GLOSSARY
Token budget
A token budget defines the processing allowance for an AI task. It can cover input, output and, depending on the model, internal processing. Limits and billing rules vary between providers and models. A tight allowance can therefore affect cost, depth and completeness at the same time.
Preparing a negotiation under time pressure
A customer briefing needs to combine contract conditions, unresolved requests and several options. If an application simply removes the final pages of a document, important exceptions may disappear. If it limits the answer instead, the assessment of a proposed concession might stop halfway through.
Business priorities should guide context selection: binding conditions, unresolved points and the immediate decision come first. Supporting material can be retrieved when needed. A difficult review may justify a larger allowance; a routine summary may require less.
Measuring cost alongside usability
Measure cost per usable briefing, omitted conditions and necessary rework. A cheap individual request loses its advantage if several attempts are needed. Truncated outputs must remain identifiable and must not be presented as complete approvals. The practical boundary depends on the context window, actual token usage and minimum business requirements. Representative cases help establish an allowance that fits the task.