Using Newton AI to Improve Customer Service for Your Small Business

Using Newton AI to Improve Customer Service for Your Small Business

Most small businesses lose customers in the gap between "I have a question" and "someone answered me." The question arrives at 9:40 PM on a Saturday, or while you're on a job site, or on the fifth page of your website where nobody is watching. By Monday, the customer has already called someone else.

Newton AI closes that gap. It's an embeddable chat agent that you train on your own content — your website, your PDFs, your FAQ, even your YouTube videos — and drop onto your site with a single snippet of JavaScript. It answers your customers in seconds, in your voice, around the clock, and it bills by usage instead of by headcount.

This guide covers what Newton AI actually does, what it costs, how to train it so the answers are right, and how to tell whether it's working.

What is Newton AI?

Newton AI is a customizable AI chat agent that you embed on your website to answer customer questions using your own business content. It is one of three products on the Chicago AI platform, alongside Hemingway AI for content and Archimedes AI for software work.

The practical difference between Newton AI and a generic chatbot is where the answers come from. A generic chatbot answers from whatever the underlying model learned on the open internet. Newton AI answers from a knowledge base you build out of your own material, so when a customer asks about your return window, your service area, or your Tuesday hours, it answers with your policy — not a plausible-sounding guess.

How it works, in three moving parts

1. Training. You point Newton AI at your content. The platform crawls and chunks it into a searchable knowledge base tied to your API key.
2. Retrieval and response. When a visitor asks a question, Newton AI pulls the most relevant pieces of your content into context and answers from them.
3. The embed. A small JavaScript plugin injects a self-contained chat window onto your page. It's isolated in an iframe, so it can't collide with your site's CSS, and it resizes itself as the conversation grows.

Why small businesses use AI for customer service

Answer time drops from hours to seconds

This is the benefit that actually moves revenue. The majority of questions arriving on a small business website are repeats: hours, location, pricing, availability, "do you service my area," "do you carry X." Those questions don't need judgment, they need speed. Newton AI answers them immediately, which keeps a browsing visitor on your site instead of sending them back to search results.

Coverage outside business hours

Your staff works business hours. Your website doesn't. Newton AI covers evenings, weekends, and holidays without overtime, and it's just as responsive at 2 AM as it is at 2 PM. For businesses whose customers shop after work, this is often where most of the value shows up.

Cost that scales with usage, not with headcount

Newton AI is billed in pre-paid usage credits rather than a per-seat license. You're charged per 1,000 tokens for chat traffic, and separately for the tokens in your training materials at the time you train. A quiet month costs less than a busy one. Custom rates and subscription plans are available on request, and the current published rates live on the Newton AI billing FAQ.

Your team stops answering the same question 40 times

Deflecting repetitive questions isn't just a cost saving — it's what frees your people to handle the conversations that actually need a human: the upset customer, the complicated quote, the custom order. Newton AI handles tier one so your staff can handle the exceptions.

Consistency

A trained agent gives the same correct answer to the same question every time. No new-hire ramp, no half-remembered policy, no contradictions between whoever happens to pick up the chat.

Training Newton AI on your business

Training is where a Newton AI instance becomes genuinely useful, and it's the step most people underinvest in. The platform accepts several kinds of source material:

A few things worth knowing before you train:

• Excluded URLs matter as much as included ones. Cart pages, tag archives, and paginated blog indexes add tokens (which you pay for) and dilute the quality of retrieval. Exclude them.
• Your unpublished documents are often your best training data. The internal one-pager that answers every question a new employee asks is exactly what your customers are asking too.
• Retraining is how you keep it accurate. When your hours, prices, or services change, retrain. An agent confidently quoting last year's pricing is worse than no agent at all.

Full details are on the Training Newton AI FAQ.

Setting up Newton AI: the actual steps

1. Create an account. Register for a Chicago AI Portal account. Creating the account is free; you fund usage credits when you're ready to train and go live.

2. Provision your instance from the AI Portal home page. You'll get an API key that identifies your agent.

3. Train it on the sources listed above.

4. Customize it — appearance, greeting, and behavior (below).

5. Embed it. The Portal generates a copy-and-paste snippet for your instance. It's two script tags, dropped in before the closing </body> tag of your site template — the plugin itself, then one call to useNewton with your API key:

   <script src="https://cdn.chicagoai.io/files/chicagoai/js/chat/newton-plugin.min.js"></script>
   <script>
     useNewton({ apiKey: 'YOUR_NEWTON_API_KEY' });
   </script>
   

   Because it's a plain script tag, it works anywhere you can edit a template or add a custom HTML block — WordPress, Shopify, Squarespace, Wix, or a hand-built site. No plugin ecosystem, no server-side install.

6. Read the first week of transcripts. This is the step people skip and shouldn't. See "Measuring what matters" below.

The Newton AI Implementation FAQ walks through the configuration options in detail.

Customizing Newton AI for your brand

The chat window is configurable from the Portal, and the settings that most affect customer experience are:

• Custom CSS, header text, and welcome message — so the widget reads as part of your site rather than a bolted-on third-party tool. The welcome message is prime real estate: "Ask me about our services, hours, or pricing" produces far better first questions than "Hi! How can I help?"
• Response randomness — lower values make answers more literal and repeatable, which is usually what you want for policy and pricing questions. Higher values sound more conversational.
• Knowledge depth and response length — how many pieces of your content get pulled into each answer, and how long the reply can run.
• Chat context (history) — lets the agent remember earlier turns in the conversation. It costs tokens, and it's almost always worth it.
• Timestamps, session indicators, and delete-history controls — small transparency touches that make the widget feel trustworthy.
• Optional tools — web search and image generation can be enabled per instance where they fit the use case.

Keeping your data where you want it

This is the part most chatbot comparisons skip, and it matters for anyone in healthcare, legal, finance, or any business handling sensitive customer detail.

Newton AI instances can run against an OpenAI-hosted model or against a self-hosted model — including a fine-tuned Llama variant served from your own Ollama instance. In that configuration, chat traffic never leaves infrastructure you control. It's the same product and the same embed snippet; only the inference backend changes.

Most small businesses start on a hosted model because it's simpler, and move to a private model if and when their data handling requirements call for it. The migration path exists either way, which means the decision isn't permanent.

Measuring what matters

Newton AI collects engagement and conversation data, and the Portal exposes metrics and transcripts for your instance. The numbers worth watching:

• Questions Newton couldn't answer well. This is your highest-value report. Every one of them is either a gap in your training data or a gap in your actual website — and fixing the second one helps your SEO too.
• Which pages drive conversations. If one product page generates a flood of the same question, the page needs an edit, not just a better chatbot answer.
• Conversations that end in contact or purchase intent. The point isn't chat volume; it's whether chat is producing customers.
• Repeat question clusters. Anything asked constantly belongs in your welc