Custom AI agent vs off-the-shelf chatbot: which do you need?
By Shaziya Shaikh, AI Engineer at Webify.AI|Updated
Short answer
Use an off-the-shelf AI agent if your help content answers most questions and your volume is modest. You usually pay a fee for each conversation it resolves, on top of seats. Build a custom agent when it must act in your own systems under your rules, keep data where you choose, or when per-resolution fees at your volume outgrow the cost of running your own.
Key takeaways
- Off-the-shelf agents launch quickly and charge per outcome, resolution or session, usually on top of seats.
- Custom agents cost more up front but can act in any system with an API, within limits you set.
- At low volume, per-outcome fees rarely justify a build. At high volume, they can pay for one within a year.
- Where data goes, which model runs and how easily you can switch often decide it, not price alone.
- You can start on a platform and add custom pieces later.
What is the difference between an off-the-shelf AI agent and a custom one?
An off-the-shelf agent is a feature of a helpdesk or platform: you configure it, point it at your content and pay per use. A custom agent is software built for you: it can use any model, act in any system with an API and run where you choose, but you pay to build and run it.
| Question | Off-the-shelf AI agent | Custom AI agent |
|---|---|---|
| Time to launch | Days to weeks, configured in the vendor's tool | A first agent for one task with one or two integrations typically takes six to twelve weeks |
| How you pay | Seats plus fees per outcome, resolution or session | Build hours, then model tokens, hosting and maintenance |
| What it can do | Answer from your content, plus the actions the vendor supports | Anything your systems' APIs allow, within limits you set |
| Where data goes | The vendor's cloud and the model providers it uses | Your choice, including your own servers |
| Which model runs | The vendor's choice | Yours: any major hosted model, IBM watsonx or an open-weight model |
| Switching later | Your set-up lives in the vendor's tool | The code, prompts and tests are yours |
What does each option cost?
Platform agents charge by use: a fee per resolved conversation, per outcome or per session, usually on top of helpdesk seats. Check the unit and the price on your own platform's pricing page. By our planning estimates, a custom knowledge-base assistant takes 240 to 480 hours to build, and an agent that takes actions 800 to 1,600 hours.
Swipe the table sideways to see every column.
| Option | Up front | Each month | Source |
|---|---|---|---|
| Platform AI agent, billed per resolution or outcome | Set-up time | A fee for each resolved conversation, plus helpdesk seats | Your platform's pricing page |
| Platform AI agent, billed per session | Set-up time | Session packs, often after a free allowance | Your platform's pricing page |
| Custom knowledge-base assistant (RAG), at our rates | $8,400 to $16,800 | Model tokens, hosting, and $280 to $560 of maintenance | Our planning estimates |
| Custom agent that takes actions, at our rates | $42,000 to $84,000 | Model tokens, hosting, and $1,050 to $2,100 of maintenance | Our planning estimates |
The units differ, so compare total cost at your volume. Platforms define a resolution, an outcome or a session in different ways, so read the definition before you compare. Our figures are planning estimates, not quotes.
Volume decides it. Take an illustrative platform price of $1 per resolved conversation. At 200 resolutions a month, the fees are $200, less than a custom assistant's maintenance alone. In our worked example, a custom assistant handling 5,000 conversations a month costs $13 to $260 in model tokens, plus $280 to $560 of maintenance. If a platform resolved 2,500 of those conversations at $1 each, its fees would be $2,500 a month. On those fees alone, an $8,400 to $16,800 build pays back in roughly 4 to 10 months, before hosting.
Fees alone are not the whole comparison. A platform's price includes its helpdesk software, hosting, updates and support, which you would otherwise build or pay for. Get a full quote for both before you decide.
When is an off-the-shelf AI agent the right choice?
Choose a platform agent when your help centre already answers most questions, you use that vendor's helpdesk, your volume is modest and you want to launch within weeks without engineers. It is usually the cheaper and faster way to start.
- Most questions are answered by your help centre, policies or product pages.
- You already use the vendor's helpdesk, so there is nothing to integrate.
- Volume is modest, so per-outcome fees stay small.
- Handing over to a person and raising a ticket are the only actions you need.
- You want to launch within weeks and have no engineering team for this.
When should you build a custom AI agent?
Build when the agent has to act in your own systems, follow your own rules, keep data in a place you choose, or run on a model you pick. Volume is the other reason: per-outcome fees grow with every conversation, while a custom agent's running costs grow much more slowly.
- It must take actions your platform doesn't support, such as issuing a refund within a limit, booking an appointment or updating a record.
- Personal or confidential data must stay in a region you choose, or on your own servers. Our private AI page covers that case.
- You need a particular model, or an open-weight model you run yourself.
- Your volume is high enough that per-outcome fees outgrow a build's running costs.
- The agent is part of your product, not only your support desk.
- You want the tests, logs and approval steps under your own control.
An agent that takes actions also needs limits. OWASP lists "Excessive Agency" among its top 10 risks for LLM applications (OWASP). Give the agent only the permissions it needs, ask a person to approve risky actions, and log every step.
Can you combine the two?
Yes. Many companies keep their helpdesk's agent for common questions and build custom pieces for the work it can't do, such as an action in their own system. Start with the platform, measure what it can't handle, and build only that.
- 1
Start on the platform
Turn on your helpdesk's AI agent for your most common questions.
- 2
Measure the gaps
Track the questions it hands over and the actions it can't take.
- 3
Price the gaps
Compare the outcome fees and the hours people spend on those cases with the cost of building for them.
- 4
Build only what pays
Build a custom agent or tool for the cases where the numbers work, connected to your helpdesk and systems.
Where does Webify.AI fit?
We build custom AI agents, so we have an interest in this comparison. If a platform agent covers your questions at your volume, it is usually the cheaper and faster choice. When you need a custom agent, our AI / LLM engineers cost $45 an hour.
- We build agents that act in your systems, with limited permissions, a person approving risky actions and a log of every step.
- We work with the major hosted models, IBM watsonx and open-weight models on your own servers. We are an IBM Partner.
- The code, prompts and tests go into your repositories, and the IP is assigned to you.
- We hold no ISO 27001 or SOC 2 certification. Our security page explains what we do instead.
Frequently asked questions
Is it cheaper to build or buy an AI chatbot?
It depends on volume. At a few hundred resolutions a month, a platform is cheaper. At a few thousand, per-outcome fees can pay for a custom build within a year, although the platform also includes software, hosting and support. Our cost guide shows the arithmetic.
How long does it take to build a custom AI agent?
A first agent scoped to one task with one or two system integrations typically takes six to twelve weeks from discovery to production. A simple knowledge-base assistant can take less.
What counts as a resolution in platform pricing?
Each platform defines it differently. Some count a conversation where the customer confirms the answer helped or leaves without asking for more help; others count sessions or credits. Read the definition on your platform's pricing page, because it decides your bill.
Can a custom AI agent run on our own servers?
Yes. An agent can run on an open-weight model on your own servers, so the data stays in your network. See our private AI page.
Can we move from a platform agent to a custom one later?
Yes. Keep your help content, test questions and conversation logs in a form you can export. They are most of what a custom agent needs to start.
Sources
Checked on 30 October 2026. Webify.AI figures are our published prices.
Talk to an AI engineer about your agent
Tell us the role and when you need someone. We'll email two named engineers with their experience and availability within 48 hours, and you choose who to interview.
- No cost to receive profiles
- Two-week trial with the developer you choose. No charge if you stop within the first five working days.
- Invoiced in USD, GBP, EUR, CAD or AUD by local transfer; IP assigned to you.