Starting point · Single module

One workflow moduleDo one thing first. Talk about the next one once it runs.

If none of the three lines fits your business cleanly, this is the page for you. Whether something can be automated isn't decided by your industry— it's decided by whether it has a clear trigger, a process you can describe, and a clear output. That test is the same everywhere.

From

¥28,000 · 1-week tier

Tiers

1 week / 2 weeks / 3–4 weeks

Deployment

Your own servers

Becomes an island

No — by design

What counts as one module

Four parts. If one is missing, it isn't a module yet.

This is the whole test. It's also what the diagnostic checks first, because a request that fails it can't be scoped or priced honestly.

One clear trigger

An email arrives, it's 8am, someone asks in a chat, a row appears in a table. You can say when it starts.

One process you can describe

The rules can be complicated, but they have to be explainable. The part nobody can explain is usually undecided rather than difficult.

One clear output

A row written, a message sent, a document produced, something written back into your existing system. When it finishes, something somewhere has changed.

One human checkpoint

Anything with real consequences — sending outward, spending money, signing off — ends with a person. That's deliberate, not unfinished.

One rough volume test on top of the four: it should happen at least twenty times a week, or be expensive enough when done wrong. Something that happens three times a month isn't worth the project that automates it.

Three things people bring us as “one module” that aren't

“Digitise our processes”

No trigger and no output — that's a project, not a module. Start with a diagnostic to break it into specific things, then pick the first one.

Three departments need to agree on the rules first

The bottleneck isn't automation, it's the rules. We'll say so directly: settle them first, or automation just cements the disagreement.

It needs your existing system's data structure changed

That's system surgery, not a layer on top. It can be done, but it's priced as a project rather than a module, and the timeline and risk are a different order of magnitude.

What tends to get built

Sixteen that come up repeatedly, grouped by function

Grouped by what the work is, not by industry — the same task in a different industry is still the same task, and grouping by industry guarantees missing people. This is a reference list, not a product catalogue; yours may not be on it and that's normal.

Documents and data entry

Documents into your systems

2 weeks

Invoices, order confirmations, customs paperwork and shipping notes keyed in one at a time — dozens a day, and one wrong digit corrupts everything downstream.

Key fields extracted and entered as soon as the attachment arrives; anything uncertain queues separately for a person. Every change is attributed.

Reconciliation exceptions

2 weeks

Platform statements, bank transactions and your internal ledger — three sets of data, compared by hand at month end, late into the night, and still with gaps.

Compared automatically every day, with only the lines that don't match pushed to you, each with both original records attached.

Email and tickets

Email triage and routing

1 week

One shared inbox, hundreds of messages a day — enquiries, complaints, suppliers and spam mixed together, with the important one sitting eightieth.

Classified, prioritised and routed to the right person on arrival, with a reminder if nobody has picked it up in time.

Reply drafting

2 weeks

Eighty per cent of replies say the same few things, and every one of them is still typed from scratch.

Drafted from your own past replies and policies; a person reads it, changes a word, sends. The send button stays with a human.

The full support agent

Document production

Quotes and document drafting

2 weeks

One quote means opening the price list, checking your discount authority and filling a template — twenty minutes minimum, with the customer waiting.

Give it the model and quantity and a draft comes back; anything beyond your discount authority is flagged and routed for approval automatically, anything within it goes straight out.

Meeting notes and action routing

1 week

The meeting ends and nobody writes it up; where notes exist, nobody chases the actions. The next meeting starts with twenty minutes of remembering.

Recording in, notes out, actions split by owner and pushed into each person's own tool, with automatic chasing when they come due.

Languages

Bulk translation with locked terminology

2 weeks

Product material has to exist in five or six languages; outsourcing takes two weeks a round, and the same component ends up with three different translations across three documents.

A glossary locks the key terms, the batch is produced at once, and people only review the pages that matter.

The full multilingual system

Inbound mail in other languages

2 weeks

German and French mail goes through a translation tool, then the reply goes back through it — two days gone on one exchange.

Incoming mail arrives in a language you read with the original attached; replies go out in the sender's language, with terminology matching your own material.

Imagery and assets

Batch product imagery and sizing

3–4 weeks

Every launch needs hero, detail and social sizes; outsourced, that's tens per image, three days of waiting, and the set still doesn't match.

A batch goes in and comes back as complete sized sets built to your visual standard, consistent across the whole batch.

The full visual pipeline

Re-versioning a whole asset set

2 weeks

One promotional change means redoing badges, prices and copy across hundreds of images.

Change the template once and the batch re-renders — no editing one by one, and no size quietly missed.

