Separate what you buy from what you sell.

Your platform bill covers part of the infrastructure used to deliver an AI service. Your client is also paying for your understanding of their business, configuration, testing, maintenance and ongoing responsibility. Treat those as different parts of the commercial model.

Start with a representative workload.

Estimate conversation volume, duration, channels and the business actions involved. Use actual client data where appropriate and authorized. Write down the configuration behind the estimate: provider choices, telephony, included usage and external services. A price per minute is only meaningful with its boundaries attached.

Count the work around the agent.

Client onboarding, source review, testing, exception handling and reporting take time. Record what your team does during the first month and during a normal operating month. This gives you a better basis for setup and ongoing service fees than copying another agency’s package.

Understand the billing unit.

Check what is metered, when charging starts and stops, and how rounding, failed actions and overages are treated. Clarify which services are billed separately. Do not assume two providers mean the same thing by a unit with the same name.

Make the client agreement understandable.

Explain what is included, what changes with usage, what support the agency provides and which outcomes the service is designed to achieve. Avoid guarantees that depend on systems or behavior outside your control. A clear agreement reduces confusion when volume or requirements change.

Review the model after real usage.

Compare estimated and actual consumption, exceptions and support work. Use the difference to improve both the service and its pricing. Healthy delivery comes from understanding the whole operation, not just finding a smaller number on an infrastructure bill.