For ordering platforms

We build back-office agents for online-ordering platforms: one that finds stuck orders and out-of-stock sales, a venue database your sales team can trust, and lead ranking inside your CRM. Each one starts as a Proof: one workflow on your real data, from 1,500 euro, 5 working days, in shadow mode.

The problems we hear

• Orders accepted but stuck. The restaurant accepted the order, but it never reached the kitchen. The first sign is the customer calling.
• Refunds by hand. The kitchen ran out of a dish, the menu still sold it, and your support team refunds each order one by one.
• Menus out of sync. A price or an item changes in the till but not on the ordering site, or the other way round.
• Prospect lists built by hand. Your sales team copies restaurants from maps into a spreadsheet and still calls venues that have closed.

What we build for each

For stuck orders, refunds and menus, we build an operations agent. It checks each order against what reached the restaurant's POS, and finds stuck orders, items sold while out of stock and menus that differ between channels. It fixes what your written rulebook allows, asks a person for the rest with the evidence, and logs every action with an undo.

For prospect lists, we build a venue database with a source per field. Venues come in from several sources, duplicates are merged, and every field keeps where it came from. A nightly self-test writes nothing when it misses too many known cases.

For the sales team, we build lead ranking inside your CRM. Each matched venue gets a score with a short reason. Thin leads stay blank instead of getting a guess. The scoring is checked on held-out records before anyone relies on it.

What our engineering team has built

• An operations agent for delivery orders, with authority tiers, a ledger and self-measured accuracy.
• A venue database for a sales and onboarding team, with the source kept for every field.
• Lead ranking for a sales team inside its own CRM.
• A visit planner for a field sales team: venues known to be open, on one map route.
• Merchant screening for a payments risk team: what a business sells and the red flags found, for staff to decide.

How the Proof runs

On day 1 we write 3 to 5 tests with you. For example: an accepted order missing from the POS after the agreed window is flagged, and a late order that did arrive is not. The Proof reads your data and reports. It writes nothing. On day 5 it runs on your data and passes your tests or it does not. The Proof is credited in full against a Pilot booked within 30 days. The Pilot, from 6,000 euro, puts the workflow live in your systems for 30 days, measured against a baseline you agree first. More than one workflow is a Build, from 20,000 euro over 3 to 6 weeks.

Half up front, the other half when it passes the tests. If it fails and we can't fix it within 5 working days, the first half comes back. You own the code.

A worked example, on synthetic data

The delivery operations agent is live to click. Every order, store and failure in it is synthetic: we made it up to look like the real thing.

On 7 replayed days, 249 planted incidents, on days it was never tuned on, it scored precision 0.98 and recall 0.99. Precision says how many of the problems it flagged were real. Recall says how many of the real problems it found. The result: 95% handled without a person, 5% handed to a person, each hand-off with the evidence attached. The weakest detector is named on the accuracy page.

Those numbers are against failures we planted, so they do not say how it would do on your orders. Your Proof is graded by your tests, on your data. How we measure explains the rest.

Send us the job in the form below and a plan with a fixed price comes back within 24 hours.