The card already knows the shop.
Nothing does the shopping.
Two million New Zealanders carry a card that records what they buy, when they buy it and what they nearly bought. It earns them points. It does not carry the bags, write the list, or notice that the thing they buy every week is not in the trolley.
This is a concept for joining the two: the loyalty programme you already run, and a personal shopper that acts on what it knows — and stops before it decides anything.
An independent concept. Not affiliated with, endorsed by or commissioned by Woolworths NZ or Everyday Rewards. Every figure on this page carries its source.
One of these is running. The other one isn’t.
Everyday Rewards is a knowledge engine that pays out in points. Personal shopping is an action engine with no knowledge behind it. Neither is missing technology. What is missing is the join.
What the card knows
- Every basket, going back years
- The shop that repeats, and the week it didn’t
- Which Boosts get activated and which get ignored
- Online, in-store, Scan&Go — the same person
- What a household looks like from its trolley
What nothing does with it
- Write the list before she opens the app
- Notice the usual thing missing from the basket
- Offer the swap she would have accepted anyway
- Fill the delivery window with something worth having
- Ask the one question a survey never gets an answer to
An agent that reads what the card already knows, prepares the shop, and then stops — holding every decision for the customer, and every send for a named person at Woolworths. Points stop being a rebate and start being the currency of a service.
One half of this can start Monday.
The other half does not exist yet.
There is no personalised-shopping API at Woolworths NZ to plug an agent into. We are not going to pretend otherwise, and we are not going to quietly design around it. It is the most interesting thing on this page.
The wait states and the rewards mechanics
The delivery window, the checkout queue, the Scan&Go loading screen, the moment at the terminal — these already happen, and Everyday Rewards already has the mechanics to pay for them: points, Boosts, Double Points Week. Filling a wait with something worth having needs read access to events your app already emits and rules you already publish. Nothing new has to be invented for this to run.
- Six weeks, one wait, one segment
- Read-only against existing events and points rules
- No new platform, no new contract, no write access
- Measurable inside the pilot window
Personalised shopping as an agentic layer
Everything in the live shop above — the list that writes itself, the missing weekly item, the substitution asked rather than decided — needs a personalised-shopping layer that nobody in New Zealand grocery has built. Not a feature bolted onto an app: a defined contract between what the card knows and what an agent is allowed to prepare from it. That is design and engineering work, done with you, not a switch we flip.
- A data contract: what an agent may read, and what it may never infer
- A preparation layer that drafts a basket and then stops
- An approval path, so a named person owns anything customer-facing
- The first of its kind in NZ grocery — which is the point
Tier one earns the right to build tier two. Six weeks on a wait moment proves the mechanics, the approval path and the appetite using only what exists — and produces the evidence that justifies the bigger build. If tier one fails its scorecard, tier two should not happen, and we will say so. If it works, you are not buying a personalised-shopping platform on faith; you are extending something you have already watched run.
One week, assembling.
Not a picture of a feature. Press play and watch the shop build itself from what the card knows — the list, the swap it asks about, the supplier slot it labels, and the points it earns on the way. Every screen here is illustrative and no real customer data was used. This is tier two — the part that would be built with you. It runs here because a concept you can press is worth more than a roadmap, not because an API exists to do it today.
The seven things this household buys every week are already on the list before the app is opened. Nothing was guessed — it is their own purchase history, read back to them.
The thing bought fifty weeks out of fifty-two, absent this week. Not upsold — flagged, because forgetting it means a second trip.
The swap a customer would probably accept is still a question, not a decision. Answer it and the answer is remembered. Ignore it and the shop still completes.
Funded by a supplier’s retail media budget, marked as funded, and relevant because of the basket rather than the highest bid. It never enters the trolley on its own.
The wait paid for itself: points for answering the one question, for the swap accepted, for the shop completed early. Loyalty stops being a rebate on money and becomes a return on attention.
Nothing is ordered, nothing is charged, nothing is sent. A prepared shop waits for one tap from the customer — and anything that would reach a customer waits for a named person at Woolworths.
Four things you could actually pilot in six weeks.
Each one lives inside a wait that already happens, uses a mechanic Everyday Rewards already has, and has a precedent somewhere in the world. None of them is a platform rebuild.
Every wait in a week of groceries.
Six moments a household waits through between the Sunday list and Wednesday dinner — online and in the aisle. Each one is somewhere a card that knows them could be doing something, and currently isn’t.
What this must never do.
Groceries are a weekly, trusted, household purchase. The failure mode is not a bad recommendation — it is a customer who stops believing the app is on their side. So the boundary is written before the feature.
It always
- Prepares, and then stops for the customer to decide
- Says when a placement is funded by a supplier
- Explains why something is on the list — “you buy this most weeks”
- Lets one tap remove anything, permanently, and remembers
- Holds anything reaching a customer for a named person at Woolworths
- Works for a household with no history yet, without pretending to know them
It never
- Adds anything to a trolley on its own
- Completes an order, takes a payment or books a slot
- Puts something in front of someone because a supplier paid the most
- Infers or acts on health, pregnancy, income or religion from a basket
- Uses a child’s presence in the data to sell anything
- Pretends to be a person, or hides that a machine prepared the list
What a pilot actually asks of you.
Written out in full, because the honest version is short and most of it is your time rather than your money.
The approver. Every draft stops with them and nothing reaches a customer until they say so. About two hours a week, reading drafts rather than managing a project.
Six weeks on the online delivery window for a single customer segment. Not the whole loyalty stack, and nothing that touches the trolley.
For tier one: read access to wait events your app already emits and the Everyday Rewards rules you already publish. No customer data leaves your environment and nothing writes to a points balance. Tier two needs no access at all in the pilot — the personalised-shopping layer is designed on paper first, against your rules, before a line of integration is written.
Your brand, legal and retail-media guidelines, so the drafts are held against yours instead of the placeholders on this page. An afternoon with whoever owns them.
Scored on drafts a CX lead ships without rewriting, whether the wait felt shorter to the people in it, and points issued that a named person approved. Fail any line and we change the design or stop.
No number is published on this page because it depends on which wait you pick. Whatever it is, it is agreed before any work starts and it does not move.
Run the pilot.
Say yes to the six weeks above and add anything you would change first. Most pilots move a line or two before they start, and the notes below are how that happens.
Opens your mail app to assembl@assembl.co.nz with your notes in it. Nothing is collected by this page.
- Six weeks, one wait, scored against a written line
- Read-only access, nothing that writes to a balance
- A named approver — nothing sends without them
- A fixed fee, agreed in writing before anything starts
- Stop any time. You keep what was drafted
US retailers lose an estimated $37.7 billion a year to customers who abandon because of long queues, split roughly $15.8b to competitors and $21.9b abandoned outright (Adyen with 451 Research). The New York State Department of Labor reported a 48% reduction in hang-ups after introducing virtual queuing. Theme parks moving to virtual queues reported a 36% lift in per-capita spending, because guests spent the wait somewhere that sold something.
Overseas figures are shown as overseas figures. We have not found a published New Zealand equivalent and we will not invent one.