How GPS-Verified Ad Delivery Works

How GPS works on a consumer phone

Every modern smartphone runs a multi-constellation GNSS receiver (GPS, GLONASS, Galileo, BeiDou) paired with a sensor fusion stack: accelerometer, gyroscope, magnetometer, barometer. The receiver locks onto satellite signals, computes pseudoranges, and resolves a 3-D position fix.

On its own, raw satellite acquisition is slow. A cold start takes minutes to first fix. Smartphones short-circuit this with Assisted GPS (A-GPS): the cellular network downloads ephemeris and almanac data to the phone over the radio link, so the receiver knows where to look in the sky and locks in seconds rather than minutes.

A-GPS plus Wi-Fi BSSID lookup is what gets a phone to its quoted accuracy floor. The figure to anchor on comes from the people who run the constellation: GPS.gov states that GPS-enabled smartphones are typically accurate to within a 4.9 metre radius under open sky, and that accuracy worsens near buildings, bridges and trees. That last clause is the one that matters for advertising geometry, and the limitations section below takes it seriously.

Smartphone under open sky

4.9 m

GPS.gov: smartphones are 'typically accurate to within a 4.9 m (16 ft.) radius under open sky'

GPS.gov (US government), GPS Accuracy

Government commitment on the signal itself

2.0 m

Daily global average user range error, the error budget before any receiver or environment effect

GPS.gov (US government), GPS Accuracy

Urban canyon (peer-reviewed)

7-13 m

Measured iPhone horizontal accuracy in a dense city grid

Smartphone GPS accuracy in urban environment, PLOS One

Operator-owned mesh vs. bid-stream-inferred location

There are two ways an ad platform can know where a phone is. They produce wildly different accuracy and economics.

Bid-stream-inferred location. The ad platform never touches the device. It listens to OpenRTB bid requests flowing through ad exchanges. Each request may carry a coarse lat/lon attached by the publisher's SDK, plus a mobile advertising ID. The platform decides whether the device is “near” a billboard or geofence and bids accordingly. The whole auction has to clear before the page or app finishes loading, which leaves the bidder a fraction of a second to decide, so the location call is necessarily a fast inference rather than a measurement.

Operator-owned mesh. WilDi Maps runs its own application on phones of people while they're driving in the cities it serves. The phone reports its GPS position directly to WilDi infrastructure: no exchange, no bidder, no SDK rake. When a person driving enters a corridor the advertiser owns, the operator's server selects the matching ad for that corridor and serves it to the operator-owned device. The location signal originates inside the operator's stack instead of being purchased downstream of three intermediaries.

The verification step: GPS to server-confirmed delivery

“Verified” is a load-bearing word in the model. It means the delivery record on the server is only written when the device's reported GPS coordinates fall inside the corridor geometry the advertiser owns at the moment the ad is rendered. The sequence is mechanical:

Stated in full: a WilDi verified delivery is not an estimated impression. It is an auditable app event in which an opted-in participant receives the advertiser's offer at the required location and explicitly acknowledges it. That claim is deliberately narrow. Most advertising currencies charge for modeled exposure, qualified impressions, or inferred attention, and several of them measure well: Geopath adjusts circulation by a Visibility Adjustment Index built from eye-tracking research, the UK's Route publishes a Likelihood to See figure, Place Exchange PerView establishes device presence deterministically, GroundTruth ties exposure to a later visit, and Quividi reads head direction on camera-equipped screens. Out-of-home has had an accreditation-grade standard since the MRC finalized its Out-of-Home Measurement Standards on December 4, 2025. What separates this model is the unit, an explicit acknowledgment by the recipient, and an acknowledgment proves an intentional human interaction rather than cognitive attention.

