Customer support AI
Answers on your product range, delivery terms, and refund policy. Handles the repeat questions end to end and passes the unusual ones on with the thread attached.
Site, WhatsApp, Instagram
Automation & AI
An AI layer that carries a real shift.
Support that answers, qualification that routes, and an admin agent you can message like a colleague. Trained on your catalogue, your terms, and your tone.
What we actually do
We build the AI layer around how you already sell.
Most businesses do not have an AI problem. They have a volume problem that looks like an AI problem. Enquiries arrive faster than anyone can answer them, the same eight questions get typed out forty times a day, and the person doing the typing is also the person supposed to be running the company.
The layer we build sits exactly there. It answers what is answerable, escalates what is not, and keeps a record of both. It knows your catalogue, your prices, your delivery windows, and the way you speak to customers, because it is built on those rather than on a template with your logo dropped in.
The test we hold it to after launch is one question: how many conversations still reach a human, and are those the ones that genuinely needed a human? If the answer is wrong, we tune it until it is right.
- Customer support AIBuilt on your actual product range, delivery terms, and refund policy. It handles the repeat questions end to end and passes anything unusual to a person with the full thread already attached.
- Enquiry qualificationEvery message and form gets asked the two or three questions that decide whether it is worth your time, then gets routed accordingly. Serious buyers reach you warm and already understood.
- An admin agent you can messageAsk what is pending, what sold today, or what needs attention, and get the answer in Telegram. No dashboard login, no opening a laptop at eleven at night.
- The workflows behind itOrder confirmations, follow-ups, stock notices, receipts. The repetitive middle of the day, written once and then running on its own.
The catalogue
Nine things we build into the AI layer.
Almost nobody needs all nine. On the call we tell you the two or three that would move your numbers this quarter, and which ones to ignore.
An agent that books calls
Reads what someone needs, checks whether they are a fit, then puts a real slot in your calendar without a back-and-forth thread about timezones.
Site, DMs, email
Enquiry qualification
Every message gets asked the two or three questions that decide whether it is worth your time, then routes accordingly. Serious buyers reach you warm.
Forms, DMs, inbox
An admin agent you can message
Ask what is pending, what sold today, or what needs attention, and get the answer where you already are. No dashboard, no laptop at eleven at night.
Telegram, WhatsApp
Workflow automation
Order confirmations, receipts, stock notices, review requests. The repetitive middle of the day, written once and then running without anyone.
Behind the scenes
Follow-up and re-engagement
The people who asked a price and went quiet. The agent picks the thread back up at the right moment, in your voice, without feeling like a blast.
Email, DMs
Internal knowledge assistant
Your policies, prices, and process, answerable in one question. New staff stop interrupting the person who knows everything.
Slack, internal web
Recommendation engine
Reads what somebody actually asked for and points at the right item from your catalogue, with stock and delivery already checked.
Storefront, chat
Reporting you can ask questions of
Instead of a dashboard nobody opens, a summary you can interrogate. Ask why last week dipped and get the actual reason.
Telegram, email digest
Wintech AI on this site is a working sample of the first two. Try it
How it gets built
Narrow first, then wider as accuracy holds.
Launching an agent across the whole business on day one is how they end up switched off in month two. We go the other way.
Week one
We read the last month of your real conversations. The tone, the top questions, and the edge cases all come out of that, rather than out of our assumptions about your industry.
Week two to three
The first agent goes live on a deliberately narrow scope, usually support, where a mistake is cheap and immediately visible. You watch every conversation during this stretch.
Week four onward
Scope widens once accuracy holds: qualification next, then the admin agent, then the workflows sitting behind both. Each expansion is a decision you make with data in front of you.
Handover
You get the prompts, the routing rules, and the escalation logic in a form your team can read and edit. If we disappeared, the system would keep running and someone on your side could change it.
Where it already runs
Two builds you can open right now.
Virelle Wellness Hub
No website, then a full operating layer.
They were running the entire business through Instagram DMs. Ads pushed people into an inbox one person could answer, so anyone who wanted to browse quietly simply left.
- E-commerce store people can order from without a conversation
- Support AI on the shop, built on the real product range
- Admin agent on Telegram reporting pending orders on request
Website, brand & AI · Live
Mylo
An agent that lives inside the product.
Rooms, shared goals, and a room agent handling Today commands, voice input, and reminders for everyone in the room. The AI is a participant here rather than a widget in the corner.
- Goals scoped to today, the week, and the month
- Voice input and reminders for the whole room
- Shared notes and polls in the same surface
Product & AI · Live
Automation projects start at $8,000 and run six to ten weeks. See a range for your brief