Most business cases for AI agents start with the model fee. That number is easy to find and easy to compare. It is also the smallest part of what an agent costs to run once it is doing real work.
For mid-sized employers, including banks, lenders, insurers, manufacturers and distributors, the larger gap is often people. Someone has to watch the agent, price it, secure it and answer for it. If nobody is hired or assigned to do that, the cost shows up later as a surprise.
Why an agent costs more than a chatbot
A chatbot answers a question. An agent finishes a task. It pulls records from several systems, tries options, calls other tools, checks its own work and keeps going until the job is done.
Each of those steps uses compute, storage and security controls. Those costs rarely sit on one line. They get spread across infrastructure, security, data and platform budgets, so no single leader sees the total.
A routine request and a messy one can differ in cost by a wide margin. A budget built on an average will be wrong in both directions.
The costs that don't appear in the business case
Beyond model access, expect to pay for:
- Orchestration and the tools that connect the agent to your systems
- Monitoring, logging and review of what the agent did
- Storage for retrieved data and records of its decisions
- Security controls and access management
- Governance and compliance work, especially in regulated businesses
- Ongoing support when workflows change
Every item on that list needs an owner with the right skills. That is a hiring question as much as a technology question.
The roles that keep the cost visible
You do not need a large team. You do need specific skills in place before agents move from a pilot into daily operations.
- Cloud cost analyst or FinOps practitioner. Tracks spend by workflow, flags drift and reports cost per outcome, not just cost per month.
- Data or ML engineer. Builds retrieval, keeps pipelines efficient and stops agents from making more calls than the task needs.
- Platform or cloud engineer. Owns the infrastructure the agents run on and sizes it for peak, not average, demand.
- Risk, compliance or controls specialist. Defines what the agent may do, what a person must review and what gets logged.
- Operations lead for the process. Knows the workflow well enough to say when the agent is wrong, slow or expensive.
In smaller organizations one person may cover two of these. The mistake is covering none of them and assuming the vendor will.
Plan the workflow, not the use case
Before you scale, map one workflow end to end and answer a few plain questions.
- How many systems does the agent touch?
- How many steps does a typical task take, and how many does a hard one take?
- What extra infrastructure supports retrieval, monitoring and security?
- Where does a person step in, and how long does that take?
Then split the budget into separate lines: model use, cloud, storage, governance tooling, monitoring and support staff. One combined AI budget hides where costs are rising and why.
Track cost per completed task and cost per business outcome. Adoption numbers alone tell you very little about whether the economics hold.
Decide who owns the number
Agent costs cross departments. Technology, finance, risk and the business unit each see a slice. Name one accountable owner for the total and give finance a regular view of it. Review assumptions every quarter during the first year, because usage tends to grow faster than the forecast.
Contract or permanent
The skills above are scarce in both Canada and the US, and the need may be uneven. Early design and cost modelling work often suits a contract specialist who has done it before. Monitoring, cost control and process ownership usually belong with permanent staff who learn your systems over years.
A reasonable pattern is to bring in contract expertise to set the foundations, hire one or two permanent people to run them, and train the existing team on the rest.
What to do next
Pick one workflow you plan to automate and list every cost and every role it needs, including the ones nobody has been assigned. Where a name is missing, you have a hiring or training gap to close before the agent goes live.
A Traverse recruiter can help you find the AI, data, cloud and finance people to fill those gaps, on contract or permanent.