Project Forecourt PWP: Agentic Offers at the Fuel Pump
While a driver fuels, the pump screen offers one store item at a purchase-with-purchase price, and the fuel transaction is left exactly as it is.
A proof of concept for a national fuel retailer: agentic AI on the existing outdoor payment terminals that invites a convenience-store purchase while the customer fuels. A driver who fuels above a threshold can add a coffee or a snack at a set price, and a voucher prints with the receipt for the counter to fulfil. Behind it sit a back office with maker-checker approval, a versioned decision table the pump reads, station controls with a kill switch, and an uplift dashboard that compares pilot stations with matched controls.
Most drivers pay at the pump and never enter the store. The retailer wanted to turn some of those visits into store sales without slowing the fuel transaction, touching card or PIN screens, or depending on a network link that is weak at some stations.
Agents plan offers, assemble campaigns and analyse results away from the payment and real-time paths. Each campaign passes 14 compliance checks, including usual-price rules, promotion time limits, a tobacco block and halal status, before a second person approves it. A site edge node on the station network keeps the offer running through a network outage, and a share of eligible sales is held back so the uplift can be measured honestly.
The proposal sets out a 24-week proof of concept in four gates, from a lab to one pump and then six stations against matched controls. An interactive mock-up runs the whole journey: the pump screen on both sizes, the store counter queue, the back office, vouchers and a hash-chained audit trail.
- Status
- Proof of concept
- Client
- Confidential
Built to the Governed Cognitive Layer: the model reads and drafts, deterministic checks decide what it may do, and a named person approves anything that commits the business.
Put the AI where it plans and checks, and keep it off the payment path. Here the agents design offers and read the results; the pump only reads a table a person has approved. That lets you try AI on a live customer journey without putting a single transaction at risk.
Proposal, C-suite deck and interactive mock-up
The proposal, the deck and the working mock-up sit behind a password, with the client's brand and real station conditions shown as they will be in the pilot.
●Live demo (on request) →The same design rules behind this build, applied to your business in half a day.
Book the WorkshopDiscuss a build like thisOther projects
Enterprise AIKELIX Enterprise AI Platform
Proprietary private cloud AI engine built for 6,000+ engineers on multi-billion dollar projects.
Read case study →
Enterprise AIMeinhardt AI Centre of Excellence
RAG platform deployed across 6,000+ engineers.
Read case study →
GovernanceGlobal AI Governance Framework
Governance standards for AI deployments across Asia and the Middle East.
Read case study →
The first version of the pump screen showed the offer at any moment in the sale. Walking through the journey as a driver changed that. An offer must never appear on a card or PIN screen, a held-back sale must see no teaser at all, and after a driver taps No thanks the screen must not ask again. The last chance comes once, when the nozzle goes back. Respecting the driver's attention matters more to the result than any model choice.
