SquidcomIn 2026 most people are on board with AI chatbots being an essential part of providing good customer service for online businesses. A few years back chatbots were often clunky and frustrating for customers, but today the latest large language models can be harnessed to provide immediate and accurate information to customers, via a conversation that feels natural. Furthermore, this service can be provided at a fraction of the cost of hiring a customer service team, opening the door for any business to vastly improve the experience customers have with their business.
However, the fact that AI chatbots are cheaper than humans has led to many providers charging far more than the actual costs of running a chatbot. We think it's worth taking a closer look at the numbers.
How Chatbots Actually Work
Let's cover some background on how chatbots work nowadays. To provide a chatbot that will respond in a natural way requires using a model from one of the big companies providing Large Language Models (LLMs), and basically that means paying either OpenAI, Anthropic, or Google to use their models. You've likely used one or more of the models provided by these companies already, i.e. ChatGPT, Claude, or Gemini. However, to get these models to provide tailored responses to your customers, i.e. respond with information for your specific products (or anything), requires a process called Retrieval Augmented Generation (RAG), which adds a step between your customer asking a question and the model responding. What RAG does in this step is search through the information you have provided (commonly referred to as the knowledge base), retrieve information relevant to the customer's question, and then send this information along with the customer's question to the LLM, which can then respond based on the information it has been provided with.
To recap: every AI chatbot provider is simply implementing RAG and calling an LLM, and packaging this up in a nice interface. And to be clear, this is exactly what they should be doing (it's what we're doing at Squidcom) because it works really well, and is by far the best way to provide a chatbot nowadays.
The Real Costs
Despite what many providers will have you believe, providing a great chatbot in 2026 is not rocket science. Get an LLM with RAG and you're 95% of the way there. And even more to the point, the cost of running a chatbot, which is almost entirely made up of the costs of calling the LLM, is really low. We're talking fractions of a penny per request. Which raises an obvious question...
Why Are They Charging So Much?
Intercom Fin charges $0.99 per resolution (successful conversation). Let's give them the benefit of the doubt and say they're using a high-quality model like Claude Sonnet 4. LLMs charge per token (roughly 1 token per word). According to Anthropic's pricing, Claude Sonnet 4 costs $3.00 per million input tokens and $15.00 per million output tokens.
Let's say a typical customer conversation involves them asking a few questions and getting some responses—about 100 words (130 tokens) from the customer, and 500 words (650 tokens) from the AI. The maths:
- Input cost: 130 tokens × $0.000003 = $0.00039
- Output cost: 650 tokens × $0.000015 = $0.00975
- Total: approximately $0.01
So when Intercom charges $0.99 for a conversation that costs roughly $0.01 to run, that's a 99x markup. Of course they have other overheads—infrastructure, support, development—but even accounting for all of that, the margins here are hard to justify. And by the way, to access their AI chatbot you already have to be paying them flat monthly fees for their wider CRM system.
Intercom are not the only ones. Many providers will try to lure you in with low starting costs, which scale poorly as your usage grows.
How We Do It Differently
Here at Squidcom, instead of saying, "Well, a human would cost you $1.20 per conversation, so we can get away with charging $1," we say, "Well, the LLM calls cost us $0.01 per conversation, so we'll charge $0.02."
We believe pricing should reflect actual costs, not what the market will bear.
Don't Take Our Word For It
If you're skeptical about the whole "every AI chatbot provider is just doing LLM calls plus RAG" thing, you can try it out for yourself. Spend an hour in OpenAI's playground on a free trial—upload your content to a vector store and start chatting with RAG pulling from your own content. You'll quickly see it doesn't take a whole lot to get a decent conversation that provides relevant information.
What we provide at Squidcom is a way to quickly integrate this technology with your business. The underlying chat functionality is widely available—we just make it simple to deploy.
Ready to see the difference? Try Squidcom free today. 