Off-the-shelf chatbot or custom AI assistant: which does your business need?
Every website builder, CRM, and help-desk product now ships with an AI chatbot toggle. Flip it on, and for a monthly fee you have "AI support." So when a business asks us to build a custom assistant, the fair first question is: why not just use the built-in one?
Sometimes the answer is "you should." We've told prospective clients exactly that, because a custom build they didn't need would have come back to bite us in a year. But the two options solve different problems, and picking the wrong one wastes money in opposite ways. Here's the difference, honestly laid out.
What does an off-the-shelf chatbot actually do?
An off-the-shelf chatbot answers common questions from a script or a help-center crawl, hands anything unusual to a human, and deploys in an afternoon. For a business with simple, repetitive questions and a decent FAQ page, that's genuinely useful, and it's the cheapest way to find out whether customers will talk to a bot at all.
The limits show up in predictable places. The bot knows what's in your help center and nothing else. It can't check a customer's order, doesn't know your returns policy changed last Tuesday unless someone updates the source page, and it answers in the vendor's house style rather than yours. Configuration options exist, but you're choosing from the vendor's menu. When customers ask something one step beyond the script, the bot either guesses (bad) or punts to your team (fine, but that was the work you were trying to reduce).
None of that is a flaw, exactly. It's what a product built for a hundred thousand customers has to be.
What does a custom assistant do differently?
A custom assistant is built on your documents, your policies, and your systems. It can answer from the contents of your actual price book, look up a live order, draft replies in your voice for a human to approve, and decline to answer the things you've told it are off-limits. It behaves like a trained employee rather than a kiosk.
That comes from work, not magic. Building one means connecting your data sources, setting guardrails so answers stay accurate and on-policy, and testing against the questions your customers really ask. A typical build for us runs four to eight weeks, and we've described what that process looks like in building a custom AI assistant your staff will actually use.
The other difference is ownership. A custom assistant is yours: your data stays where you decide it stays, and you're not stuck if a vendor raises prices or sunsets the feature. For businesses handling sensitive information, that control is often the deciding factor all by itself.
The comparison, side by side
| Off-the-shelf chatbot | Custom AI assistant | |
|---|---|---|
| Time to launch | Hours to days | Weeks |
| Up-front cost | Minimal | A real project budget |
| Knows your business | Only what's on public pages you point it at | Your documents, policies, and connected systems |
| Connects to your tools | Rarely, and only the vendor's chosen partners | Yes, that's the point |
| Voice and behavior | Vendor's defaults, lightly configurable | Yours, with guardrails you set |
| Data control | Vendor's terms | Your terms, self-hosted if needed |
| When it breaks down | Questions outside the script | Mostly when a connected system changes, which is maintainable |
| Lock-in | High; it lives inside their product | Low if built sensibly |
A simple rule for choosing
If your customer questions are simple and your answers are public, start with off-the-shelf. If the value depends on your private data, your systems, or your voice, the packaged bot will disappoint you and a custom assistant will pay for itself.
A more concrete test: write down the last twenty questions your team actually answered. If most could be settled by quoting your FAQ page, the built-in bot covers you. If most needed someone to check an account, interpret a policy, or write a careful reply, that's the work a custom assistant takes over, and no off-the-shelf product is going to do it.
There's also a sequencing answer. Plenty of businesses run the cheap bot first, learn from it which questions pile up, and use that evidence to scope a custom build that targets the expensive ones. The off-the-shelf bot becomes the research phase for the real project. Nothing wrong with that path at all, and the usage data makes the custom project better-scoped than it would have been.
The mistake in the middle
The worst outcome we see is the business that pays custom prices for off-the-shelf results: a "custom" assistant that's really a thin wrapper around a generic model with none of their data connected. It demos well, answers vaguely, and gets quietly abandoned by month three. If a proposal for a custom assistant doesn't talk specifically about which of your documents and systems it will be grounded in, and what happens when it doesn't know an answer, you're being sold the wrapper.
Grounding and guardrails are the substance of a custom build. Everything else is chrome.
Where to go from here
If you're weighing this decision for your own support inbox or internal helpdesk, our custom AI assistants page describes what a grounded build includes. And if you're not sure your question volume justifies either option yet, say so when you book a call. "Not yet" is an answer we give regularly, and it's free.