How to Automate Otter

One kitchen can be five restaurants at once

Otter is the order layer for restaurants that sell mostly through other people's apps. It pulls DoorDash, Uber Eats, Grubhub and the rest into one screen, and adds channels of its own: QR ordering at the table, a kiosk, an online ordering page, catering.

It is the software that made the ghost kitchen practical. One unit, one team, one extraction hood, and five virtual brands on the marketplaces, each with its own name, menu and rating. Before a system like this, that meant five tablets on a shelf, each chirping at a different pitch, and somebody keying every ticket into the till a second time.

So the tablet holds the whole demand side of the business. Every order and what was in it, which brand and which channel it came through, the prep time promised, the items that got cancelled, the ratings left afterwards, and what each marketplace says it is going to pay.

Everything that is not an order lands on a person

Orders are the part that runs itself now. The rest of the week is unchanged.

The produce order goes in on the wholesaler's own site, from a count somebody did on a phone in the walk-in. The meat invoice is a PDF attached to an email that nobody has filed. Friday's payout is a few hundred short and nobody is sure which orders it covers. The rota gets built against a guess at when the spike will come. A one-star rating arrives at midnight about a brand the manager on duty has never cooked.

And underneath sits the question the tablet cannot answer. The sales screen says the burger brand had a good week. Whether it made money depends on what chicken cost this week, what the marketplace took and how many orders came back refunded, and those three numbers live elsewhere.

Otter

Each brand costed properly What a virtual brand actually earned once commission, refunds and this week's ingredient prices come off, rather than what the sales screen showed.

Money that did not arrive Payouts read against the bank and refunds read against the tickets, so a short week gets queried while the evidence still exists.

Low ratings triaged first The one-star orders pulled out with the items and the timings attached, so a reply takes two minutes instead of an investigation.

Otter reaches as far as the channels it connects

Within its own boundary Otter automates a great deal, and a kitchen using it as a ticket printer wastes most of what it pays for.

Order Manager shows every online and manual order in one chronological list on a single tablet, accepts them automatically against the prep time you set, prints at the pace the kitchen can work, and syncs with the point of sale. Menu Management updates prep times, takes an unavailable item off the online menus or pauses the store, and pushes that to every connected channel at once. Analytics reports sales across every channel and location and tracks order issues, cancellations and product mix. Financials consolidates the delivery partner payouts and breaks out the taxes, commissions and fees.

That is a genuinely useful spine, and it has one thing in common throughout. It is all fed by orders that came through a channel Otter connects to.

The wholesaler's ordering site is not one. Neither is the meat supplier's invoice, the bank account the payout lands in, the scheduling app, the accounting package, the commissary landlord's portal, or the regional delivery app nobody has integrated. Every one of them is a separate login, and the login is where Otter's reach ends.

A brand slipping under water can be caught in a week

Once those logins can be opened too, the numbers that decide the business stop being a quarterly discovery.

The two or three ingredients that carry a brand checked weekly on the wholesaler's site and put against the menu price of the items using them. A brand whose margin has quietly gone becomes a Tuesday conversation rather than something an accountant mentions in April.

Refunds and cancellations from the past week pulled into one list beside the tickets they belong to, so a claim gets made while the evidence exists. The payout read against what actually reached the bank, with the difference named. Tomorrow's produce order drafted on the wholesaler's site from the mix that sold rather than last week's count.

The rota checked against the hours orders actually arrive rather than the hours they used to. Low ratings collected across every brand with the order and the timings attached, and a reply drafted in the voice of that brand rather than the unit.

Keep the line where the guest and the money are. Counting, comparing and drafting can run unattended. Sending a supplier order, replying to a guest or changing a live menu price should be prepared and read by a person: all three are hard to take back.

The prices behind the menu do not check themselves

The real margin on a virtual brand is scattered across a wholesaler, a bank and each marketplace back office, and nothing adds it up. In a kitchen that runs to midnight, whoever does add it up is already tired.

WebRun is an agent that works a real Chrome browser, signed in as you. It opens the supplier site, the bank or the marketplace, reads what is there, and puts it next to what Otter recorded, with the gaps named.

It runs on your schedule in your own private environment, and you can watch a run and stop it partway. Anything that reaches a guest or commits money is drafted and waits for you.

Each of the workflows below is already built, and says which sites it opens before you switch it on.