Hiring an AI agent means paying a provider to set up software that takes a defined job off your team, such as answering enquiries, qualifying leads or booking appointments, and then keeping it running on your own data and systems. You can buy a ready-made platform, pay a developer to build one from scratch, or use a provider that configures an existing platform around your workflow. Each path costs money in different places.
This guide breaks down where the cost of an AI agent actually sits, how providers price their work, how to compare platforms fairly, and the questions worth asking before you sign anything. It is written for owners and operations managers who are close to a decision, not for developers.
What does it mean to buy or hire an AI agent?
The phrase covers three quite different purchases. Knowing which one you are making changes who you talk to, what you pay and how much of the work stays with you.
Buying a platform. You subscribe to software that already does most of the job, connect your channels, load your documents and configure the rules yourself. This is usually the fastest and cheapest start, but you carry the setup work.
Hiring a developer to build one. An agency or freelancer builds an agent for your exact process, often on top of a large language model and an automation tool. You get full control and full responsibility, including who maintains it after launch.
Hiring a provider to configure and run one. A provider takes an existing platform, sets it up around your workflow, connects it to your systems and stays accountable for keeping it working. This sits between the other two on both cost and control.
Many searches for “hire an AI agent” are actually looking for the third option: someone who will make it work, rather than software to learn or code to own.
How much does an AI agent cost?
There is no single price, because an AI agent is a bundle of costs rather than one product. Comparing quotes only makes sense once you can see each part separately.
Setup. Mapping your process, writing the knowledge the agent answers from, connecting channels and systems, and testing. This is often quoted as a one-off fee and is where scope creep tends to hide.
Subscription or licence. The ongoing fee for the platform, typically monthly or annual. Longer commitments usually lower the monthly rate.
Usage. Many agents pay a language model provider per amount of text processed. Some providers include this in the subscription up to a limit; others pass it through. Ask which applies and what happens when volume grows.
Integration and maintenance. Connections to your booking system, CRM or website break when those systems change. Someone has to fix them, and that time is a real cost even when it does not appear on an invoice.
Your own team’s time. Writing and updating the documents the agent relies on, reviewing the questions it could not answer, and approving actions it is not allowed to take alone. This is the cost most often left out of comparisons.
For a sense of how a provider packages these, InCard lists its plans in US dollars on the pricing page, with lower monthly rates on six- and twelve-month terms.
Which pricing models do AI agent providers use?
Providers price in four main ways. None is better in every case; what matters is how each one behaves as your volume and team change.
Per seat. You pay for each staff member who logs in. Predictable, but it can discourage you from giving access to the people who should review the agent’s work.
Per conversation or message. Cost tracks volume. Fair when volume is steady, but a busy season or a spam wave can produce a surprising bill unless there is a cap.
Flat plan with limits. A fixed monthly fee covering a set number of conversations, channels or agents. Easy to budget; check what happens when you exceed the limit.
Project fee plus retainer. Common for custom builds. The build is quoted once, then a monthly retainer covers hosting and maintenance. Make sure the retainer scope is written down, including response times when something breaks.
Is an AI agent cheaper than hiring a person?
It is the wrong comparison for most small and mid-sized businesses, because an agent and a person do different parts of the job.
An agent is strongest at volume and timing: answering the same questions many times a day, replying instantly at eleven at night, and recording every conversation in the same format. A person is strongest at judgement: handling a complaint, negotiating, reading what a customer is not saying, and taking responsibility for a decision.
The useful question is which hours of your team’s week are spent on work an agent could take, and what those people would do with the time back. In practice, the businesses that get value from agents rarely reduce headcount. They stop losing enquiries that arrive out of hours and give their staff more time for the conversations that need a human.
How do you compare AI agent platforms fairly?
Feature lists make every platform look similar. A fair comparison scores each option against your own process, using the same test for each.
Channels. Does it connect to the channels your customers actually use, through each platform’s official connection rather than a workaround? Ask which channels are fully supported and which are still being assessed.
Systems. Can it read from and write to the tools you already run, such as your booking system, CRM or calendar?
Handover. When the agent passes a conversation to a person, does the person see the full history? Does the agent stop replying once a human has joined?
Boundaries. Can you define, in writing, what the agent may do alone, what needs approval, and what it must never touch?
Visibility. Can you see every action the agent took and why, after the fact?
Exit. If you leave, can you export your conversation history and knowledge base?
What should an AI agent demo show you?
A demo built on the provider’s own sample questions tells you very little. Ask for a demo that uses your material, and watch for four things.
First, send it questions your customers asked last week, including the awkward ones. Second, ask something it should not know and see whether it admits that or invents an answer. Third, trigger a handover and look at what the person receiving it would see. Fourth, ask to see the record of what happened, the log or audit trail, not just the chat window.
If a provider is reluctant to run a demo on your own questions, treat that as information.
What questions should you ask an AI agent provider?
These questions separate providers who have run agents in live businesses from those who have mostly built demos.
Where is customer data stored, and who can access it? For Australian businesses this links directly to your obligations under the Privacy Act, covered in more detail below.
What happens when the agent is wrong? Who notices, how quickly, and how is the answer corrected for next time?
How is the agent kept up to date? Prices, policies and stock change. Who updates the knowledge, and how long does an update take to apply?
What is the realistic timeline? A narrow, well-defined first workflow can go live in days to a few weeks. InCard quotes a reference rollout of five to fifteen days for a defined scope. Be wary of any quote that promises a broad, multi-department deployment in the same time.
Can I switch it off instantly? There should be a clear way to stop the agent replying at any time.
What are the warning signs when choosing a provider?
Some patterns show up repeatedly in projects that disappoint. Spotting them early is cheaper than unwinding a contract.
Promises of a specific percentage of enquiries handled without people, before the provider has seen your enquiries. Claims that the agent will “learn on its own” with no explanation of who reviews what it learns. No written list of what the agent is not allowed to do. Pricing that only becomes clear after onboarding. And no case study from a business that looks like yours.
On that last point, it is reasonable to ask for proof in your market. InCard’s work with an Australian real estate consultancy is an example of an agent trained on client-supplied material about the local property market and running around the clock.
What do Australian businesses need to check about privacy?
An AI agent handles names, phone numbers and the details of what customers want, which makes it a privacy question as much as a technology one.
The Privacy Act 1988 and the Australian Privacy Principles set out how personal information is collected, used and disclosed. Australian Privacy Principle 8 deals specifically with sending personal information overseas, which matters if a provider stores or processes data outside Australia. If you send marketing messages, the Spam Act 2003 also requires consent and a working way to opt out.
You do not need legal training to ask the right questions: where the data is held, which sub-processors handle it, how long it is kept, and how a customer’s request to access or delete their information would be handled. A provider used to working with Australian clients should answer these without hesitation. For legal advice specific to your business, speak to a qualified adviser.
How do you start without overcommitting?
The lowest-risk route is to start with one workflow that clearly costs you money today, measure it, and only then expand.
Pick the single job where enquiries are being lost, often out-of-hours messages or slow first replies. Write down how it works now, including the exceptions. Agree with the provider what the agent may and may not do. Decide before launch how you will measure success, such as first-response time or the number of enquiries turned into bookings. Run it for a few weeks before adding a second workflow.
If you want a provider to configure and run an agent around your own process rather than build from scratch, the Custom AI Agent page explains how InCard scopes, connects and hands over each agent.
And if you would rather see it working on your own questions first, you can book a demo and bring a week of real enquiries to test against.