External data

Competitor price and stock monitoring

1 week

Someone checks a few sites by hand, incompletely and late. By the time you notice a competitor's price drop, it's a week old.

Collected and compared daily, with an alert only when a change exceeds the threshold you set. History is kept, so you see a trend rather than a snapshot.

Scheduled collection of public data

1 week

Exchange rates, raw material prices, policy notices and tender listings live across a dozen sources, and exactly one person sees all of them. When they take leave, it stops.

Collected on a schedule into one table, outliers flagged automatically, and material changes pushed into the channel you already watch.

Internal knowledge

Q&A for one department

3–4 weeks

HR, finance and IT answer the same dozen questions every day, and their real work arrives in fragments between interruptions.

Just this one department's policies and common questions loaded first; colleagues ask in the group chat and get an answer with its source, escalating when there isn't one.

The company-wide version

Capturing what the veterans know

3–4 weeks

Two people in the company can tell whether a batch of material will cause problems, and one of them retires next year.

Structured interviews surface those judgements and turn them into entries a new hire can find and ask about.

The same method at scale

Chasing and reporting

Approval chasing

1 week

A document sits with someone for three days, nobody knows where it stopped, and asking around reveals they're travelling.

Chased automatically when due, escalated to their manager after two rounds, with a full trail of who released what and when.

Automatic daily and weekly reports

2 weeks

Someone spends half a day a week copying numbers out of several systems into one table that nobody reads closely.

Generated on schedule with every figure sourced; anomalies surface on their own rather than waiting to be spotted in a table.

How one module becomes several

Four modules over nine months, never scoped as one project

An industrial parts exporter, a bit over twenty people, customers in Europe. No storefront of their own, no existing knowledge base — none of the three productised lines fits them cleanly. They started with the single most annoying thing.
Module 12 weeks

Enquiry email produces a draft quote

Previously, one enquiry meant opening the price list, checking discount authority and filling a template — twenty minutes minimum, twenty enquiries a day in season, and quotes always half a day behind competitors. Afterwards: the model and quantity are pulled from the mail automatically and a draft is in the salesperson's inbox within two minutes, with any discount beyond their authority flagged on the spot. The send button stays with the salesperson.

This was the entire scope. Nobody was discussing a “full line” at this point.

Two months innoticed by a person

The real bottleneck surfaced on its own

Quotes did get faster, but deals didn't. The run log showed why: on about a third of quotes, the customer came back asking about technical specifications and compatibility, so the salesperson had to go and find an engineer — another day gone each way.

Nobody raised this requirement at kickoff. It only became visible once the first thing was running.

Module 22 weeks

Product material into a small knowledge base

Specification tables, compatibility relationships and the few dozen most-asked questions were collected into a small base, and the draft quote started carrying the relevant specifications and compatibility notes with it. The answer is in the first quote the customer receives.

This didn't replace module one — it attached to it. The interface was built into module one from the start.

Module 31 week

Follow-up after a quote goes out

Seven days without a reply and the responsible salesperson is reminded, with the outcome recorded. No more quotes disappearing into silence in peak season, and the quarterly review can finally say at which step deals were lost.

Module 43 weeks

Website enquiries with first replies in three languages

The small knowledge base was connected to the website enquiry form, with automatic first replies in English, German and French carrying specifications and common answers, tagged by source into the sales follow-up list.

By this point four modules share the same product material, and this is effectively part of the storefront operations line — except it was never scoped as one project.

You can stop at any step

Four separate projects, each costing what one thing costs. If a step isn't worth it, stop there — what's already live keeps running and doesn't depend on the rest of the plan.

What to do next gets discovered, not designed

Modules two, three and four only became visible after the previous one was running. No requirements workshop produces “quotes got faster but customers ask about specifications next.”

Is four projects more expensive than one? Slightly

There's extra scoping and integration each time, and we're not going to pretend otherwise. What you buy with it is the ability to stop at any point, with every step already returning something. If you already know you want the whole line, take the whole line — that's what we'll recommend, rather than selling it to you in four pieces.

The path above is an illustrative example, rewritten and anonymised: the structure and the cause-and-effect are real, the company details and figures are a sample.

The question worth asking

Does doing one small thing leave us with an island?

It does — if it's built the way an automation script is bought. The reason those are abandoned after three months is always the same: the data lives with the supplier, there's no interface out, and any change means rewriting the whole thing. Three commitments close off all three.

The data lands in your own store

Every record the module produces is written to tables on your own servers — not held by us with a lookup interface bolted on. Stop using the module and the data is still there, clean and exportable.