1. The phone reports its position while driving. The operator's app reads the OS-level location service (Core Location on iOS, FusedLocationProvider on Android) at a configured cadence. The fix is A-GPS-assisted with Wi-Fi BSSID and inertial sensor fusion.
2. Corridor membership test. The server runs a point-in-polygon test against the corridor geometry. If the point is inside the active corridor, the ad slot is eligible.
3. Ad selection. The advertiser who owns that corridor at that hour wins the slot. There is no auction. Corridor rights are fixed-rate, time-boxed, and exclusive.
4. Render and confirm. The ad is rendered on the screen of the person driving. The server writes a delivery record stamped with the device ID, the GPS fix that authorized it, the corridor ID, and the timestamp.
5. Bill the advertiser. The per-delivery rate is debited from the advertiser's prepaid balance. Only confirmed deliveries bill. No fix in the polygon, no charge.

Privacy and data handling

The privacy posture is structurally different from consumer SDK location collection because the device is operator-owned. The member opts in once when joining the network, the same way a courier opts into a logistics platform, and the GPS stream is used to fulfill the operator's contract, not sold downstream to data brokers.

On the advertiser side, the only signals leaving WilDi infrastructure are aggregate delivery counts and (if the advertiser opts in) attribution joins via the lead-capture pipeline. Individual member coordinates do not leave the operator's stack, are not joined to consumer identifiers, and are not exposed in the OpenRTB ecosystem.

On the consumer side: there is no consumer side. Members are network participants, not targeted users. The model does not require Apple App Tracking Transparency consent for cross-app tracking, because there is no cross-app tracking. The ad is shown on the operator's own surface to an operator-contracted device. iOS ATT governs IDFA access for cross-property advertising; that question doesn't arise here.

For OS-level context, see Apple's App Tracking Transparency documentation and Android's location permissions developer guide. Both frameworks distinguish between app-owned device usage (what WilDi does) and cross-property tracking for advertising (what bid-stream targeting does).

Limitations: urban canyon, indoors, edge cases

The model inherits the same physics every other GPS system inherits. Pretending otherwise would be dishonest.

• Urban canyon multipath. In dense downtown grids, satellite signals reflect off glass towers before reaching the phone. GPS.gov says as much in one line, that accuracy worsens near buildings, bridges and trees, and peer-reviewed iPhone measurements published in PLOS One put the horizontal error at 7 to 13 metres in a dense city grid, against 4.9 metres under open sky. WilDi corridors are sized to absorb that: a corridor isn't a 5-metre line, it's a street-aligned polygon.
• Indoor delivery. GPS effectively does not work indoors. People parked in covered garages, tunnels, or warehouse loading docks will not produce in-corridor fixes during those windows. The model does not pretend to deliver impressions to phones whose GPS is dead.
• Cold start. First fix after a phone reboot or extended power-off can take seconds; A-GPS shortens this dramatically but does not eliminate it. Deliveries during the cold-start window are not counted.
• Spoofing. A bad actor running a mock-location app could try to fake corridor membership. The operator runs server-side anomaly checks against speed, heading consistency, and inertial sensor data; flagged sessions don't bill.
• Coverage geography. A corridor only delivers when a contracted member is physically inside it. Off-peak hours and low-density geometries deliver fewer impressions per dollar of corridor ownership. Pricing is per delivery, not per hour, so the advertiser is never paying for empty corridors.

The CPVD pricing model

Cost Per Verified Delivery (CPVD) is the unit. A delivery is one ad rendered on one phone, from someone driving, with one server-confirmed GPS fix inside the corridor the advertiser owns at the moment of render. The rate is set by the account's monthly deposit tier rather than by an auction: $0.25 per background delivery at Enterprise ($3,000 a month and up), $0.32 at Pro ($1,000 and up), $0.40 at Local ($250 and up), $0.50 at Starter below that. Tunnels and zones cost more than background, so 'from $0.25' is a floor, not a flat price.

There is no CPM. There is no bid. There is no exchange fee, SSP fee, DSP fee, ad-server fee, or verification-vendor fee. The advertiser is billed the tier rate for the delivery and nothing is skimmed on the way, because there is no middleman stack between them.

For the full pillar treatment of how this compares to standard programmatic OOH, see Cost Per Verified Delivery. For why the model matters financially, see What is the Middleman Tax?