Fleet Management Software: How It Is Built in 2026

Fleet management software is the system that turns what a fleet of vehicles reports about itself into decisions about where the vehicles are, how they are driven, what they cost and whether they are legal to operate. Its job is to make the stream of telematics mean something to a dispatcher, a mechanic, a safety manager and an auditor.

This guide is written from the builder's side. Every figure is sourced to the page it was read on, linked in the text or named under its table, and our own arithmetic is labelled; prices were read on 3 October 2026.

The short version

• Fleet management software is the layer above GPS tracking: it joins position with engine data, driver identity, fuel transactions and maintenance records, and it is the system of record a regulator inspects.
• The data comes from four places with different guarantees: an aftermarket device on the diagnostic port, the manufacturer's built-in modem, a certified electronic logging device, or a tachograph. Only the last two are specified by law, and only the aftermarket device, plus a few manufacturer feeds such as Ford Pro's, reports often enough for live operations.
• A vehicle reporting every second produces about 40,000 position records in an 11-hour driving day; at one position every 5 seconds the cloud cost of ingesting a vehicle is cents per month in our model, while the SIM card costs dollars, so sampling policy is a cost decision before it is a data decision.
• The compliance layer is concrete: the FMCSA specification tells you which engine signals to read, how precise the location may be, and how edits must be preserved; the EU smart tachograph records position every three hours of driving and applies to vans over 2.5 tonnes in international work from 1 July 2026.
• The measured evidence on safety is narrower than the marketing: coaching backed by video cut safety events by 38 to 52 percent in an FMCSA before-and-after study without a control group and roughly halved the odds of a risky event in a controlled NIOSH study, warning lights alone made no significant difference, and the ELD mandate itself did not reduce crashes for small carriers.
• Public plan prices run from about $2.50 to $55 per vehicle per month depending on whether maintenance, telematics or video is included, and about $65 to $67 when a camera add-on is stacked on a tracking plan; the enterprise vendors quote only, and their filings put contracts at three to five years (Samsara) and typically three (Motive).

What fleet management software is, and what it is not

Fleet management software is a platform that ingests vehicle telematics and driver activity, stores it as a history, and runs five kinds of workflow on top: tracking and dispatch, maintenance, fuel, driver safety and regulatory compliance. The line that matters is between tracking and management: a GPS tracker tells you where a vehicle is, while a fleet platform adds engine hours, odometer, fault codes, fuel level and who is driving, and turns the combination into a work order, an hours-of-service record, a fuel exception and a safety score.

Four neighbouring systems get confused with it, and the confusion costs money in procurement.

The transportation management system and last-mile delivery software guides cover the neighbouring layers and AI route optimization the routing engine; this article stays with the vehicle, the driver and the regulator, treating the fleet platform as one piece of the wider logistics management software stack.

The fleet it manages, in numbers

The population a fleet platform serves is large, old and overwhelmingly small-scale, and those three facts shape the product more than any feature list. In the United States the Federal Highway Administration counted 12,599,614 single-unit trucks and 3,482,896 combination trucks registered in 2024, 16.08 million vehicles that drove 329.6 billion miles; the average combination truck covered 55,276 miles and burned 7,771 gallons of fuel at 7.1 miles per gallon, and the average single-unit truck did 10,880 miles. The FMCSA's 2024 Pocket Guide counted 787,189 active interstate and hazardous-materials carriers at the end of 2023 with 5,421,013 power units between them. Of those, 418,526 ran a single truck, 53.2 percent by our arithmetic, and 90.2 percent ran ten or fewer; only 5,062, 0.6 percent, ran more than a hundred.

In Europe, ACEA's Vehicles on European Roads 2026 report counts 31.1 million vans, 6.16 million medium and heavy trucks and 699,238 buses in the EU in 2024, with the average truck 14.0 years old and 96.3 percent of trucks still diesel. Where a truck that old has no manufacturer modem, its data path has to be an aftermarket device. The UK Department for Transport licensed 4,951,370 light goods vehicles and 521,099 heavy goods vehicles at the end of June 2026.

The Census Bureau's 2021 Vehicle Inventory and Use Survey, an owner-reported national survey, found telematics "for fleet management" fitted to 872,500 of 8.9 million trucks other than pickups, minivans, light vans and SUVs, 9.8 percent by our calculation, and to 16.0 percent of Class 8 trucks. It did not ask about electronic logging devices, so it understates the share that transmit something.

