Convert any data and fix every addressCopy link
Source files rarely arrive in the importer's shape: the warehouse exports one
column called SHIP-TO, half the addresses are missing a city, and three of
them are typed the way the customer says them on the phone.
Turn any format into Routing24 ordersCopy link
Give your assistant the source data and tell it what each field means. It can either write a ready-to-import workbook, or feed the orders straight into the optimizer through the WebMCP tools.
Your WMS exports one CSV per day that looks nothing like our template:
DOC_NO;CONSIGNEE;ADDR1;ADDR2;PCS;WGT_KG;WINDOW;NOTE
88431;Cafe Mirabel;12 rue des Lilas;75011 Paris;4;38;08-12;ring bell
88432;Bistro Nord;;Paris 18e, 4 av. de Saint-Ouen;2;11;;back door
Ask for the conversion:
Here is today's WMS export. Convert it into a Routing24 orders sheet:
DOC_NO is the order id, CONSIGNEE plus ADDR1 and ADDR2 make the address,
WGT_KG is the load in kg, WINDOW is a delivery time window in hours
(08-12 means 8:00 to 12:00), and NOTE goes to the driver note.
Every order needs 10 minutes of service time. Rows with an empty
WINDOW have no time window at all. Give me the file, and list any row
you had to guess about.
Two things to state explicitly in the prompt:
- The column names come from us, not from the model's memory. Point the assistant at the spreadsheet column reference or at More > Download template, so it writes headers the importer actually accepts.
- Ask for the guesses. "List any row you had to guess about" turns a
silent transformation into a short review list.
88432above has noADDR1and no window.
The same works from a PDF manifest, a pasted email, a screenshot of an order list, or a Google Sheet: the assistant reads it, you review the mapping, and the result goes in through Import from Excel or CSV.
Ask the assistant to save the converted workbook, not only to push the orders into the app. Tomorrow's export will have the same shape, and "apply the same mapping as yesterday" is a one-line prompt.
Skip the file entirelyCopy link
The assistant can also hand the orders straight to the optimizer, with no file and no import dialog:
Take the 34 deliveries from this email thread, use our depot at
14 Chaussee de Mons, Brussels, plan them on 3 vans with a 20 pallet
capacity each, 8:00 to 17:00, and show me the routes.
That creates a new plan, geocodes everything, optimizes, and draws the routes on the map in your tab. See Describe constraints in plain language for the constraints you can express this way.
Fix the addresses the geocoder could not placeCopy link
Routing24 returns, for every address, either a match with coordinates or a clear "not found", and it tells the assistant which one you got. On a failure or a doubtful match, the assistant can go look the address up.
Geocode all the stops first. For anything that fails, or that resolves
to a different city or a house number that does not exist, search the
web and Google Maps for the business name plus the postcode, tell me
the address you think is right and why, and only then use it.
What that catches in practice:
| What is wrong in your file | What the assistant does |
|---|---|
Bistro Nord, Paris 18e with no street number | Finds the business, proposes 4 avenue de Saint-Ouen, 75018 Paris |
A misspelt street (Chausee de Mons) | Matches the real street and shows you the corrected spelling |
| A postcode that belongs to another town | Flags the conflict instead of silently trusting one of the two |
| An address that resolves to a field 8 km out | Reports the suspicious match so you can check it on the map |
| A site with no usable address at all | Takes coordinates you paste, or the ones it found, and uses those |
The skill instructs the assistant to follow two rules:
- You confirm, the assistant does not decide. Ask it to show you the matched address for anything it repaired before it optimizes.
- Corrections travel as coordinates. Once a stop is resolved, the assistant passes latitude and longitude into the plan, so the fix cannot be undone by a second guess later.
Everything above has a manual equivalent in the app: unrecognized addresses show up in the Address Book with a warning, and you can drag the pin or retype the address. See Fix unrecognized or mislocated addresses.
Cross-check a whole file before you planCopy link
Address repair can run as its own pass, before you optimize:
Geocode every address in this file but do not optimize. Give me a table
with three groups: matched cleanly, matched but suspicious, and not
found. For the suspicious ones say what looks wrong.
You get a review list instead of a plan, and a file you can fix at the source.
Next stepCopy link
- Describe constraints in plain language turns your operating rules into solver constraints.
- Query, edit, and export a plan with AI covers everything after the optimizer finishes.