Price the workflow, not the word AI
A useful quote starts with a defined piece of work: what triggers it, what information it needs and what counts as completion. A draft-only assistant and an agent allowed to update customer records have different testing and integration needs. There is no single price that describes both. Azen scopes projects individually; this guide explains the questions behind that scope rather than presenting an unverified market average.
Separate the initial build from ongoing costs
Ask for discovery, data preparation, integrations, implementation, evaluation and staff handover to be itemised. Then ask what happens after launch: model usage, hosting, third-party licences, monitoring and support. Check who holds each subscription and which costs change with volume. A low initial fee can leave substantial work outside the scope.
Find the integration work early
A documented API and consistent records can make a connection straightforward. Duplicate customer records, unreliable identifiers or a system without a supported integration introduce additional work. Share representative examples before requesting a fixed scope. Include the exception cases, not only the cleanest record.
Use a worked estimate with labelled assumptions
Suppose a team handles 600 requests a month, each taking eight minutes. If a pilot releases three minutes per request, that is 30 hours of potential capacity a month. These are illustrative inputs, not Azen client results. Subtract the time spent checking outputs and resolving failures. Released capacity is not automatically a cash saving: decide how the team would use it.
Agree what a successful pilot means
Choose a bounded workflow and a representative evaluation set. Agree the acceptable output, actions that require approval, response time and failure handling. Ask whether revisions needed to meet those criteria are included. Compare proposals against the same scope and acceptance criteria so the cheaper option is not simply delivering less.
Questions to take into a scoping call
Bring your monthly task volume, typical handling time, systems involved and examples of difficult requests. Ask for the assumptions, exclusions, recurring costs, ownership arrangements and handover deliverables in writing. A sensible next step may be an audit when those inputs are unclear, rather than committing immediately to a full implementation.
Explore AI consulting, audits and training ↗
See the Prep Point project · See the Superior Accounting project