Service · E-commerce automation

Support, content and translation
running off one knowledge baseNot four subscriptions that have never met each other

Most storefront teams already have tools for each of these. The problem is that each tool keeps its own copy of the truth. Put them on one knowledge base and something new becomes possible: what support gets asked decides what gets written, and what gets written is what support answers with next.

Live in

3–4 weeks

Languages

10+ from one source

Deployment

Your own infrastructure

Priced per seat

No

What this is for

Four problems that are really one problem

Each of these looks like a separate tooling gap, and gets solved by buying a separate tool. They're the same gap: no single place that knows what your business says.

The same question, thirty times a day

Sizing, compatibility, shipping, warranty. Every answer already exists somewhere — in a policy doc, in an old ticket, in one person's head. It just isn't anywhere a machine can reach.

Content stalls on topics, not on writing

Most storefront blogs die because nobody knows what to write next, not because writing is slow. Meanwhile your support inbox is a list of exactly what your buyers want explained.

Every language drifts apart

One policy change, ten storefronts, and for the next week each of them says something slightly different. Nobody notices until a customer quotes the wrong version back at you.

The tools don't know each other

A helpdesk, a translation service, a content tool, an analytics dashboard. Four subscriptions, four copies of the truth, and no single place that knows what your business actually says.

What gets built

Three lines, sharing one foundation

Described by what they do for you, not by how they're built. The specific components and models are listed line by line in the proposal — along with anything that would leave your boundary, which you can veto individually.

Support agent

Trained on your own tickets, policies and product facts — not on a generic model's guesses about your business. Most enquiries close without a person, and the ones that shouldn't be automated aren't.

  • Answers cite what they're based on, so a reviewer can check them
  • When it can't find an answer it escalates instead of improvising
  • Anything involving a commitment, a refund or an exception goes to a person by design
  • Unanswered questions become a queue of knowledge to add — reviewed by a human before it takes effect

Content production

Articles and product copy produced as a pipeline, with topics coming from evidence rather than a brainstorm. A person signs off; nothing publishes unreviewed.

  • Topics from three sources: search gaps, recurring support questions, and site searches that returned nothing
  • Once an article exists, support cites it back — the same question stops being answered thirty times a day
  • Product copy, imagery and listing prepared together rather than in three separate queues
  • Translated from one source of truth, with a shared glossary across every language

Ads, social and reporting

We do the homework and leave the spending decisions with you. The phrase we use internally is: we don't run your account, we finish your homework.

  • Channel data reconciled against real on-site orders, not platform self-reported ROAS
  • Diagnosis: which sets burn without converting, which creatives are decaying, which audiences have room
  • Ad copy in multiple languages and variants for testing, and creative through the image pipeline
  • Social scheduled and published through official APIs using your own authorisation, revocable at any time
  • One dashboard where anomalies surface on their own and the weekly summary is a conclusion, not a table

The loop worth noticing: what support gets asked most becomes the next article, and once that article exists support answers with it. A question goes from being answered thirty times a day to one piece of content plus one knowledge entry. That loop only closes when support and content read from the same knowledge — which is exactly what buying two separate products cannot give you.

Measured, not projected

Four figures, each with its source

A performance number without a stated source reads as marketing to anyone technical. So each line names the record behind it.
Content throughputsource: publishing logA long-form article went from 8 hours to 16 minutes, of which 15 minutes is human final review.
Support automationsource: agent transcripts80%+ of incoming enquiries close without a human; the rest are escalated by the agent itself.
Product imagerysource: image production logPer-image cost fell from ¥200 outsourced to ¥10, with a batch of 24 going from 10 days to 0.5 hours.
Multilingual syncsource: translation workflowA policy change reaches 10+ languages within 1 minute, down from 5 days.

These are measured results, not a commitment about your numbers. Where they came from and the full set of seven is on the case study →

Where people stay

Four things that never happen unattended

These aren't limitations we're working to remove. They're where the system is designed to stop and wait for someone.

Anything that commits you

Refunds, exceptions, goodwill, anything that changes what a customer is owed.

Anything that spends money

Budget and bid changes arrive as a recommendation with its reasoning, never as an action.

Anything published outward

Articles, product copy and social posts pass one review — the same checkpoint for all of them, not one per channel.

Anything written into knowledge

New entries queue for confirmation. Nothing writes itself into the source of truth unreviewed.

Boundaries

Four things this doesn't include

Stated up front, because finding out at the proposal stage wastes both our time.

We don't promise search rankings

What you get is a content pipeline and the capacity to run it. Positions depend on your competitors, your domain history and a search engine nobody controls, so we don't put it in a promise.

No link building, no scraped or spun content

No purchased links, no private blog networks, no rewriting other people's articles. If that's the growth plan, we're the wrong supplier.

We don't hold your ad accounts

We don't take custody of ad accounts and we don't set budgets or bids. Not a capability limit — an accountability one: the decision to spend has to sit with the person answerable for it.

No social community management

Scheduling and publishing, yes. Replying to comments, running the community and negotiating with creators stay with your team.

Questions

What people ask about this one

The first is the most common, so it's first.
01Will AI-written articles get us penalised?+

What gets penalised is content with nothing behind it, in any language and by any author. The pipeline is built so that doesn't happen: topics come from questions your buyers actually asked, the facts come from your own knowledge base rather than a model's memory, and a person reviews before publication. That last step is 15 minutes of the 16 for a reason.

02What if the support agent answers something wrong?+

Three things reduce it and one accepts it. It answers from your own knowledge rather than improvising; it escalates instead of guessing when it can't find something; anything involving a commitment goes to a person by design. And every conversation is logged, so a wrong answer becomes a knowledge entry rather than a mystery.

03Do you have to rebuild our storefront?+

No. Where a storefront already exists we connect to it. Building the site ourselves lets the hooks go in at the foundations, which raises the ceiling — but it isn't a precondition, and starting with one workflow on your existing site is the common case.

04How long until it's live?+

3–4 weeks. Whether it's three or four is almost entirely decided in week one: how organised your existing policies, tickets and product data are. The diagnostic happens before that clock starts and isn't counted in it.

05Can our own team maintain it afterwards?+

That's one of the delivery goals. Wording, rules, knowledge entries and notification targets are configurable in the back office, the orchestration layer is visual, and the source code lands in your repository with deployment docs, a runbook and training.

06Does adding a workflow later cost extra?+

Yes, and it's written into the contract at signing rather than raised mid-project. Day-to-day maintenance — parameters, wording, knowledge entries, model and API version upkeep, routine checks — your own people can handle, or you can buy an annual maintenance subscription. A new agent, a new workflow or a new system integration is new development, scoped and priced as a new project. Splitting it this way keeps the accounting honest: folding new work into a maintenance fee either inflates the fee or means the new work gets done carelessly.

Next

Start by finding out which part is worth doing first

You don't have to take all three lines. Most engagements start with whichever one is costing the most hours right now, with the interfaces for the others built in from the start. The diagnostic fee is credited in full against the project.