AI Chatbots for Business: Costs, Options and When Not to Bother

Most guides to buying an AI chatbot for business are written by companies that only sell one kind of chatbot, so the answer is always yes. This one is not. We build these systems, and often enough the honest advice is to spend the money somewhere else. Here is how the options actually differ, what you are really paying for, and the cases where a chatbot will not move a single number that matters to you.

What “AI chatbot for business” actually means

The phrase covers three quite different things, and the gap between them is the gap between a few hundred dollars and a proper build. Before you compare quotes, work out which tier a vendor is quoting, because a rule-based widget and a grounded language model are not the same product with different logos.

Tier one: rule-based widgets

These are decision trees dressed up as chat. The visitor picks from buttons or the bot matches keywords, then it follows a script you wrote. There is no real language understanding. Ask it something the script did not anticipate and it either loops or hands off to a human.

They are cheap, they are quick to launch, and they are genuinely useful for narrow jobs: booking a callback, qualifying a lead with three questions, pointing people at the right page. If your need is “catch the enquiry and route it”, a widget is often the right call and you should not pay language-model money for it.

Tier two: platform bots

These sit on a subscription platform (the well-known support and marketing tools) that bolts a language model onto a builder. You get better conversation, canned integrations with your help desk or CRM, and a dashboard your team can manage without a developer.

The trade-off is that you live inside the platform’s limits. Pricing usually scales with conversations or resolved tickets, so a good month for your website is also a more expensive month. For a lot of NZ businesses this tier is the sensible middle, especially if you already pay for the platform.

Tier three: custom LLM chatbots grounded in your knowledge base

This is a language model connected to your own content: product docs, policies, past tickets, pricing rules, whatever the answer actually lives in. Done properly it retrieves the relevant material and answers from that, rather than guessing, which is what keeps it accurate and on-brand. It can also take actions through your other systems, which is where a chatbot starts to look like an AI agent rather than a FAQ box.

This is the tier we spend most of our time in, and it is the one worth doing when the chatbot is core to how customers reach you. It costs the most and needs the most care, but it is the only tier that can answer questions specific to your business in your words. You can see the kind of grounded systems we run in our live deployments.

How AI chatbot cost really works

Every quote breaks into two parts, and vendors love to show you one and hide the other. Ask for both.

• Build. A one-off cost to design the conversations, connect your systems, load and structure your knowledge, test it and launch. A widget is a small build. A grounded custom bot is a real project with discovery, integration and QA.
• Monthly running cost. This is the part people forget. It covers model usage (you pay per conversation or per unit of text the model processes), hosting, and maintenance as your content and products change. A chatbot is not a plug-in you install once. It needs feeding.

The single biggest driver of ongoing AI chatbot cost is volume multiplied by how much the bot has to read to answer each question. A chatbot that scans your whole knowledge base on every message costs far more to run than one built to retrieve only what it needs. That engineering choice does not show up in a demo, but it shows up on the invoice every month, so ask how retrieval is handled before you sign.

Here is the shape of the three tiers. No prices, because anyone quoting a flat number without seeing your setup is guessing.

Tier	Build effort	Monthly cost driver	Best for
Rule-based widget	Low	Flat, low	Routing, simple lead capture
Platform bot	Low to medium	Conversations or seats	Support teams already on a platform
Custom grounded LLM	Medium to high	Model usage plus hosting	Answers specific to your business

The right question is not “what does an AI chatbot cost” but “what does the tier that fits my problem cost to build and to run for a year”. A cheap build with an expensive, badly-engineered monthly bill is not cheap.

When a chatbot is the wrong answer

This is the part the sales pages skip. A chatbot amplifies whatever is already there. If the underlying situation is weak, it amplifies the weakness. Three cases come up again and again.

You have no meaningful traffic

A chatbot answers people who show up. If barely anyone visits your site, a chatbot has nobody to talk to and it will not fix that. The money is better spent getting the right people to the page in the first place. If that is the real gap, AI-assisted marketing and automation or plain demand generation will do more for you than any bot.

Your knowledge is not written down

A grounded chatbot is only as good as the material it reads. If your policies, pricing and answers live in your team’s heads or in a dozen inconsistent emails, the bot has nothing solid to stand on, and it will either refuse to answer or make things up. The fix is to document the knowledge first. That work is valuable on its own, and until it exists a custom chatbot is premature. This is often where an AI consulting conversation starts, before anyone writes a line of bot.

The real problem is slow human follow-up

Plenty of businesses think they need a chatbot when what they actually have is a response-time problem. The enquiries are arriving fine. They just sit unanswered for a day, or fall between two staff members, or never make it into a system anyone checks. A chatbot bolted on top will collect more enquiries and let them rot faster. Fix the follow-up first, usually with routing, notifications and tighter integration between your tools, and you may find you never needed the bot at all.

Choosing well, in one page

Work through it in this order and you will avoid the expensive mistakes:

1. Name the job. Route enquiries, deflect support tickets, or answer detailed product questions. Each points at a different tier.
2. Check the prerequisites. Real traffic, documented knowledge, and follow-up that works. Missing one is a reason to pause, not to buy a bigger bot.
3. Get both numbers. Build and monthly, with the monthly cost driver named. If a vendor cannot explain how retrieval works, keep looking.
4. Start at the lowest tier that does the job. You can always move up. Paying for tier three to do a tier-one job is the most common waste we see.

So which one should you build

For most NZ businesses the honest answer is: the smallest chatbot that actually solves the named problem, on a build you understand and a monthly cost you can predict. If you want a second opinion on which tier fits, or whether a chatbot is even the right tool for what you are trying to fix, that is exactly what our AI chatbots service exists to sort out before you spend anything.