The American Transportation Research Institute's 2026 operational cost update put the average marginal cost of running a truck in 2025 at $2.336 per mile, $1.854 excluding fuel, with tolls up 13.2 percent on the year. The Energy Information Administration's weekly on-highway diesel price was $6.382 a gallon in the week of 28 September 2026, $2.63 above a year earlier, so the average combination truck's 7,771 gallons cost about $49,600 a year, our calculation, and a one percent change in consumption is worth roughly $500 per truck. NHTSA counted 5,340 people killed in crashes involving large trucks in 2024, 70 percent of them in the other vehicle.

Where the data comes from: CAN, J1939 and OBD-II

Every vehicle built in the last two decades carries an internal network, and fleet management software begins with the question of what that network will tell an outsider. It is the controller area network, standardised as ISO 11898-1, whose 2024 edition added CAN XL frames of up to 2,048 bytes to the classic 8-byte frame; the classic bus is limited to 1 megabit per second, and CAN FD lifted the payload to 64 bytes. Cars and light trucks expose it through the OBD-II connector, heavy trucks through the SAE J1939 network, which uses 29-bit identifiers at 250 kilobits per second on older vehicles and 500 kilobits per second on newer ones. A device therefore has to detect the bus speed automatically before it reads anything; Geotab's GO9+ gateway lists 125, 250 and 500 kilobits per second.

The bus carries parameter groups with fixed addresses; the table lists the ones a fleet platform reads, with the ISOBUS registry as the public mirror of SAE's J1939 digital annex. Fault codes arrive unasked: the diagnostic message DM1 in PGN 65226 is broadcast every 1,000 milliseconds while a trouble code is active, so the platform knows about a sensor failure before the driver does.

The light-duty side is narrower than vendors imply, because the law that created OBD-II is an emissions law. The federal rule for heavy-duty engines, 40 CFR 86.010-18, requires engine speed, coolant temperature, torque and time since start on a diesel engine's data stream, lists vehicle speed and fuel level only "if used" by an emissions monitor, and does not put the odometer on the data stream. California's light-duty regulation, 13 CCR 1968.2, added odometer and fuel rate to the mandatory list in steps of 30, 60 and 100 percent of model years 2019, 2020 and 2021, and the federal rule for light vehicles adopts California's 2013 requirements for model year 2017 onwards, so the odometer and fuel-rate items are verified as California mandates only. Geotab's developer documentation states the result: "seat belt and odometer requests are proprietary on most passenger cars", so a device vendor's value is largely a library of reverse-engineered manufacturer messages, maintained per make, model and year.

Two transitions are under way. SAE J1979-2, diagnostics over the Unified Diagnostic Services protocol, becomes mandatory for model year 2027 light and medium vehicles and for heavy-duty engines from the 2027 engine model year, so a device library written against classic OBD-II PIDs has a shelf life. In Europe the eight truck makers behind the FMS-Standard interface, Daimler Truck, MAN, Scania, DAF, IVECO, Volvo, Renault and Ford Otosan, publish a common gateway specification, version 05 dated November 2025, that gives third parties a defined subset of J1939 without touching the powertrain bus; what each truck provides still "depends on manufacturer, model, production year, and installed ECUs", in the standard's own words.

Four ways to get vehicle data, and what each guarantees

A fleet platform can receive vehicle data through an aftermarket device, the manufacturer's embedded modem, a certified logging device or a tachograph, and the four differ in cadence, in legal standing and in who controls the feed.

The device path is the default because it is the only one that reports at operational cadence on every vehicle regardless of age. Samsara's gateway datasheet describes GPS updates every second, Motive's updates as frequent as every two seconds, and Geotab samples second-by-second GPS and hundreds of accelerometer readings per second. But a builder gets the vendor's API, not the device: Samsara's API delivers GPS every 5 seconds when the vehicle is on and every 5 minutes when it is off, engine speed roughly every 2 minutes and fuel on each 1 percent change, tells integrators not to request updates more often than every 5 seconds, and limits each token to 150 requests per second; a Kafka connector serves those who need sub-minute data without polling.

The embedded-modem path removes the hardware and the installer. Ford Pro's April 2026 release notes state that position, speed, odometer and heading now arrive every 5 seconds instead of every 30, and the basic product is complimentary on 2021 and newer Ford and Lincoln vehicles. GM offers OnStar data through an API with no dongles and ships the basic fleet service as standard on most 2025 and newer vehicles; Mercedes-Benz sells per-vehicle data packages through a Fleet Partner API that pushes events to a Kafka consumer the integrator must run; Stellantis's data arm, Mobilisights, counts 14 million connected vehicles in 2024 across 14 brands. The weakness shows in Smartcar's data freshness table: 15 to 30 minutes for Ford's cloud, 30 to 45 minutes for GM brands, 30 to 60 minutes for Toyota in the US and 1 to 3 minutes for Mercedes-Benz, BMW and Volvo in Europe, and fetching more often does not help because "the data won't change". A manufacturer feed is excellent for odometer, fuel and fault codes and, outside Ford's new cadence, unusable for live dispatch. Smartcar prices access from $1.99 per vehicle per application per month; Geotab's OEM page shows the logos of more than twenty manufacturers, our count.

