Case study · Live · Client delivery

Seven departments,
one foundationA retailer running physical stores alongside its own storefront

Before this, the problem wasn't an absence of systems. It was that one order travelled between seven departments as screenshots and group messages — every handover meaning somebody retyped it. Those seven departments now sit on one foundation, sharing one knowledge base, one permission model and one run log. The system runs on their own servers, and when a new role needs help, they add it themselves.

Departments

Seven

Where it runs

The client's own servers

New roles

Built by them, on the platform

Figures on this page

None — see below

Before

Not an absence of systems — facts breaking between departments

None of the four below is “missing a tool”. They're four breaks in the same thing: the facts not reaching the person who should see them.

Head office waits for the facts

Every store kept its own records and filed its own summary, so head office saw a version that had been through somebody's hands a day or a week later. Who adjusted what along the way was not recoverable afterwards.

Answering “which store had the most returns yesterday” started with asking someone to build a sheet

The same job, done again in seven places

One order moved from greeting to sale to invoice to accounts to fulfilment to after-sales through screenshots and group chats. Every handover meant somebody retyping it, and every mistake was found by the next department along.

Entering it took longer than doing it

Online and in-store were two sets of books

Storefront orders, products and content lived in one back office; footfall, samples and transfers lived in another, or in nothing at all. What one customer had browsed online and tried in a store never met.

The most valuable signal in the business was never connected

Small improvements couldn't get scheduled

Business teams could name several half-day savings every month, but each one joined the same development queue. By the time one was built, the way that work was done had often changed.

Not a capacity problem — every small thing shared one queue

Seven departments

What each one actually handed over

The left of each card is the specific work handed over; the closing line is the judgement that stayed in that department, in full. How that line gets drawn is on the role map.

Finance

  • Purchase orders, invoices and statements extracted into tables instead of retyped
  • Store takings reconciled against system orders, with only the mismatches pushed to a person
  • Expense pre-checks: over-policy, missing receipt, duplicate claim, flagged on the spot
  • Ageing and collection reminders sent by customer tier
  • Month-end aggregation and a first draft of the reports

Stays with people: Payment and credit approvals, what a discrepancy means, tax treatment — none of it moved.

People

  • Inbound CVs screened and structured against each store role's requirements
  • Interview scheduling and candidate status kept in sync, so nobody is dropped
  • Joiner and leaver checklists dispatched, each department receiving only its own line
  • Attendance and rota anomalies across stores surfaced
  • Policy questions: leave, expenses, contributions, store allowances

Stays with people: Hiring, pay bands, performance, conversations, and every exception involving a person.

Engineering

  • Business requests and store feedback condensed into tickets: classified, de-duplicated, linked to a module
  • Interface and configuration docs updated alongside the code
  • First drafts of code, test cases and migration scripts
  • Production logs clustered, with similar past incidents and their fixes surfaced
  • Release notes and changelogs drafted

Stays with people: Architecture and technology choices, what ships first, calling an incident. Review time unchanged.

Sales

  • Leads de-duplicated, scored and assigned to a person
  • Customer background and history assembled into a single page
  • Follow-up reminders, escalated to a manager when they slip
  • Quotes and contract documents drafted
  • Won and lost reasons captured and grouped

Stays with people: The negotiation, the concession, the relationship, and the quote that actually goes out.

E-commerce

  • Product launch pipeline: copy → imagery → listing, with humans only on final review
  • Content and terminology kept in sync across languages
  • First-response support, escalating what it can't answer
  • Order anomalies and storefront health flagged automatically
  • Storefront and campaign data collected into one board

Stays with people: Pricing, promotions, channel strategy, and complaints that need a concession.

Design and production

  • Product imagery re-rendered at volume: background, scene, consistent lighting
  • Hero, detail and social sizes exported in one pass
  • Scene renders and short-form video source material at volume
  • Style held to the brand standard so a batch doesn't drift
  • Assets filed and searchable by attribute instead of by folder

Stays with people: Setting the visual standard, selecting from each batch, physical photography.

Store management

  • Daily trading summaries with anomalies flagged: footfall, conversion, returns
  • Walk-in records structured, with follow-up assigned to a named person
  • Sample and stock count lists, inter-store transfer requests routed
  • Inspection photos and remediation items filed, overdue items chased
  • Product knowledge and talk tracks staff can ask for, answers with sources

Stays with people: Serving the customer in front of them, rotas, merchandising decisions, complaints on the floor.

