Commercial model

Priced per workflow, not per seat.

We do not publish a price list, because a number without a scope would be misleading: an engagement is sized to one specific queue. What we can set out plainly is the structure, what actually drives the cost, and what you own at the end. That should be enough to build an internal business case before you ever speak to us.

An abstract cost against volume visualisation of rising luminous steps — The fixed cost sits at the start. The running cost scales with the work actually done.

Three stages, and you can stop after any of them.

Each stage is separately committed. Nothing rolls over automatically, and the first stage is deliberately small enough that finding out we are the wrong answer is cheap.

Fixed fee, two to three weeks

Decision audit

A scoped, fixed-price piece of consulting on one workflow. Priced the same whether or not you continue afterwards, because it is a real deliverable in its own right.

Sessions with the people who actually do the work
Your decision framework for that workflow, written out in full
The gaps and contradictions we found, named plainly
An honest recommendation, including "do not automate this yet"
A firm quote for design and deployment, if it is worth doing

What you own afterwards. The written decision framework. Yours to keep and use however you like, including with another vendor.

Quoted per workflow, five to eight weeks

Design and deployment

The engineering engagement: tuning the operator on your material, connecting it to your systems, building the guardrails and running it live beside your team until it earns trust.

The operator fine-tuned on your documents, precedent and vocabulary
Connection to the systems in scope, with its own named account
Workflow Guard configured to your actual policy
Evidence trail on every action, from the first supervised run
Shadow running, then supervised live running, with weekly review
A measured comparison against how the queue ran before we arrived

What you own afterwards. The framework, the precedent library and the full evidence trail. The tuned operator runs on infrastructure you control.

Monthly, optional

Ongoing

Only worth taking if you want us to keep improving it. Plenty of clients run a deployed workflow without ongoing support once it is stable.

Quarterly review of the framework and the guardrails
Model migration when a materially better model ships
A named engineer who already knows your workflow
Additional workflows quoted separately, at a lower cost than the first

What you own afterwards. Everything above. Ending the arrangement does not switch anything off or take anything away from you.

What actually moves the number.

Roughly in order of impact. If you want to estimate where your own queue sits before talking to us, these are the six things to look at.

How much of the decision framework already exists

The single largest driver. A workflow with a maintained, accurate procedure document is materially cheaper to codify than one that lives entirely in two people's heads. Most sit somewhere between, and the decision audit is what tells us which.

How many distinct case types the queue contains

A queue with three recognisable shapes is quicker to tune than one with twenty. Breadth costs more than volume, because each shape needs its own thresholds, its own escalation rule and its own test cases.

How many systems the operator has to touch

One system with a clean interface is straightforward. Six systems, two of which are browser portals with no API, is a different amount of connection work. This is usually the second largest driver.

How strict the governance requirements are

A queue with a simple single approver costs less to guard than one with segregation of duties, tiered spending ceilings and a regulator with audit-trail requirements. Stricter is not worse, it is just more to build and verify.

Where it has to run

A cloud deployment has no hardware line. A local, air-gapped deployment on your own server does, plus the setup that goes with it. We size and specify it with you rather than marking up hardware.

Case volume, once it is live

Model inference is the running cost and it scales with how many cases the operator handles. Higher volume raises the running cost but almost always improves the economics per case, because the fixed engagement cost is spread wider.

Four things we will put in the contract.

These are the commercial terms clients most often ask us to confirm, so we would rather state them up front than wait to be asked.

No seat licences

You are not billed per user. Pricing tracks the engagement and the running volume, not how many of your staff happen to look at the output.

No lock-in on your own material

The decision framework, the precedent library and the evidence trail are yours, in an open format, at any point. Leaving does not mean starting again from nothing.

No charge to move models

When a better model ships and we migrate your deployment onto it, that is part of the ongoing arrangement, not a re-implementation fee.

No markup on hardware

If a local deployment needs a server, we specify it and you buy it. We do not resell it to you at a margin.

How to build the internal business case.

The comparison that matters is against how the queue runs today, not against a vendor benchmark. These are the numbers your own finance team will ask for, and all of them are ones you already hold.

Hours currently spent on the queue

Not headcount. The actual hours per week that go into the repetitive, reviewable part of the work, separated from the genuinely judgement-heavy part your people should keep.

Cost of the errors you already catch

Rework, duplicate payments, missed deadlines, the reconciliation that had to be redone. You are already paying this; it is just not on a line item anywhere.

What inconsistency costs you

How often two competent people would decide the same case differently, and what the downstream consequence of the wrong one is. This is the cost AutoServa is most directly aimed at.

The key-person risk on that queue

What happens to throughput and accuracy when the person who really knows it is on leave, and what it would cost you if they left permanently. Most teams have a number in mind and have never written it down.

Get a real number for your queue.

Bring one workflow and we will tell you what the decision audit costs for it, what design and deployment would likely run to, and whether we think it is worth doing at all. The first conversation costs nothing and we will not send a proposal you did not ask for.

Book a working session Compare the alternatives