Interfaces are built in from the start

The first module's inputs and outputs are designed on the assumption that something will connect to them later. So the second thing attaches instead of replacing — and that isn't a line item, it's the default.

It runs alone and can also be called

Each module is complete in itself and doesn't depend on anything not yet built, while also being callable as a step by later orchestration. Stopping at the first one works; carrying on works.

The three together say one thing: the value of the first thing doesn't disappear because you skip the second, and doesn't have to be rebuilt because you want it.

What stays under your control

Three things that have to be in your hands

No exception for a small module — otherwise it's another black box you don't govern, which is no different from a subscription.

Where it runs

On your own servers or private cloud, with data inside your boundary. We don't operate it and we don't keep a back-door account; you grant temporary access if you want help.

When it stops and waits

Which actions need a human nod and which can pass automatically is yours to set, effective immediately, not hard-coded.

Source code and run records

Code into your repository, built from open standard components. Every run and every human confirmation is logged, traceable and exportable.

Timeline and cost

Three tiers, plus two honest things about money

The diagnostic happens before the project and isn't inside these timelines. Which tier applies depends on whether the data already exists and how many systems have to be connected — both known by the end of the diagnostic, so what you get is a figure, not a range.
Single module from¥28,000one-week tier · quoted in CNY
1 weekData already exists, one trigger and one output, no third-party system to connect. This is the tier the starting price corresponds to.
2 weeksData needs organising first, or one existing system has to be connected. The most common tier in practice.
3–4 weeksSome groundwork is needed — collecting material, building a small knowledge base, or connecting several systems.

A second module is a separate project

This is a billing rule written into the contract at signing rather than raised mid-project. Day-to-day maintenance — parameters, wording, an extra rule, thresholds, routine checks and incident response — your own technical people can carry, or you can buy an annual maintenance subscription. Adding a second module is new development and is scoped and priced on its own.

Upgrading to a full line credits what you've paid

If you later take an entire line, modules already delivered are credited at their original price against it, and that's in the contract. It exists to answer the obvious suspicion — that splitting work into pieces is a way to charge more.

Questions

What people ask about this one

The first is the most common, so it's first.
01We're not e-commerce and we have no knowledge base. Can this still work?+

Yes, and that's precisely why this page exists. The three productised lines were organised around the two most common kinds of client — they aren't the limit of what can be built. Whether something can be done isn't decided by industry, it's decided by whether it has a clear trigger, a process you can describe, and a clear output. Manufacturing, wholesale, export trade, engineering, clinics, law firms, training providers — we apply the same test. During the diagnostic we'll tell you which few are worth doing first and which aren't worth doing at all.

02What does one module cost?+

From ¥28,000, which corresponds to the one-week tier: data already exists, one trigger and one output, no third-party system to connect. The two-week and 3–4 week tiers are quoted against how many systems have to be connected rather than carrying a flat price — the same task can differ threefold between two companies depending on how many systems are involved, whether the data exists, and how settled the rules are. But which tier it lands in and how many systems are involved are both known by the end of the diagnostic, so what you receive is a figure rather than a range. The diagnostic fee is credited in full against the project.

03If we start small, are we locked in to you?+

The opposite, and it's one reason we suggest starting with one thing: evaluating a supplier costs least while the commitment is small. On delivery the source code goes into your own repository, built entirely from open standard components with no proprietary black box, and the data sits on your own servers. So after the first module you already hold everything that can be handed over — and whether there's a second one is a decision you make with real run logs in hand rather than on our word.

04We have our own developers. Can you do only the hard part?+

Yes, and it's a common arrangement. Your people take the business-system side, we take the model, orchestration and pipeline side, with the interface agreed between them. There's a handover session at delivery, after which day-to-day adjustments are yours and we only come back for new development.

05Can we change it ourselves after launch?+

Two categories. Parameters, wording, an extra rule, thresholds and who gets notified are configuration — your own people can change those, and we teach that at handover. Changing the processing logic, adding a new output or connecting a new system is development and needs someone who writes code, which can be your team or us.

06When it fails while running unattended, who notices?+

Not you, checking a dashboard. Failure modes are handled three ways at build time: what can retry, retries itself; what's uncertain doesn't proceed and queues for a person; a genuine fault pushes an alert into the channel you already watch, naming which item and which step. Every run is logged, so afterwards you can find when it started going wrong. This layer isn't an add-on — it's there by default.

Next

Bring the most annoying thing your team does by hand

You don't need a shortlist. One task, roughly how often it happens, and what goes wrong when it's done badly is enough for us to say whether it passes the four-part test. The diagnostic fee is credited in full against the project.