How often a vehicle should report, and what each interval costs

Reporting cadence is the single parameter that sets data volume, cellular cost, cloud cost and the quality of every derived metric, and it should be chosen per signal rather than once. A US property-carrying driver may drive 11 hours in a duty period; the table shows how many position records each documented cadence produces in that day. A compact telematics record sits between the 45-byte minimum and 255-byte maximum of a widely used device protocol, so a second-by-second vehicle produces 1.8 to 10.1 megabytes a day before transport overhead and a 5-second vehicle 0.36 to 2.0 megabytes. Our calculation throughout.

Vendors do not store every sample. Geotab's gateway samples each second but logs with its curve algorithm, a Ramer-Douglas-Peucker simplification that keeps a point only when the line between kept points would deviate from the true path by more than an allowed error, and says it cannot state a fixed logging rate. Geotab reports about 6 million vehicles and assets and 100 billion data points a day, roughly 16,700 points per vehicle per day by our division. Samsara's data platform processed over 25 trillion data points in the fiscal year ended January 2026, up from over 14 trillion the year before and over 9 trillion the year before that, and Motive's engineering blog described 260 million GPS and telemetry points a day in 2021 through an ingestion path built because "we cannot tolerate any data loss" and peaking at almost 800 requests a second into Kafka.

The cost side is lopsided. At AWS IoT Core's published prices, $1.00 per million messages for the first billion a month, $0.15 per million rule evaluations and $0.15 per million actions, with connectivity at $0.08 per million connection-minutes, a fleet of 1,000 vehicles reporting every 5 seconds for 8 moving hours on 22 days generates 126.7 million messages a month and a bill of about $175, including a year of raw positions in object storage: about $0.17 per vehicle per month, our model. The SIM in the same vehicle moves about 51 megabytes a month with protocol overhead, which in the same model costs $2.52 at Hologram's self-service rate of $0.03 per megabyte plus $1 per SIM, $3.03 on Soracom's 50-megabyte US plan, or $7.07 at the prototyping rate KORE publishes for the former Twilio Super SIM. Connectivity is 14 to 40 times the cloud ingest line, and a 10-year, 500-megabyte flat SIM at $14 lasts about ten months at 5-second reporting, so the cheapest tier only works at one position a minute.

The expensive mistake is downstream. Snapping every raw fix to the road network through a paid API, at 100 points per request, means 1.27 million requests a month for that fleet: about $7,950 a month at Google's Roads API tiers and about $1,920 at Mapbox's map-matching tiers, the latter only if results are shown on a Mapbox map. Both exceed the entire ingest bill, so a platform matches per trip, after the trip closes, and only where it needs road attribution.

The telematics pipeline, from the vehicle to the dashboard

The pipeline is the product. Every screen in fleet management software is a view over the same ordered stream of timestamped events, and the engineering is in keeping that stream complete, de-duplicated and queryable at both ends of its life: the last 30 seconds for a live map and the last seven years for an audit.

The gateway buffers, so the server must expect bursts. In tunnels, garages and rural dead zones the device keeps logging: Geotab's GO9+ holds up to 80,000 logs out of coverage and, on a detected collision, sends the last 72 second-by-second records; Motive's gateway advertises up to four months of log memory. When coverage returns the device replays its backlog out of order, so ingestion has to accept events hours old, insert them into the right trip and never count a replayed batch twice. The protocols help only partly: MQTT 5.0, an OASIS standard messaging protocol, offers quality-of-service levels up to exactly-once between client and broker, and Kafka's idempotent producer deduplicates by producer ID and sequence number, but neither knows that two messages sent before and after a reboot describe the same second. The deduplication key has to be application-level, device identifier plus sequence number or device timestamp, checked at the point of durable write.

