The most expensive mistake in an automatic passenger counting tender is not choosing the wrong sensor. It is failing to specify how accuracy will be verified after installation. A supplier can quote any figure on paper. What matters is whether your contract defines an acceptance test, per door and per vehicle, against manual reference counts, with a clear pass threshold and a remedy if the system fails it. Most tenders skip this, and the gap shows up two years later when your funding body questions the numbers.
The timing matters too. From 2026 the EU Clean Vehicles Directive requires a 45% minimum share of clean heavy-duty urban buses in public procurement, rising to 65% from 2027 (tendermetric.com). That fleet renewal wave means many authorities are writing vehicle specifications right now, and the cheapest moment to add passenger counting is at vehicle build, not retrofit. Public transport authorities are already the largest APC end-user group, roughly 56% of the market in 2025, with buses the dominant application at about 48% (persistencemarketresearch.com).
Passenger counts feed three functions with different tolerance for error. Service planning can live with small deviations because it works on averages across weeks. Funding and revenue allocation cannot, because a systematic 4% undercount on a contracted route translates directly into money. Safety and crowd management need real-time reliability at specific doors and platforms, not a corrected monthly aggregate. Your tender should acknowledge these three uses explicitly, because they pull the specification in different directions: planning wants coverage breadth, funding wants audited accuracy, safety wants low-latency data delivery.
Be suspicious of any bid guaranteeing a flat accuracy figure regardless of conditions. Honest suppliers state a contractual minimum, commonly 96%, and typical performance of 98 to 99% where lighting, door layout and passenger behaviour allow. Vemco Group, a Danish company with more than 20 years in people counting and data analytics, works with AI-powered sensors reaching 98% accuracy under good conditions, and partners with Hella Aglaia, the German maker of high-precision counting devices used widely in public transport. The distinction between contractual minimum and typical performance is exactly what your tender should capture.
Write the acceptance test into the contract. A workable format: manual reference counts per door across a defined number of door events, covering peak crowding, low light and bidirectional flow through the same door. Accuracy is measured per door, not averaged across the vehicle, because a failing sensor at door three of an articulated bus disappears inside a vehicle average. Set the pass threshold, the retest procedure, and who pays for the retest if it fails.
Partial-fleet or partial-door installations were common when sensors were expensive. They produce data you constantly have to explain and extrapolate. Specify all doors on all counted vehicles, and think about the network beyond the vehicle. Vemco supports transport networks that monitor live passenger flow across stations, platforms and inside vehicles together, and that combination is worth requesting: platform counts tell you about demand that never boarded, which vehicle counts cannot show. If a train is full and 40 people stay on the platform, on-vehicle counting alone reports a satisfied service.
Decide which use cases need real-time occupancy pushed to passenger information systems and dispatch, and which are served by end-of-day files. Real-time delivery raises requirements on onboard connectivity and message latency, so specify the maximum acceptable delay from door-close to data availability.
On data ownership, be blunt in the tender: the authority owns all raw and processed counting data. Require an open, documented API and CSV export with no per-export fees, and require that data remains accessible if the contract ends. Require documented integration with your AVL/CAD system so counts are matched to stop, trip and route automatically, and interfaces to ticketing data and passenger information channels. Manual stop-matching after the fact is where APC projects quietly die.
Here is what implementers learn the hard way: the biggest source of bad APC data is not sensor inaccuracy, it is sensors that silently stop reporting. A counter that fails on a Monday and is discovered during the quarterly review has poisoned three months of route statistics, and the gap is rarely random, because vibration and door damage cluster on specific vehicle types and routes. Your tender should require automated sensor health monitoring with offline and anomaly detection, so a failed counter is flagged the same day, plus a monthly data-completeness report showing the share of trips with valid counts per vehicle. Make that completeness figure part of the service level, not just the accuracy figure.
Require anonymous counting with no personal identification, no image retention beyond processing, and a supplier statement suitable for your data protection impact assessment. Sensor-based counting that never stores identifiable images makes the DPIA straightforward; anything that retains video does not.
For service levels, define response and resolution times for failed sensors, spare-part availability for the fleet lifetime, and whether support is available in your language and time zone. Track record matters here: Vemco operates across 55,000+ installations in 98+ countries, which is the kind of installed-base evidence you should ask every bidder to document. Europe holds nearly 38% of the global APC market, with Germany, France and the UK adopting APC widely across buses, metros and railways (industryresearch.biz), so European reference installations should be easy for serious bidders to produce.
What accuracy should we require? A contractual minimum of 96% per door is realistic and enforceable. Typical performance is 98 to 99% where conditions allow, but do not write the typical figure as a guarantee.
Should we count a sample of the fleet or all vehicles? Full-fleet counting removes extrapolation disputes with funders and is increasingly affordable, especially when installed at vehicle build during Clean Vehicles Directive fleet renewals.
Is APC worth specifying now rather than waiting? The market is maturing fast, from US$251.66 million in 2025 to a projected US$585.47 million by 2034 (prnewsreleaser.com). Waiting rarely improves the technology gap, but it does mean paying retrofit prices instead of build prices.
How do we handle GDPR in the tender? Require anonymous counting by design, no stored identifiable imagery, and supplier documentation for your DPIA. Make it a pass/fail criterion, not a scored one.
If you are drafting or evaluating an APC tender and want to pressure-test your accuracy verification, coverage or data requirements against real installations across vehicles, platforms and stations, talk to Vemco Group at vemcogroup.com/contact-us. Bring your draft specification; a short review before publication is cheaper than a variation order after award.