SamePal
Margin erosion
Everyone gets the same cut. High-intent shoppers who would have paid full price are subsidized. Low-intent shoppers leave when the sticker comes off.
Retail neural network · store-local
Stop buying loyalty with mass discounts.
Retailers spend billions on blanket promotions that thin margins and teach shoppers to wait for the next sale. SamePal turns ten visits into a habit — with a personal offer created for that shopper, in that store.
~10 visitsto a shopping habit
15 minutespersonal offer window
0 photosstored. Vectors only.
Entrance agent
Anonymous vector captured
Dummy profile D-4821
No RGB kept
Structure only — no photo file
Nothing to search as a face
The expensive reflex
Mass discounts do not create customers. They rent them for a weekend.
A yellow sticker trains one behavior: wait. It does not remember that this household always buys oats and tomatoes, arrives after work on Thursdays, and will add olive oil if the offer is theirs — not the store’s weekly broadcast.
No memory
Classic promotions are blind to repeat baskets. They cannot tell a first-time passerby from a neighbor who shops here twice a week.
Privacy backlash
Shoppers and regulators are wary of biometric dossiers and hidden “surveillance pricing.” SamePal is built to stay on the other side of that line.
Three agents, one loop
Entrance, tablet, and checkout learn together.
SamePal is a neural network for a single retail space. Three stations watch movement and baskets — not identities — and only start personalizing after a shopper voluntarily interacts with an in-store tablet.
01 · Entrance
A vector, not a portrait
ToF and camera extract anonymous structure. No video or photo is stored. The vector is slotted into a dummy profile. Staff wearing tags are excluded from shopper counts.
02 · Tablet
Consent unlocks the offer
Legal access to a working profile begins when the shopper chooses to interact. Then they receive a 15-minute offer built from what people like them actually buy in this store.
03 · Checkout
The basket trains the store
Purchases attach to the dummy profile. Reinforcement learning improves the next offer — for that shopper and for lookalike local groups. No global data broker required.
Interactive walkthrough
A shopper’s first ten visits.
Advance the visits. Consent appears only at the tablet. The 15-minute offer stays locked until the shopper opts in.
Store memory
Entrance records an anonymous vector and opens a dummy profile.
Checkout attaches the first real basket. The profile is still unnamed.
Tablet is available. Offer stays dark until voluntary interaction.
Repeat items form a staple pattern. Lookalikes in this store start to cluster.
Offer timing can shift with store load — quieter hours, better space to convert.
Habit window: the shopper expects a relevant offer here, not a generic flyer.
SamePal · Aisle tablet
Store-local
Consent gate
Your offer is ready when you are.
We do not open a working commercial profile from a face in the doorway. Tap to attach this visit to the dummy profile already learning from baskets in this store.
Personal 15-minute window
Built from this store’s baskets, not a national file.
15:00
Redeem before the window closes. Timing can flex with queue load so the offer is useful, not noisy.
Oat drink 1L + granola−18%
Cherry tomatoes 500g−12%
Olive oil 750ml · add-on−15%
Designed against the dossier
Anonymous vectors. Dummy profiles. Consent before commerce.
SamePal records anonymous vector data. It does not store video or photos. Over time, repeat visitors are grouped — not named — and offers stay inside the four walls of the store that earned the data.
No RGB archive
Entrance sensing keeps structure, not a picture. There is no searchable face gallery sitting on a server.
Dummy, not dossier
Profiles are placeholders. They hold vectors and baskets. A legal working profile starts only after tablet interaction.
This store, these groups
No market-research panel. No global broker. Correlation happens among local dummy profiles that shop here.
A discount with a clock
SamePal replaces generalized sales with a short, individual offer — not a hidden higher price for the person next in line.
Sensing stack
ToF at the door. Cameras that count. A dashboard that stays operational.
Identification for offers is vector-based. Camera analytics sit beside that loop for store operations — footfall, dwell, conversion, staffing — without turning the aisle into a biometric file.
In and out, cleanly
- Entry / exit at 98% accuracy
- Staff elimination via wearable tags
- Centralized operational dashboard
- Depth structure for anonymous vectors
Three stations in the supermarket
- Entrance, tablet, and cash register
- Unique and group visitor counting
- Open / close and hourly lean vs peak
- Dwell and conversion via two-camera or API setup
- Optional aggregate age / gender footfall
- Employee attendance when entry and exit are covered
What the retailer sees
Upside without another storewide markdown.
Each accepted offer trains the local model. Similar dummy profiles get sharper predictions. Offers can be timed to floor load so the store sells more without flooding an already busy hour.
98%In / out accuracy
15mOffer window
LocalTraining data only
3Synced agents
Hourly load · offer timing prefers lean slots
Retailer outcomes
- Protect margin: stop giving the same cut to every cart
- Create upsell from real baskets, not guesswork
- Time offers to capacity — resilience on peak days
- Build a habit instead of a coupon reflex
- Keep learning inside the store that paid for the sensors
Straight answers
What this is — and is not.
Is this facial recognition?
No gallery of faces is kept. The system stores anonymous vectors and dummy profiles. A commercial profile is not activated from a walk-in alone.
Is this hidden personalized pricing?
The product is a timed personal offer in place of a generalized sale. The shopper sees it on a screen they chose to use.
Where does the data live?
In the store’s loop. Dummy profiles are correlated with other dummy profiles from that location so local groups get better offers over time.
When does learning start?
Baskets can train an anonymous profile from the first checkout. Personalized offers require voluntary tablet interaction.
Pilot a floor
SamePal is the path from an impulse discount to a long-term habit of shopping.
Tell us the store format and we will walk the three-agent loop against your entrance, tablet point, and checkout.