A few months ago, a business owner came to us with a problem that's more common than most people admit: their inbox was full of the same customer questions, over and over.
Where's my order? Is this item in stock? What's your return policy? Can I change my shipping address?
Their team was handling it — but barely. These weren't bad employees. They were good at their jobs. Customer support just wasn't their job, and it was eating hours every week that should have gone toward running the actual business.
Hiring a dedicated support person wasn't on the table. The volume didn't justify a full-time hire, and a part-time hire felt like a band-aid. So they lived with it.
We built them an AI that handles all of it. It runs around the clock, responds in their brand voice, checks real-time order status, and escalates only when a human judgment call is actually needed.
Total cost to run last month: $1.02.
What the business looked like before
This is a 30-person business. Not a startup, not an enterprise — the kind of company that's been around for a while, has real customers, and is growing steadily. They sell physical products and ship orders regularly.
Before we built anything, their customer support workflow looked like this:
- Customers emailed a general inbox or submitted through a contact form
- A team member — whoever had a moment — would check the inbox and respond
- Answering an order status question meant logging into the order system, looking up the order, copying the tracking info, and writing a reply
- Response time was anywhere from a few hours to the next business day
- Nothing happened on nights or weekends
The team estimated they spent 8–10 hours per week on customer emails. Most of it was the same handful of questions.
The cost of that time — even at a modest hourly rate — was well over $1,000/month. The actual dollar cost of what we built to replace it: about one dollar.
What we built
The short version: an AI that reads incoming customer messages, figures out what they need, and responds — pulling from their order system and a knowledge base we built together.
Here's how it actually works, in plain terms:
Step 1 — A customer sends a message
Doesn't matter when. Doesn't matter how. Email, contact form, it all routes to the same place.
Step 2 — The AI reads it and categorizes the intent
Is this an order status question? A product question? A return request? A complaint? The AI identifies what the customer actually needs before deciding how to respond.
Step 3 — It looks up the relevant information
For order status questions, it connects directly to their order management system and pulls real-time data — tracking numbers, shipping status, estimated delivery. No logging in. No copying. Live data, pulled automatically.
For product and policy questions, it references the knowledge base we built with them — a structured document covering their most common questions, return policies, shipping options, and product details.
Step 4 — It responds in their voice
We spent time at the start of the project calibrating the tone. The AI doesn't sound robotic. It doesn't say things their team would never say. We wrote example responses with the client, defined their brand voice, and trained the system to match it.
Step 5 — It escalates when it should
Not every message gets resolved automatically. If a customer is angry, if the situation is complicated, if the question falls outside what the AI can confidently answer — it flags the message and notifies the right person on the team. They get a summary of the conversation and context so they can jump in without starting from scratch.
The numbers
Response time went from hours to under 30 seconds. The team went from spending 8–10 hours a week on email to reviewing a short escalation queue — usually 3–5 messages per day that actually need them.
Why does it cost so little to run?
This is the question we get most often, and it's a fair one.
The underlying AI technology is priced per use — specifically, per word (or "token") processed. Not per seat. Not per month. When a customer sends a message and the AI reads it and responds, that exchange costs a fraction of a cent.
For a typical small business receiving a few hundred customer messages per month, the total AI cost comes to a few dollars at most. This business happens to run on the lower end of that range — hence the $1.02.
The infrastructure to run it (servers, integrations, the logic that ties it all together) involves small ongoing costs too, but the total is still negligible compared to what any equivalent staffing would cost.
What this took to build
We spent two weeks on this project start to finish. Here's roughly how that time broke down:
- Week 1: Understanding their business — how their orders work, what their common questions are, what their team's voice sounds like, what escalation scenarios matter
- Week 2: Building, testing, and calibrating — the integrations, the knowledge base, the response logic, the escalation rules
- Handoff: A training session with their team so they understood exactly what they had, how to update the knowledge base as things change, and what to do if something behaves unexpectedly
They own it completely. It runs on their accounts. There's no ongoing fee to us unless they want us to keep finding new projects to build.
What this means for your business
If your team is spending time answering the same questions from customers — whether by email, phone, or chat — there's a good chance AI can handle most of it. The technology is there. The cost to run it is negligible.
The gap for most small and mid-size businesses isn't the AI itself. It's the implementation: figuring out what to build, building it to work with your actual systems, calibrating it to sound like you, and making sure your team knows how to manage it.
That's the work. And it's the work we do.
Curious what this could look like for your business?
That's exactly what our first call is for. We'll look at how your business operates and tell you where AI creates the most value, fastest.
Book a Free Call