Twelve Questions to Ask Before Hiring an AI Consultant
Is an AI consultant worth it? Direct answers on cost, audits, internal builds, code ownership, timelines, and running costs, with real pricing.
These are the questions owners, operators, and finance leads ask before they hire anyone to build an AI system. We collected them from the language buyers use in search and from the conversations that open our engagements. The answers use our published numbers. Where the right answer is not us, we say so. It is worth it when one recurring process costs real hours or errors every week and nobody inside the business has the time to redesign it. It is not worth it when the goal is to find somewhere to use AI. Size matters less than repetition. A 40-person company with two people matching invoices every Monday has a better case than a large company with no named process. Our pricing has three parts. The Operational Design Session is free. The AI Workflow Audit is $10,000 for two weeks, credited in full if you proceed. Fixed-fee builds typically run $150,000 to $400,000 over 8 to 16 weeks, depending on scope and integration depth. We do not bill hourly, so the price does not grow when the work takes longer than planned. A readiness assessment scores the organization. Our audit decides one build. It maps the workflow as it actually runs, prices it against the status quo, sets what the system does and what people still decide, and defines the tests the release must pass. The plan is portable. You can hand it to your own team or another vendor. Often, yes, for individual tasks. A person drafting faster with an assistant is a real gain. The gap appears when the work spans systems: reading from one, deciding, writing to another, routing exceptions, and keeping a record. That needs integration, permissions, and testing. A chat window does not provide those. Hire when you have a backlog of specified workflows and someone who can own them for years. A single hire usually arrives without the process knowledge and spends the first months learning it. The hybrid often works better: we ship the first release inside your environment while your future owner works alongside us and inherits the runbook. You do. We build in your repository and your cloud account. Your data stays in your systems. You can take the system fully in-house at any point, and the build includes the documentation and evaluation harness your team needs to run it without us. Then the audit did its job. A valid result is build now, assist only, keep human, or defer. Each one is cheaper than a $150,000 build aimed at a workflow that cannot support it. If every audit ends in a proposal, it is sales qualification with a document attached. Rarely in the way people fear. The question is narrower than data readiness across the company. Can the system reach the records one workflow needs, and are they consistent enough to decide on? The audit answers that for one process. Messy inputs add extraction and reconciliation work, which moves the build price. They do not automatically block it. The audit takes two weeks. Builds run 8 to 16 weeks. The first release is deliberately small: one workflow with human approval at the points that matter. Scope widens after the release proves itself on real work, not before. Typically $3,000 to $8,000 a month for model usage, hosting, and monitoring, depending on volume. The model itself is a small share of the total. On our current builds it accounts for about 20 percent of build cost. Integration, architecture, and evaluation account for the rest. It will. The design question is what happens next. Every release names the cases the system handles, the cases it routes to a person, and the actions that require approval. Uncertain cases stop and wait. Every decision is logged, so a wrong one can be traced, reversed, and used to improve the rules. We agree on the proof before we build. The audit defines acceptance tests and a baseline: hours, turnaround, error rate, or volume handled. The release is measured against those numbers on real work. If a number cannot be measured today, the plan names the missing input instead of estimating it. Most of these questions get easier once a specific workflow is on the table. The Operational Design Session takes one recurring process and returns a workflow map, a first-release scope, human checkpoints, and proof criteria. It is free and takes a few minutes.