Your support inbox is answering the same forty questions forever.
We build AI agents that handle the repetitive part of customer conversation — sizing, stock, order status, returns policy, product fit — and escalate anything they should not be deciding alone.
Free scoping · pilots from $12,000 · tested before it talks to a customer · you own the system
Off-the-shelf chatbots made this worse, not better.
The generic widget answers from a help-centre article, gets the specifics wrong, and the customer asks for a human anyway — now annoyed. So the team turns it off, and support goes back to answering “will this fit me” by hand, at volume, at the exact moments when conversion is decided.
The difference is not the model. It is whether the system is connected to your actual catalogue and order data, and whether anyone tested what it does at the edges before a real customer found them.
Grounded in your data, not the open internet.
Answers from your catalogue
Product attributes, sizing, materials, availability and policy come from your store data, so the agent is not improvising plausible-sounding detail.
Recommends, with reasons
Takes a vague request — “something light for summer, under a budget” — asks the one or two clarifying questions a good salesperson would, then narrows to real products.
Handles order status and returns
The highest-volume, lowest-judgement tickets, resolved without a person, inside the channel the customer already used.
Escalates deliberately
Complaints, refunds outside policy, anything involving a payment dispute or an upset customer goes to a human with the conversation attached.
Works where your customers are
On-site chat, email, Instagram or WhatsApp — whichever channels actually carry your volume, rather than all of them by default.
Tested before launch
Defined behaviours, adversarial cases and a review pass before it speaks to anyone. This is the part most agencies skip and it is the reason these projects fail.
We built it for ourselves first.
Before pitching this to anyone, we built a commerce agent end to end as an internal project — vague query in, clarifying questions, catalogue-grounded recommendation, human handoff when needed.
It is a concept build, not a client deployment, and it is labelled that way wherever it appears on this site.
3–6 weeks to pilot
One channel, one defined job, measured against how it is handled today.
Yours outright
Code, prompts, configuration and accounts. No platform you cannot leave.
Agreed overlap
A fixed daily window that matches your team’s hours, set before the engagement starts.
Published starting prices.
A fixed price agreed in writing before any work begins. These are the starting points; the exact figure follows the scoping conversation.
Free
We read your most repetitive support threads and tell you which an agent should handle — and which it should not.
From $12,000
Three to six weeks. One channel, one defined job, tested against adversarial cases before launch.
At cost
Passed through with no markup, and shown to you before you commit to them.
Shown in USD. Quoted and invoiced in your own currency — GBP, EUR or AUD. You own the code, prompts and configuration outright.
Before you commit
How is this different from the chat widget we already tried?
Most widgets answer from help-centre text and have no live connection to your catalogue or order system, so they are confidently wrong about specifics. We connect the agent to your real data and define, in advance, what it is allowed to decide and what it must escalate.
What stops it from inventing product details?
The agent answers from your catalogue and order data rather than generating from general knowledge, and we test it against cases designed to make it guess. Where it cannot ground an answer, it is built to say so and hand over rather than improvise.
Which platforms does it work with?
Shopify and WooCommerce are the most common starting points, and we have tested commerce plugins and integrations across WordPress, Drupal, Shopify and BigCommerce. If your stack is custom, the integration work is scoped explicitly rather than assumed.
What happens when it gets something wrong?
Escalation paths are part of the build, not an afterthought: a human gets the full conversation and context. We also review the cases it handled badly after launch and tighten the behaviour, which is why the pilot is scoped to one channel first.
How do you price this?
Pilots start at $12,000, quoted in your own currency, as a fixed price against a defined deliverable agreed before work begins. The scoping session is free. Ongoing model and platform costs are passed through at cost with no markup, and shown to you before you commit to them.
What if an agent is not the right solution?
Then we will tell you. A better FAQ page, a sizing guide or a fixed automation solves a surprising number of these problems more cheaply and more reliably than an AI system does.
Start with the forty questions.
Send us your most repetitive support threads and we will tell you honestly which of them an agent should handle — and which it should not.
Scope a pilot →