Why this page has no figures

Not an omission — publishing one would drag the others down

Case studies sell better the more specific they get, so this section runs the other way. It applies to you too — how we treat this client's numbers is how yours will be treated.
01

They know it got faster; there's no measure both sides would sign

Internal roles have no natural counter the way order volume does. “Expenses went from next week to same day” is something the client says themselves — but it isn't a figure a third party can check, and converting it to a percentage is the step where it becomes an estimate.

02

So we publish none

The figures on our storefront case each carry the record they came from, which is exactly why that page is worth reading. Invent one here and the checkable ones lose their value alongside it. That trade isn't worth making.

03

The percentage isn't what you should be asking anyway

Ask this instead: how many times a week does this repeat, what does one mistake cost, and who is waiting on whom. You can answer all three from your own data, and you'll know more than any case study figure would tell you.

Three rules govern how we write these: no brand name and no industry, no figure without a checkable source, no back-office screenshots. This page is all three at once.

After delivery

They're the ones adding to it now

Delivery isn't a finish line, it's the system changing hands. When a role opens and needs help, they build the block themselves — no waiting on our schedule. How that layer works is on the service page.

A new role, and they configure its Agent

The platform is theirs. When a role opens and needs help, the trigger, rules, output and human checkpoint get picked in the interface. They do the day-to-day faster themselves; anything they're unsure about, we look at together.

A new one runs alongside first

It gives suggestions and touches nothing real until its judgement has settled. Look first, let go after — the people using it know where they stand, which is why it spread.

The prompts are theirs to edit

Delivered as a library per department, written alongside them on site, with a stretch of time afterwards. Wording and rule changes are now made by the departments themselves, the same day.

It sits on their own servers

Installed on the client's own hardware from day one, with the source and the data in their hands. When they want us to help they grant access and close it again afterwards — so it's always clear who did what, and when.

This is the third system built on a client's own servers. As with the other two: the source goes into their repository and their own people run it day to day; when they want us to help, access is granted and closed again afterwards.

When this doesn't apply

The boundary, stated

On the case page, rather than in a meeting.

Departments whose process still changes weekly Automating unsettled rules just produces the mess faster. We'll say leave it alone.

Anyone wanting to start with just one department Perfectly fine — it just sits closer to a single-module engagement, and scoping it as this line wouldn't be worth it. We'll say so.

Anyone expecting it to replace an ERP or people system It sits on top of them, reads from them and writes back. It doesn't replace them.

Anyone whose goal is headcount This hands over actions, not people. Used as a redundancy tool it gets quietly resisted into failure.

Questions

What you're likely to ask

Answered directly.
Were all seven departments done at once?+

No — and doing it that way is how this fails. The first one is whichever has the highest volume and the most reversible mistakes; the rest follow once it works. That order isn't a technical preference, it's about momentum: the run log from the first department is the only material that convinces the second, and it works better than any proposal document. The foundation is built once, so later departments are lighter. How much lighter we won't put a number on — it depends on the state of that department's data.

Do people in the stores actually use it?+

That's the largest delivery risk in this kind of project, far larger than anything technical. Two things are non-negotiable: the entry point sits in the tool they already use — no app, no new password — and what ships first is whatever saves them time (asking about a product, generating a count list), with reporting and oversight coming later. We've seen the result of the opposite order: however good the system, the floor treats it as something built to measure them, and quietly stops using it.

Isn't the listing and imagery part the same as your other cases?+

The method is the same, which is why it isn't explained twice here — in this client's flow it's two lines inside the e-commerce card. For how that part actually works and where its limits are, the visual pipeline service and the content pipeline case both go into far more detail than this page would.

After delivery, do they come back to you to add things?+

Not for the everyday things: the platform and the source are theirs, and a new role Agent is configured in the interface — trigger, rules, output, checkpoint. What we would sit down about is the other kind — connecting an external system nobody has connected before, or a capability the platform doesn't have yet. Scope and pricing for that are agreed before anything starts, rather than raised afterwards.

Next

Find the two departments your facts break between

The starting point is the same as everywhere else: how often does this repeat, what does one mistake cost, who is waiting on whom. The diagnostic fee is credited in full against the work that follows.