find the mean. Most people are surprised, and the surprise is the business case.

Then work out where the days go. If most of the delay is one step, automate that step and stop. A narrow automation that removes six days is worth far more than a programme that promises to fix everything and lands next year.

This is the same principle behind the analytics work we did for TRAMERICAA and the routing and stock work at Trade America: the data needed to make the decision already existed, and what was missing was the time to act on it.