Telematics is often the last line item a fleet buyer negotiates and the first one that becomes a three-year problem. Whether the connectivity module sits on the controller PCB or arrives as a bolted-on box decides four things at once: what the bill of materials costs per vehicle, what the operator pays per vehicle per month in data and platform fees, who owns the resulting data, and who carries the risk when a remote immobilise command is wrong. Because those four items feed directly into total cost of ownership and into what the fleet is worth at resale, the choice belongs in the RFQ and the technical specification, not in a retrofit project scheduled after delivery.

Architecturally the two options are not interchangeable. An integrated module draws power from the controller and sits on the same CAN or UART bus that already carries battery state of charge, state of health, controller temperature and fault codes, so the platform can report vehicle condition rather than position alone. A retrofit terminal usually taps the harness or the main battery, reads GPS and accelerometer data, and stops there: it knows where the bike is but not whether the battery is degrading, whether the controller has logged a recurring fault, or whether the rider is in a limp-home mode. That gap shows up first in maintenance planning and warranty discussions, where vehicle-state history is the evidence.

A fair cost comparison needs the same three lines for both options. The first is the hardware delta per vehicle, which for an integrated module is a BOM line and for a retrofit is a purchase order plus fitting labour. The second is connectivity: an eSIM inside the module versus a SIM and terminal supplied by a third party, with tariffs that scale by fleet size and reporting frequency. The third, and the one most often left out of the spreadsheet, is the platform subscription per vehicle per month. Extend all three across the intended holding period, typically three years for a leased fleet, and the option that looked cheaper at delivery often loses, because a retrofit saves BOM but rarely saves subscription.

Supply chain questions follow the same logic. An integrated module is tied to the controller generation: when the module or its radio variant changes, the controller part number changes with it, carrier certification has to be re-confirmed, and spare controllers bought two years earlier may no longer match the current platform. A retrofit terminal decouples that cycle and its spares are common across brands, which is exactly why mixed-brand fleets tend to start there. The compatibility boundary is real, though: one retrofit terminal reading proprietary CAN dialects across several controller brands will expose only the signals all of them share, and the deepest diagnostics stay locked to whichever brand the supplier knows best.

Lock control is where the specification gets physical. An electronic wheel lock, a frame lock, an alarm and a remote immobilise command are four different mechanisms with different failure modes, and only some of them can be actuated without the controller being awake. Any design that allows a vehicle to be stopped remotely also has to answer what happens when the module loses power or the network drops mid-command. The mechanical key or override must remain the final authority, and the bike must never be left in a state the rider cannot resolve at the roadside. Ask suppliers to describe that fallback explicitly, then test it by pulling the module fuse.

Remote disable carries legal weight that a location feed does not. If a finance company, a shared operator or a fleet manager can stop a vehicle from a dashboard, the user agreement must state the trigger conditions, the notice given and the appeal route, and the purchase contract should allocate responsibility between the telematics vendor, the platform operator and the fleet owner. Insurance conversations go more smoothly when the platform can produce a defensible event record rather than a position ping, so specify the log fields before signing and confirm they survive an offline period.

Minimum anti-theft event log: timestamp in UTC, position and accuracy, lock state before and after, shock or movement alert, geofence entry or exit, command source, command result, and the network state at the time of the event.

Data ownership is a contract question, not a product feature. Three categories need to be named separately: raw telematics records generated by the vehicle, aggregated or derived data the platform builds from them, and personal data tied to an identified rider. Buyers should secure the right to export raw records in a usable format at any time and, critically, on termination of the service, including a defined window after the supplier stops supporting the hardware. Without an export clause the fleet history is effectively held by the platform, and changing supplier means rebuilding the maintenance and warranty record from zero.

On the interface side, REST for queries and MQTT for streaming are the common baseline, with webhooks for event-driven alerts; shared operators often expect GBFS-format availability and status feeds on top of that. The specification should fix the transport, the payload schema, the push frequency, rate limits and the SLA attached to each endpoint, because a generous API in the demo becomes a throttled one under load. Data protection adds a second filter: where rider-linked data is processed, GDPR obligations and the legal basis for cross-border transfer decide which hosting regions are acceptable, and retrofitting a compliant data flow onto a closed platform costs far more than choosing one.

Acceptance testing should be written before the first unit ships. Measure position accuracy and cold-start time in the open, then repeat in an underground garage and a dense urban canyon; confirm the declared IP rating and operating temperature range against the actual installation point; and force an OTA update to fail deliberately to see whether rollback works. Fix the degradation strategy in the SLA as well: how long the unit buffers data offline, whether locking and riding still work without connectivity, the fault response time and the spare replacement cycle. Then run 20 to 50 vehicles for four to eight weeks and compare measured uptime, false alarm rate and the real data bill against the supplier's promises before committing the rest of the fleet. That pilot is also the cheapest place to discover that an interface, a protocol and a fallback were never specified.