Tools

Seven questions about what happens in the process today. The answers show where automation will deliver results and where order needs to be restored first.

How many turns a bot with a connected knowledge base can hold before the start of the conversation falls outside the window — from six numbers about the instruction, the retrieved passages and message length.

Calculating the turn at which a chatbot hits the model's window, the share of the window used by the end of an operator's shift, and the monthly cost of compacting history.

How many messages, conversations and document pages fit into the window of a chosen model, and at which message the history starts getting evicted.

Six questions about the task and the data. The answers show what closes it faster: an assistant, document processing, an autonomous agent, or a private model inside your own perimeter.

An estimate of the time from a customer's question to a finished answer: by model, prompt length, output volume and number of calls — checked against a person's patience in chat and against the telephony window.

A comparison of paying per token versus renting GPUs for an open-weights model — with the volume at which owned hardware becomes cheaper.

An estimate of monthly spend on AI automation for a single process and the payback period — from five figures already known about the company.

An estimate of the one-time investment in a first AI pilot and how long it runs — based on five process parameters that are known before the first conversation with a vendor.

An estimate of the annual cost of manual work at a single operation and the share of hours automation realistically removes — based on request volume, minutes per item and headcount.

How much a move from the current model to a cheaper one costs, and how many weeks the token price difference needs to repay the investment in re-testing prompts.

How many hours and how much money a year go into manually checking model answers, and at what error rate grounding against sources pays for itself.

A calculation of how many document pages and dialogue turns fit into a model's context in a single request, and what a month of operation costs with full context versus fragment retrieval.

A monthly token bill calculated across three architectures — full text in the prompt, retrieval over a knowledge base, and fine-tuning — based on knowledge base size and request volume.

Calculating the monthly charge for model calls based on dialogue volume, conversation length, and model class — with the price of a single dialogue and the share of context in the bill.