Documents that enter the system on their own

Why documents sit in the inbox

Monday, 8:30. A dozen or so attachments are waiting in the office inbox: client orders as PDFs, date confirmations from suppliers, delivery notes from subcontractors. Each one looks different, because every business partner has its own template. Someone has to open each file, find the part numbers, quantities, prices and dates in it, and then retype them into the system field by field.

This retyping has three costs that never show up on an invoice. The first is time: several to a dozen or so minutes per document, day after day, all year round. The second is errors: a typo in a part number, transposed digits in a quantity, a wrong date. An error like that only comes to light in production or at dispatch, and by then it costs many times more than a minute of careful retyping. The third cost is the least visible: as long as the document sits in the inbox, the system does not “see” it. The delivery confirmation arrived two hours ago, yet production planning is still working on old data.

There is a fourth problem as well. The knowledge in these documents stays locked in attachments scattered across mailboxes. When, three months later, someone needs to find a confirmation with a specific date or an agreed price, the search begins through folders and people's memories.

A solution in three parts

A mechanism like this can be built into a manufacturer's system, and it has three parts.

Part one: a single entry point for documents. Everything that arrives by email or from the scanner lands in one place in the system. AI recognises what the document is: a client order, a delivery confirmation or a delivery note, and which business partner it comes from. The variety of templates stops being a problem, because AI reads the content of the document, whatever the page layout.

Part two: automatic reading and saving straight into the database, exactly where the data belongs. From the recognised document, AI extracts line items, quantities, dates, order numbers and prices, and enters them into the right records: a client order creates order lines, a delivery confirmation updates the date on the right purchase line. A person sees the source document and the extracted data side by side, approves with one click or corrects a single field. The machine reads and enters, the human approves. When the data in a confirmation differs from the order, for example in quantity or date, the system switches on a “red light” and routes the discrepancy for clarification wherever that is needed.

Part three: searching the content of documents across the whole database. Every document is attached to the record it concerns, and its content becomes searchable together with the rest of the data. A question written in plain language, for example “show delivery confirmations for raw material due next week”, returns a list of documents in a few seconds.

What this means for the business owner

The calculation is best done on your own numbers. Count how many documents from suppliers and clients come into the company each week and how many minutes it takes to retype one into the system. Multiply one by the other, and the result by the cost of an hour of office work. That is the first and simplest part of the calculation.

The second part matters more. Think back to the last retyping error: the wrong quantity of raw material in an order, a mixed-up date, the wrong part number. Work out what it cost to sort out: phone calls, corrected documents, a batch put on hold, an unhappy client. Sometimes a few such situations a year are enough for their cost to exceed the whole first part of the calculation.

The third part is the speed at which information flows. A delivery confirmation that reaches the system a minute after the email arrives, instead of the next day, means production planning on current data. This part of the calculation rarely makes it into the cost spreadsheet, yet on the shop floor it shows up fastest.

Where to start

Automation makes sense on an orderly document flow. The first step is one place where documents arrive: a shared mailbox or folder instead of attachments scattered across employees' personal inboxes. The second step is choosing the starting stream. It is best to begin with the documents that are most numerous and have repetitive content, for example delivery confirmations from a few main suppliers. The third step is the reading mechanism with human approval, launched on that one stream. Once it works and has earned trust, it is extended to further document types and further business partners.

This is exactly the kind of mechanism I build in systems based on Claris FileMaker, connected to AI for reading documents. If you would like to see how this could work with your company's documents, write to me or book a short call in the calendar below.

More ideas about what can be built in your company are in the Knowledge section.