The problem
The work that runs your company lives in spreadsheets, shared inboxes, and an admin screen someone built years ago that nobody dares touch. Every request for a better tool waits behind customer work, so the back office keeps retyping.
What goes wrong
- The tool is scoped from a wish list. Six months later it does forty things badly and the team is back in the spreadsheet.
- AI is bolted on at the end. The tool is a form with a database behind it, and the reading, matching, and drafting still happen in someone's head.
- Nobody owns it after launch. The person who built it moves on, the integrations drift, and the admin panel becomes the next legacy system.
The approach
- Pick the one screen your team opens most and build that first. One queue, one job.
- Put AI in the first version. It reads the document, drafts the record, and flags the mismatch. A person approves.
- Build on your systems, in accounts you own. The tool reads from your transportation management and accounting systems, runs scheduled read-only checks that post mismatches to Slack or Teams, and ships with an admin panel for seats, permissions, and connectors.
- Ship every ticket with tests, a runbook, and a named owner on your side, so the tool stays maintainable after the pod steps back.
When it works
When a team does the same back-office job every day from a spreadsheet or an inbox, the data already exists in your systems, and one person on your side will own the tool after it ships.
Why internal tools first
Customer-facing software gets the budget. The back office gets a spreadsheet. That is where a logistics operation actually runs: the dispatch sheet, the carrier onboarding checklist in a shared inbox, the invoice exceptions tracked in a tab nobody else can read. Each one works until the person who owns it is out for a week.
An internal tool fixes that. It gives the job one screen, one record, and one owner. The part most teams miss is that the tool should do some of the work, not just hold it.
What to build first
Example: the exception board
Picture a freight broker's billing desk that runs freight audit in a spreadsheet: carrier invoices checked against rate confirmations, one row per load. The first version of the tool is a queue. Illustrative example data:
- Load 48213. Carrier invoice shows four hours of detention. Rate confirmation allows two. Flagged, with both documents attached.
- Load 48227. Invoice total matches the rate confirmation. Proof of delivery signed. Ready for payment, one click to approve.
- Load 48240. No proof of delivery on file. Held, with a request to the carrier already drafted.
AI reads the invoice and the rate confirmation, matches them to the load, and writes the flag. A person on the billing desk approves, holds, or sends. The tool logs who did what and when.
Then the next screen
Once the first queue is live, the next one is obvious. Carrier onboarding: insurance certificate in, expiry date read, a reminder set before it lapses. Dispatch: the load board with the fields your team keeps retyping, pulled from the transportation management system instead. Each one ships as its own ticket, in production, before the next starts.
How we build it
A forward-deployed engineering (FDE) pod joins your team, and the first ticket is a working slice of the first screen.
Anything in the tool that moves money or contacts a carrier goes through an approval owned by a named person on your team, and every action is logged. The admin panel, including who can approve what, is not a second project. It ships in the same build.
When it does not work
It does not work as a replacement for a system of record you have not chosen yet. Pick the transportation management system or the accounting system first. The internal tool sits on top of it, not instead of it.
Where to start
If the back office is where the hours go, the FDE Pod builds the first tool and the ones after it. What a forward-deployed engineer does all week shows how, and a week in a pod has the rhythm. If you are not sure which screen comes first, ask Roo, Webaroo's AI: Which back-office spreadsheet should become a tool first?
Book 15 minutes with Connor