What drives the cost
- Number of systems the agent connects to
- Each integration such as a CRM, ERP or ticketing system adds authentication, data mapping and testing. This is usually the single biggest cost driver.
- Read-only versus taking actions
- An agent that only answers questions is cheaper than one that updates records, sends emails or issues refunds, because actions need permissions, guardrails and approval steps.
- Quality bar and evaluation
- Customer-facing agents need a test set, automated evaluations and red-teaming before launch. Internal agents can launch with lighter testing.
- Data preparation
- If the agent needs your documents or data, cleaning, chunking and indexing them for retrieval adds time, especially for scanned or messy files.
- Compliance requirements
- HIPAA, SOC 2 or financial regulations add logging, access controls, data retention rules and sometimes private model hosting.
- Channels
- Each channel such as web chat, Slack, email, SMS or voice adds its own interface work. Voice is the most involved.
The short answer
Most custom AI agents cost $15,000 to $60,000 to build. The low end is a focused agent that does one job for internal users. The high end is a customer-facing agent that takes real actions inside several business systems with compliance controls.
After launch, expect $50 to $500 a month in model and hosting costs for a typical small business agent. That can grow into the thousands for high-volume agents.
Where the money goes
A typical build spends its time roughly like this:
| Phase | Share of effort | What happens |
|---|
| Discovery and design | 10–15% | Map the workflow, decide what the agent may and may not do, define success |
| Integrations | 30–40% | Connect to CRMs, ticketing, databases and email with proper permissions |
| Agent logic and prompts | 15–20% | Tools, instructions, memory and the decision flow |
| Evaluation and guardrails | 15–20% | Test sets, automated scoring, red-teaming and fallback behavior |
| Deployment and handoff | 10–15% | Hosting, monitoring, alerts, runbook and training |
Most first-time buyers expect the model and prompts to be the expensive part. In practice integrations and evaluation take the most time, and they are what separate a demo from an agent you can trust.
How to keep the cost down
- Start with one job. An agent that handles one well-defined workflow ships faster and teaches you what to build next.
- Keep a human in the loop at first. Approval steps cost less to build than the guardrails needed for full autonomy, and you can remove them as trust builds.
- Use the platforms you already have. If your systems already connect to n8n or have clean APIs, integration work drops substantially.
- Get your data in order. Clean, current documentation and well-structured CRM data cut preparation time.
How Lumeor prices agent work
We scope every agent as a fixed-price project after a short discovery call. You get a written scope with the integrations, actions, evaluation plan and timeline before any work starts. If an off-the-shelf tool fits your need better, we will say so in that scope.
Frequently asked questions
Why do AI agent quotes vary so widely?
Because "AI agent" covers everything from a prompt connected to one API to a system that takes actions across a company's core software. Ask any vendor to list the integrations, the actions the agent can take, and how they will test quality. Those three answers explain most of the price difference.
Is it cheaper to use an off-the-shelf AI tool?
Often, yes, if one fits your process. Tools such as Intercom Fin or Zendesk AI are faster and cheaper for standard support use cases. Custom agents make sense when your workflow, data or systems are specific enough that packaged tools cannot handle them, or when per-seat or per-resolution pricing gets expensive at your volume.
What does it cost to run an agent after launch?
Model usage and hosting for a typical small business agent runs about $50 to $500 a month. High-volume customer-facing agents can run thousands per month in model costs. Ongoing maintenance and improvement, if you want it, typically runs $750 to $5,000 a month.
Can we start smaller?
Yes. Many clients start with a two-to-three-week AI readiness assessment or a focused agent for one workflow, prove the value, then expand. A focused first agent is usually the fastest way to learn what a larger build should look like.
Who owns the agent when it is done?
You do. Lumeor delivers the code, prompts, evaluation sets and infrastructure in your accounts, with documentation, so you are not locked in to us.