By 2026 most transport software claims to read orders with AI, from a PDF, a spreadsheet, an email. In the demo the vendor shows a clean file, one row per order, every address spelled right, and the result is flawless every time. Your orders do not arrive like that. This article is about what to bring to the demo and the five tests to run, with any vendor, to find out whether order entry holds up on real documents.
The worst document you receive
Before the demo, open last week’s order folder and pick the three ugliest. They are usually these.
- The photo of a hand-filled form, taken on a phone by the sales rep or the customer, slightly crooked, one line crossed out and a quantity corrected in pen.
- The scanned PDF of a long order: thirty lines, two pages, a note in the margin saying “deliver after 3 pm”.
- A customer’s spreadsheet in their own layout: street in one column, number in another, merged cells in the header, two orders on the same sheet.
Ask for the demo to start from those rather than from the vendor’s file. If the answer is “send them over first and we’ll prepare them”, you have already learned something.
For a reference point, look first at how Magic Order in OptivoRoute reads an order in a chat: attach the file or paste the text, and watch what it asks when a value is missing. Then run the same five tests on anyone.
Five tests to run with any vendor
1. A missing value
Remove the street number from an address, or leave a quantity blank, and watch. There are four possible behaviors: the system invents a value, leaves the field empty and moves on, looks it up in that customer’s history, or stops and asks. The first two cost you a wrong delivery. The third is fine only if the result is shown and confirmed. The fourth is the right one, as long as the question comes before the order enters the plan.
2. Two orders on the same page
A sheet with two customers, or a PDF with ten orders in a row, has to become two or ten deliveries, with the right lines under the right recipient. Many tools read a single-order document well and get confused when several recipients share a page. Bring a real one.
3. A new customer and one you have served for years
For the long-standing customer the system should already know the address, the time window and the delivery notes, and use them to fill what is missing. For the new customer it has to ask for everything. Ask the vendor where the history comes from, since when, and whether it is updated with the deliveries you make.
4. A correction
The system read “15 Oak Street” and the right number is 50. Where do you fix it? By typing in the same chat, in a table to re-check line by line, or by uploading the file again? The time a correction takes is the real time of the feature: if fixing one value means leaving the flow, the promise of no retyping is only half kept.
5. What comes out the other side
The last test is the output. From the order that was read you should get a delivery with recipient, address, quantity, time window and notes, already where routes get calculated. If what comes out is a file to import into another tool, the work has only moved one step. Ask to see the delivery in the plan, not the extraction table.
What not to ask
Do not ask for an accuracy percentage. Nobody can verify it on your documents before seeing them, and a high number on the vendor’s sample says little about your scanned PDF. Ask for the mechanism instead: what the system does with a missing value, with one that does not add up, with a line it cannot read. A vendor who answers “it asks” and shows you is worth more than one who answers with a number.
Two more questions that get forgotten in demos and matter afterwards. Where the documents are processed, meaning in which country and with which model. And what happens to them after reading: how long they are kept, who can see them. If you distribute pharmaceuticals or food, your customers will ask you the same.
Where Optivo stands
In OptivoRoute, order entry is Magic Order, a chat inside the software. You attach the files as they are, as Excel, CSV, PDFs including scans, a photo taken on your phone, or you type or paste the text, several files at once and in more than one language. A document with many orders becomes many deliveries. If a value is missing, the AI looks it up in that customer’s history and, if it finds nothing or the result does not look plausible, asks in the same chat; corrections are typed there too. Deliveries come out with recipient, address, quantity, time window and notes and go straight into route calculation with the constraint library. Documents are processed by an AI model run by Optivo on servers in Europe.
We are happy to run the five tests on your documents: book a demo and bring the three ugliest.
Frequently asked questions
Does the AI read a hand-filled form?
It depends on the tool and on the handwriting. In OptivoRoute the photo of a hand-filled form is one of the formats Magic Order accepts; whatever it cannot read, it asks about. The real test is your own form, in a demo.
Do I need to prepare a template before uploading orders?
Not with chat-based order entry: you attach the file as it is, in the formats it arrives in, or paste the text. If a vendor asks you to fill in a template first, you are retyping again, just somewhere else.
What happens to the documents I upload?
Ask every vendor in the demo: where they are processed and how long they are kept. At Optivo the documents are used to create the deliveries in your plan and are processed by an AI model run by Optivo on servers in Europe; if you have specific requirements on retention and access, we go through them together in the demo.
The next step
The other questions to ask a vendor, on constraints, on the trial and on scale, are in the guide on how to choose route optimization software. If your orders already arrive structured from your ERP, the route is the API or a CSV export; the chat is for everything else, which is usually the part that costs the most time today.