Methodology
What must be true for this to be right?
BillingEngine is only useful if you can see where every number came from. This page states exactly that.
What comes from you
Volume, handling time, labour cost, target automation, escalation expectations and any architecture details you already know are your inputs. They are labelled as user input everywhere they appear.
You are never required to know a technical figure. Every question accepts 'I don't know'.
What BillingEngine estimates
Where you do not supply a value, BillingEngine fills an explicit default and labels it a BillingEngine estimate. Defaults are visible and editable on the results page, never hidden inside a calculation.
Model prices are illustrative and stored separately from the calculation logic. They are not vendor quotes and they change frequently.
Why ranges rather than a forecast
Deployment economics depend on workload characteristics nobody can know in advance: how much of your traffic is genuinely simple, how often the agent needs a human, how much context each answer requires.
So BillingEngine produces three scenarios, conservative, base and optimistic, by moving automation, escalation, resolution rate and consumption together. Presenting a single number as certain would be the wrong kind of precision.
Why AI cost is more than tokens
Total AI workflow cost is model inference plus supporting infrastructure and data, tool and system calls, human intervention through escalation and review, the cost of failures and retries, and an allocation of implementation cost.
In a realistic customer-service deployment, human intervention is frequently larger than inference. A model price change can be economically irrelevant while a five-point move in escalation is decisive.
What cost per outcome means
Cost per token measures consumption. Cost per successful business outcome measures whether the workflow is worth running.
For customer service, the outcome is a successful resolution. An interaction the AI attempted and failed still consumed cost and produced no outcome, which is why it belongs in the denominator's failures rather than its successes.
How confidence works
Confidence reflects the provenance of the material inputs, how many came from you rather than from our defaults. It is a mechanical measure, not a judgement about your business.
Alongside it, BillingEngine ranks which inputs your result is most sensitive to, so you know what to validate before committing money.
Limitations
BillingEngine cannot predict your exact deployment economics, and does not claim to. It models one workload, a customer-service AI agent, and it models it from assumptions you can see and change.
Treat the output as a structured business case with named assumptions, and validate the two or three that drive it before spending.
Ready to model a workload?
V1 covers customer-service AI agents. You will see results before anything is asked of you.
Analyze an AI use case