Pricing

Each of those has a page of its own above, and so does every other job listed there. Reading them and agreeing is not the way to find out, though. Bring one of them and a month of the records behind it.

Every one of those pages ends with the measures to take out of your own systems first — days from a box landing to the call being made, claims filed against claims available, forecast error by line. Write yours down before anything changes, because the baseline is gone for good once it does.

If the losses you carry are not the shape of the ones described there, we will say so.

Records you already hold. Each scenario names what it wants in order to start: one month of returns and your credit notes; one lane or one carrier and a quarter of invoices; a year of weekly history for one product family; one closed claim file and your delegated authority schedule; one contract and a question your team answered from memory.

Extracts in whatever format your systems produce them are enough to begin with. The joining is the work.

A named person. The paperwork is gathered — the entry, the port, the reason it is held, the documents that are missing, the days it has been held and the per-day charge — and the letter is drafted. Then it stops.

Somebody reads the case and approves it, edits it or throws it out, and that sign-off is kept on the record. Nothing goes to the carrier before it. The same shape holds elsewhere: what arrives is a proposed action, and an action nobody approves is an action that has not been sent.

It says so, rather than answering anyway.

A return grade it does not recognise is refused instead of filed under a best guess. A demand series it cannot fit comes back saying it could not be fitted, instead of as a confident-looking line with nothing underneath it. Where part of a drafted action cannot be carried out, the reply names the step that did not happen instead of reporting the work as done.

That matters more than it sounds. The failure this replaces is a plausible answer that nobody downstream could tell apart from a real one.

You can see why it was wrong. Competing methods are tried against periods your own history already contains, and the one that predicted those periods best is the one used. The answer carries the name of the method that won.

So a miss is something to look into — which method, against which history, and what changed — rather than something to take on trust. Your own forecast error by line is the figure to write down before any of it is running.

The record. Each proposed action waits in a queue as a record of its own, and approving it is the step that carries it out.

That approval is written down with the name of the person who gave it and what they decided, kept together. So the answer to why a claim was filed or an entry was held comes out of the record rather than out of whoever still remembers.

Onto machines you control, and no further. The order files, the customs papers, the sensor readings and the reasoning about them all run on hardware you run. Nothing goes to an outside model provider.

That is the answer a security review asks for before it will let a supplier hold its order data. It is also why the three licences above are told apart by whose machine they run on: that question is the first one a buyer in a regulated business has to settle.

Seats. A seat is one person who uses Runink. You count the people who need it, multiply by the price above, and that is the licence.

Each seat also comes with an allowance of computing capacity — the machine time Runink uses to read your documents, check your records and draft the work. That allowance is included in the seat price. You are not billed by the question, the document or the report.

It is the meter for machine time, the way a kilowatt-hour is the meter for electricity. Runink counts capacity in Compute Units so that what you were given and what you have used are stated in the same terms, and both are on screen in the console rather than arriving at the end of the month.

Every Dedicated seat carries 1,000 units, and your organisation gets a further 2,000 units for every 10 seats you hold. Those units are pooled, so a heavy week for one person draws on the same allowance as a quiet week for another.

Extra capacity is charged at $0.10 per 100 units, or $5.00 per hour of machine time. In practice that line stays empty for ordinary day-to-day work and appears when you run something very large in one go — reprocessing a year of documents in an afternoon, for example.

You can see the running total in the console and set a budget against it, so the first you hear of a heavy month is not the invoice.

Count your people first.

Under ten, the Lite Licence is the fit. It runs on a machine shared with other customers, and it is the only one you can take a month at a time — so an evaluation does not need a year’s commitment.

Ten or more, the Dedicated Licence costs less per person and runs on machines kept for your company alone, with your own web address and first call on the capacity you pay for. It is taken a year at a time.

If your information cannot leave your own building, that is Enterprise, and the conversation starts with where it has to run.

Only for Dedicated and Enterprise. The Lite Licence can be taken month by month at $86 per person, or a year at a time at $75 — the same 15% difference shown by the toggle above.

Dedicated and Enterprise are a year at a time because both involve setting machines aside for your company specifically, and that capacity is reserved whether or not you use it in a given week.

Because the reasoning runs on hardware rather than on somebody else’s metered service. The cost of a question is the electricity to answer it.

The practical consequence is a budgeting one. Your spend is a function of the capacity you run, decided once, rather than a number that moves with how many questions your team asked last month. A team that finds heavy use for Runink does not discover a cost that grows with that success.