A city considering a sensor deployment now gets a competing pitch: skip the hardware, buy the data layer. The providers are credible and the coverage numbers are large. INRIX reports 48 million parking spaces across 22,000 cities in 145 countries, drawing on data from 300 million connected vehicles, with availability predicted at estimated time of arrival and demand forecasts extending seven days out. Parkopedia publishes benchmarking reports comparing its parking data quality against Google and INRIX. Both sell satellite and positioning-derived layers alongside their sensor and operator feeds.
At the far end of the same technology, PayBySky’s Skymeter — an in-vehicle device relying on real-time connectivity and satellite geo-positioning to detect the start and end of a parking session — has been running autonomous parking payment since its launch with Calgary’s ParkPlus system, billing by the minute at posted rates on a fixed monthly subscription.
These are genuinely useful products. They are also answering a different question than a sensor deployment answers, and the procurement mistake is treating them as substitutes on a spreadsheet.
What positioning data is actually good at
The underlying signal in a connected-vehicle data layer is a vehicle that stopped, stayed stopped, and then moved. From a large enough fleet of probe vehicles you can infer a great deal from that: which blocks fill and when, how long the typical stay is, how demand shifts by day of week and by event, and — with a model on top — what availability is likely to be when a driver arrives twelve minutes from now.
That inference is strong for the questions cities and drivers most often ask.
Demand patterns over time. Excellent. Probe data across months produces a better picture of temporal demand than a sparse sensor deployment does, because it covers the whole network rather than the instrumented subset.
Wayfinding and driver-facing guidance. Strong, and improving. A driver wants to know where to go, not the exact count. Predicted availability at arrival time is the right product for that question, and precision beyond “likely available” adds little to the driver’s decision.
Coverage economics. Unmatched. Nothing else covers 22,000 cities. A sensor network covers the blocks you paid to instrument, and instrumenting an entire city’s kerb is not a budget any city has.
Validation practice. Worth noting as a differentiator. INRIX describes monthly bench testing against satellite imagery, municipal data, and APIs, plus quarterly real-world ground-truth validation — reporting validation across 46 city instances in 12 countries between January and April 2026, with roughly 99% coverage and accuracy in that ground-truth testing. That is more methodological transparency than this category usually offers, and a provider who cannot describe an equivalent process should be asked why.
What it does not do, and where that matters
The limitation is structural, not a matter of model quality: inference is not measurement, and several functions require measurement.
Enforcement. You cannot cite a vehicle based on a probability. Overstay enforcement, permit verification, and loading-zone abuse all require a ground-truth observation tied to a specific vehicle at a specific space. Probe-derived availability cannot produce that and does not claim to.
Per-space billing and dynamic pricing at space granularity. Charging a specific driver for a specific space requires knowing that the space was occupied. Block-level or zone-level inference supports zone pricing well and space-level pricing not at all.
The under-represented tail. Probe penetration is not uniform. Connected-vehicle data skews toward newer vehicles and toward markets where those vehicles concentrate. A block where the parking population is older vehicles, commercial vehicles, or fleets that are not in the probe pool is a block where the inference is weaker — and it is frequently the block where the policy question is hardest.
Off-street interior detail. Probe positioning degrades in structures. Multi-level garage occupancy by floor, the classic wayfinding application, remains a sensor or a gate-count problem.
Accountability at the space. A city that reports occupancy to residents, or that ties a policy commitment to a measured outcome, needs a number it can defend under challenge. Ground truth has a different standing than a model output when the question is contested.
The layered architecture most cities land on
The framing that works is not sensors-versus-data. It is: buy inference broadly, measure narrowly, and use the measurement to check the inference.
Buy the data layer for the whole city. It is the only affordable way to have any visibility outside the instrumented core, and it is the right product for driver-facing guidance everywhere.
Instrument where a decision depends on the number. Blocks under active dynamic pricing. Loading zones where enforcement policy is contested. Corridors where a pilot’s outcome will be reported. Those are small fractions of a city’s kerb, and they are where measurement earns its cost.
Use the instrumented blocks as a permanent validation set. This is the step most cities skip and the one that makes the whole architecture defensible. A few hundred sensed spaces give you a continuous read on whether the purchased data layer is accurate in your city, on your streets, with your vehicle mix — rather than in the provider’s validation cities. When the inference drifts, you find out.
The procurement questions that separate the products
Ask what fraction of the coverage in your specific city is sensed, operator-fed, or inferred — the global number tells you nothing about your market. Ask for the accuracy figure broken out by those categories rather than blended. Ask what the ground-truth validation method is and when it was last run in your region. Ask how the model behaves during an event, a closure, or construction, which is when guidance matters most and historical patterns help least.
And ask what happens to your data if you also supply into the platform. An operator or city feeding occupancy into a provider’s layer is improving a product that is sold to others, and whether that is a fair trade depends on terms that are rarely in the first draft of the agreement.