Storage splits by access pattern. Positions and engine samples are time-series data at a volume where row storage is expensive to query, so the usual choices are a time-series extension on PostgreSQL, a columnar time-series engine, or object storage with a query layer. Tiger Data claims 90 to 98 percent compression for TimescaleDB's columnstore, and a competitor's 2026 benchmark, the vendor's own result, measured TimescaleDB at about 1.3 million rows per second of ingest on a 32-vCPU instance against 7.1 to 8.6 million for QuestDB. Geotab described in 2022 nearly 3 billion raw records a day landing in BigQuery and tens of petabytes under management. Managed options move: Amazon closed Timestream for LiveAnalytics to new customers on 20 June 2025, steering new builds to its managed InfluxDB, and AWS IoT FleetWise is no longer open to new customers. Trips, drivers, vehicles, work orders and fuel transactions stay relational, and retention has a legal floor set in the compliance sections; Geotab documents hourly snapshots for a day, daily for a week, weekly for a month and monthly for ten years.

Our supply chain visibility software guide covers the shipper-facing view of the same stream, where position is joined to orders rather than drivers.

Spatial queries get their own index: positions are bucketed into hierarchical cells and only candidate fences are tested, rather than a polygon test per point. Uber's H3 grid gives a 531-metre edge at resolution 8 and 75.9 metres at resolution 10; Google's S2 cells run from about 1.27 square kilometres at level 13 to 1,237 square metres at level 18. The same cells feed the IFTA and tachograph jurisdiction logic.

Turning signals into events: trips, harsh driving and fault codes

Raw telematics becomes useful when the platform derives events from it, and each derivation carries thresholds that are engineering decisions with safety and labour consequences. A trip runs from ignition-on to ignition-off. Geotab logs the odometer at every ignition on and off and every two kilometres and engine hours every hour and at ignition off, the minimum needed to reconcile GPS distance against the vehicle's own count; the two drift apart in cities, where an iPhone 6 tested in a city showed 7 to 13 metres of average horizontal error against the 4.9 metres under open sky that GPS.gov cites, and the error correlates with building height.

For harsh events, Geotab sets GPS-based thresholds per vehicle class, 0.30 g for acceleration and braking on a passenger car and 0.20 g on a heavy-duty truck, with cornering at 0.42 and 0.24 g. Motive's hard-acceleration thresholds are 0.44 g for light, 0.41 g for medium and 0.38 g for heavy vehicles, computed, according to Motive's harsh driving page, from vehicle speed, GPS speed and the gateway's inertial unit together; 0.44 g is 9.65 miles per hour per second, matching Motive's own conversion. The spread between 0.20 g and 0.44 g for the same word, harsh, means an event count is a product of policy, and driver scores cannot be compared across vendors. A current gateway carries an accelerometer at 1.66 kilohertz with a range of plus or minus 8 g, and GPS speed must be cross-checked against CAN speed to avoid false events under bridges.

Fault codes are the one event class the vehicle volunteers, and a DM1 message with an active trouble code is the earliest possible trigger for a work order. Across all these events the design rule is to store the raw samples that produced them: a score that cannot be re-derived when a threshold changes is a liability, and the evidence that an event happened is what an insurer, a court or a driver's union will ask for.

The compliance layer in the United States: ELD, hours of service, DVIR and IFTA as engineering requirements

For a commercial fleet in the United States the compliance layer of fleet management software is not a feature but a specification with numbers in it, and most of those numbers live in one document: Appendix A to Subpart B of 49 CFR Part 395, the technical specification for electronic logging devices. Carriers have had to use a registered device since 18 December 2017, except for drivers who keep records on no more than 8 days in any 30, driveaway-towaway operations, vehicles manufactured before model year 2000, and short-haul drivers who stay within a 150 air-mile radius and return within 14 hours.

What a certified ELD must do, in numbers

Three parts of the specification shape a platform's data model. The first is the edit model: an edit must not overwrite the original record, automatically recorded driving time cannot be shortened except when unidentified records are assigned or team drivers swap with both confirming, every edit carries an annotation of at least 4 characters, the driver certifies each day and re-certifies after any change, and a carrier's edit is only a proposal until the driver confirms or rejects it. That is an append-only event log with status codes; a platform that stores hours as a mutable row will fail its first audit. The second is the unidentified-driver account: driving without a login goes to a single account labelled "Unidentified Driver", which the carrier must review and keep for 6 months. The third is transfer: the regulated output is a single comma-delimited file with a 25-character filename built from the driver's surname and licence digits, checksummed line by line, sent over SOAP 1.2 with TLS 1.2 to an FMCSA endpoint or as an AES-256 encrypted S/MIME email, or handed over on USB 2.0 or Bluetooth 2.1 at the roadside.

The location rules set the privacy balance. FMCSA's 2015 rule keeps positions coarse, reduced to a 10-mile radius during personal use "to further protect the driver's privacy", and says ELDs "are not required to track a CMV or driver in real time"; 49 CFR 390.36 makes it harassment to use ELD information in a way the carrier knew would push a driver into violating fatigue or hours rules, while permitting productivity monitoring short of that. Drivers must be able to access their own records without going through the carrier, and usernames may not contain a licence or Social Security number.

Certification is by the vendor's own declaration: a provider can self-certify that each model and version "has been sufficiently tested" and register it online, with no third-party laboratory. The registered list carried 939 entries, 813 distinct device names, when the Internet Archive captured it on 19 August 2026, and the revoked list 385 entries on 31 August 2026, our count of the table rows: roughly 29 percent of every device ever listed has been removed. By its own bulletins FMCSA had removed 79 devices since January 2025 by 20 May 2026, and three further bulletins through 6 August 2026 add 16 more, 95 in nineteen months by our sum; a carrier has 60 days to replace a revoked device before its drivers are placed out of service. In December 2025 the agency overhauled the vetting process, saying the old process made it easier to "re-register devices that had been revoked".

Hours of service as a rules engine

The hours-of-service rules are shorter than their reputation. A property-carrying driver may drive 11 hours after 10 consecutive hours off, within a 14-hour window from coming on duty, must take a 30-minute break once 8 cumulative hours of driving have passed, may not exceed 60 hours on duty in 7 days or 70 in 8, and resets with 34 consecutive hours off. Sleeper-berth time can be split into 7 and 3 or 8 and 2 hours, adverse conditions add up to 2 hours, and passenger carriers run on 10 hours of driving and a 15-hour duty limit. The 2020 rule behind the current text widened the short-haul radius from 100 to 150 air miles, moved the break to 8 hours of driving rather than on-duty time and added the 7/3 split; the agency has announced two pilot programmes for 2027, a more flexible sleeper split and a pause in the 14-hour window, with 256 drivers each. A rules engine that computes remaining hours in real time from the event log, visible to a dispatcher before a load is assigned, is the whole of the hours module; the hard part is the exceptions.

DVIR and maintenance records

Under Part 396, a driver prepares a vehicle inspection report at the end of each day, but since 2014 for property carriers and since September 2020 for passenger carriers only when a defect is found; the carrier certifies the repair, or that repair is unnecessary, before the vehicle runs again and keeps the report for three months, and the next driver reviews and signs it before the trip. A rule effective 23 March 2026 states that the report may be created and kept electronically with electronic signatures. The carrier must also keep a systematic maintenance record per vehicle for a year and for six months after the vehicle leaves its control, prove an annual inspection within the preceding 12 months with the report kept 14 months, and certify corrections from a roadside inspection within 15 days. The Pocket Guide shows what goes wrong: among 2023 driver violations, false logs numbered 60,442, failure to certify the ELD record 29,537, and a device that could not transfer its data 20,504, and 190,975 vehicles were cited for missing proof of the periodic inspection. Each is a software failure as much as a driver one.

IFTA mileage by jurisdiction

Fuel tax is a geofence problem. The International Fuel Tax Agreement covers 48 states and 10 Canadian provinces, applies to vehicles over 26,000 pounds or with three or more axles that cross jurisdictions, and requires a quarterly return within a month of quarter end apportioning fuel tax by miles driven in each jurisdiction, with records kept four years. State auditors have published what a GPS distance record must contain: a position at least every 10 minutes while the engine is on for IFTA and every 15 for the International Registration Plan, latitude and longitude to at least four decimal places, the engine odometer at each reading, and a spreadsheet export, because, as Maine's Bureau of Motor Vehicles puts it, "data in a static format such as a PDF, JPEG, PNG, or Word is not acceptable"; Nebraska's notice gives the same intervals. The platform's part is the spatial join: each qualifying position assigned to a jurisdiction polygon, distances summed between readings, gaps flagged, and the result reconciled to the odometer. The FMCSA's Safety Measurement System scores carriers on seven categories, Unsafe Driving, Crash Indicator and Hours-of-Service Compliance among them; a revised methodology adopted in November 2024 still had no launch notice by 3 October 2026.

The compliance layer in Europe: smart tachograph 2, driving-time rules and the 1 July 2026 van line

In the European Union the regulated record is the tachograph, the rules it enforces are Regulation 561/2006, and the two deadlines that matter to a fleet platform in 2026 are 24 December 2025, after which only a full version 2 smart tachograph may be fitted to a newly registered vehicle, and 1 July 2026, when vans over 2.5 tonnes in international carriage joined the regime. Article 2 of Regulation 561/2006 as amended by the first Mobility Package applies the driving-time rules to goods vehicles over 3.5 tonnes and, "from 1 July 2026, of goods in international transport operations or in cabotage operations", to vehicles over 2.5 tonnes; