Omnichannel clienteling
You paid to win this customer once. Next quarter you'll pay to win them again.
Stitchwork joins your website, your own stores and the counters you run in other people's buildings around one record of each customer — so the person who bought in March isn't a stranger in September, and the second purchase gets asked for instead of waited on. One customer, one record, every counter you trade from.
For omnichannel brands — online, your own floors, concessions and kiosks · orders worth a conversation · 1–15 locations
This is the customer every brand says it wants. Watch a year of them go unrecognised.
Three weeks of looking, then the first order: the reading chair, £1,480. A name, an email, a delivery address. This is the person the whole funnel was built to find.
They come in to choose the fabric in daylight, and leave with the footstool as well. The till opens a fresh record: new customer, first purchase.
The counter you staff in someone else's building. They spot the lamp on a lunch break and buy it. That receipt lives in the host's tills — as far as your systems know, it never happened.
Twenty minutes on the new collection, on their phone, sitting in the chair you sold them. The site meets them like a stranger. So does everything you send.
"We miss you — here's 10% off your first order." Their third purchase was five months ago. Nobody knew it was their third.
One person, quietly becoming a regular — on systems that can't see it happening.
One customer became a regular this year. Nobody in your business got to notice — so nobody got to ask for the next one.
Repeat revenue isn't a marketing problem. It's a memory problem.
Every plan you've written says the growth is in the second and third purchase, and the economics agree: the first sale pays for the finding of them, and every sale after that should be bought with memory instead of media. But a second purchase has to be asked for — by an associate who knows what this person already owns, by a site that leans towards what they're circling, by a call the week they go quiet. And nobody can ask for what none of your systems can see.
The wall runs both ways, and your team feels it before you do. The store holds onto customers, because a sale that closes online counts against the floor. The website team optimises a checkout your best customers only used once. The concession is a monthly statement from the host, not a place your regulars go. Everyone is defending their half of a customer who thinks of you as one shop.
A brand is a memory the customer assumes is mutual. They remember the chair. They expect whoever sells them the lamp to remember it too.
This is what Stitchwork is for. Not another channel and not a campaign tool — the memory underneath all of it, so your site, your floor and your counters stop meeting your regulars as strangers. The clienteling screens come with it. The record they run on is the point.
A clienteling app assumes the record already exists. Stitchwork is where the record comes from.
Point a repeat-purchase problem at the market and it answers with clienteling — Tulip, Endear, Salesforce Retail. The demos are slick. Ask one question in them: where does everything on that screen come from?
An associate opens the app and knows the customer — what they own, what they browsed, what to offer next. The right idea. But the app is a window, and the record behind it — web browsing joined to store purchases joined to what the concession sold — is assumed. You bring your own data. That's the integration project, and the integrator that comes with it, before the first associate ever sees a real customer in it.
- The window without the record shows a stranger. Clienteling stands on an omnichannel record: one customer recognised across the website and every till. If your data can't say that today, the app can't either — it displays whatever you manage to pipe into it, and the piping is the project.
- Stitchwork starts one layer down. One script tag on your site. Last year's orders in by CSV, tomorrow's straight from your checkout. Every store, kiosk and counter set up as a channel in an afternoon. The record assembles itself from there — and the screens your team sells from stand on top of it, already filled in.
- The proof is a week on your own trade, not a pitch. Seven days, everything switched on, no card. Your catalogue in on day one, last year's orders on day two, the tag live on day three — and the record starts assembling from customers you already have, before anyone has signed anything.
One customer, assembled from every place they meet you. Without the data project.
The site watches, the counters report, and identity does the joining — the same person recognised across all of it, with everything they did before you knew their name attached afterwards.
Browsing that remembers who it belongs to.
One script tag, behind the consent banner you already run. Stitchwork sees what visitors view, add and search — and the moment a browser identifies, the whole anonymous trail joins their record, retroactively. The regular on their phone at midnight stops being a stranger with a cookie.
- Views, cart adds and searches, joined to the customer the moment they identify
- Identify at checkout — or from your newsletter, live chat or a download, no order needed
- Recommendations and category grids re-ranked for the person looking
- Interest signals your email stack can read, to pick the banner or the segment
Stores, kiosks and concessions that report to one memory.
Every place you trade is a channel — the flagship, the pop-up, the counter in someone else's building — each with its own staff, country and currency. A code on the shelf edge, printed as a QR or written to an NFC tag, puts the person standing at it onto their own record. The till's orders land there too.
- Channels typed and owned your way: owned, franchise, authorised, wholesale, marketplace
- Scan or tap, and the next half hour of browsing belongs to that counter
- Orders in by CSV or from your own back office — idempotent, safe to re-import
- Store visits on the customer's timeline, in sequence with their browsing
- A Trade view per counter — revenue, orders and average order, per currency, and what sells there
Six sources, one stream. Nothing collapsed into a tally.
Web browsing, campaign arrivals, store visits, quotes, orders and the moment an anonymous trail was claimed — one timeline per customer, newest first. Six views over five weeks and six views in four minutes are different customers, so the sequence is kept, not counted. And the thread survives housekeeping: delete the quote, the customer, even the channel, and the link doesn't break — because the alternative is a customer you have to meet twice.
Who's due back — and whose Tuesday it lands on.
Knowing is half. The other half is a short list with a name against each row and a day it's owed by — and proof at the end of the week that it got worked.
Ask the AI you already use — Claude, or any MCP agent, connected by pasting one URL — who's worth a call this morning, and it answers from your own records, not its imagination: the basket abandoned on Thursday, the person circling the same piece for three weeks, the quote still open — and the sharpest cut there is, the past buyer who went quiet in spring and started browsing again this week. These aren't strangers from a form; they're your own regulars, mid-decision. Ask for the brief before the call and it answers with what they're worth, what they're circling, what they already own down to the actual items, and the last few things your colleagues wrote about them. Then it puts each one on a colleague's list with the reason attached — and a draft of the opening message, for the colleague to make their own and send from their own tools — and the rules below take over.
- Having bought is a reason to call, not a disqualification. A repeat buyer circling something new makes the high-intent list — every signal is counted on one window, so buying from you two years ago doesn't file someone under finished. The second purchase is the point of this list, not an edge case on it.
- Nobody opens with "you didn't finish your order" to someone who bought on Tuesday. Before anyone is proposed for a call, Stitchwork checks what they've bought since — in the room, through the concession, under different SKUs. Answered baskets come off the list, and the ones that survive say what the customer did buy instead.
- The work lands with whoever they've been dealing with. Read off your own records — the associate on their open quote, at the store it was raised — never guessed, and always with the sentence explaining why. A store is sticky; a person who leaves or moves branch doesn't take your regulars with them.
- A follow-up the customer answers retires itself. Buy before anybody rings, and the row comes off the list the moment the order lands — marked apart, with nobody paid for a call that never happened. It's held to the real figure: a £30 cushion doesn't settle a £4,200 sofa.
- Closing one means writing what happened. So Monday's review reads as a record of the week — raised, done, dismissed, settled, still overdue — not a count of pressed buttons. And when the sale lands, the credit splits across the people who earned it, frozen at the sale, with a manager able to overrule it on the record.
Your site can say something to a returning customer. And it holds people back, to tell you whether it worked.
One line across the top of your own site, shown to a visitor whose browsing matched a rule you wrote — a returning customer three visits into the new collection, say. Nothing is emailed and nothing is queued: the message exists only in the session it appears in.
Any tool can show a banner to interested people and take the credit when they buy — interested people were always the likeliest buyers. So every nudge holds a share of its matches back and shows them nothing at all. The difference between the two groups is the only honest number in this category, and it's the one we lead with.
- The report is allowed to say it did nothing. If the shown group doesn't beat the held-back group, it says so — and until there are thirty visitors in each arm it says "still counting" instead of quoting a percentage nobody should act on.
- A phone number becomes somebody's Tuesday morning. Ask for a callback and it lands as a follow-up on a colleague's list, with the reason attached. Ask for an email, and three weeks of anonymous browsing attaches to a customer record the moment they type it.
- The thresholds come from your own buyers. Nobody knows how many product views counts as interested, so it doesn't ask you to guess: it reads what customers who actually bought had done first, and fills the numbers in. Under 25 orders to learn from, it says so and uses defaults.
- It rides the tag you already installed. No second script, no popup builder. It waits for the same consent an event does, comes down if consent is withdrawn, and sits in its own shadow root so your CSS can't break it and it can't break yours.
- Your AI can draft one. Only a person can run it. An agent connected to the workspace can propose a nudge — it arrives as a draft, named as the machine's, and cannot appear on your site until somebody reviews it and turns it on. Then the holdout referees the AI's copy exactly as it would yours. Agent proposes, person approves, the control group keeps everyone honest.
Where the number actually moves.
An order value you steer.
Your catalogue is ranked by signals you weight yourself, and one of them is a target average order value. Set the number your business runs on, and every surface — the site, the grids, the list in your associate's hand — leans towards it. A dial, not a hope, and never a black box: the Rankings page shows each product's score, signal by signal.
A site that leans towards the regular.
Every customer gets a personal overlay on the ranking — what they've viewed, carted and been quoted — blended with the catalogue's own score by a dial you set. And the chair they just bought stops following them around: a purchase settles the interest that led to it, so the suggestions move on to whatever comes next. The returning customer lands on a shop that's quietly about them.
Quotes that close on a phone.
The considered basket doesn't die in a cart. Your team builds the quote in the room or after the visit, sends it as a link, and watches it get opened. The customer accepts and pays a deposit into your own Stripe account — or clicks through into the checkout you already run.
A floor with no reason to hoard.
When the conversation opens in store and closes online at midnight, the associate who opened it is named on the sale — and credit splits across opener and closer, frozen when it lands. The store stops fighting the website for customers they share.
The ones a demo glosses over.
The counter, the site and the store — answering to one record.
The whole thing is written down. Send it to whoever runs your systems.
The script tag and the consent behaviour. How an anonymous browser becomes a named customer. How orders get in from a CSV or your own checkout, and how channels, codes and currencies work. Every endpoint, every webhook payload, every failure mode that actually happens — written for the person who has to make it work on a Tuesday afternoon, not the person signing the contract.
Read the documentationNo form, no email gate. Twenty-two pages of plain HTML you can read on the train.
We'll show you the parts that work and tell you which ones aren't built yet — and you'll notice this page quoted you no lift figures, because we don't have a merchant's numbers to quote. We'd rather lose the deal than the second meeting.
The second purchase is the cheapest revenue you'll ever book. It deserves better than luck.
Give us thirty minutes and one real customer from your own books — somebody who bought twice. We'll walk what your business knew about them at each moment across your site, your floor and your counters, and show you exactly where the thread breaks today. You'll leave with the map whether or not you ever use us.
By hand, not by checkout. There's no plan catalogue and no public price, because there's nothing here you'd buy without talking to someone first.
Seven days free with every feature switched on. No card, no reduced tier, no feature gates. Long enough to get your catalogue and a year of orders in, and watch one associate use it.
Terms agreed per client — flat, per seat, or both — in your currency, on your interval. Set once, written down, and not moved without a conversation.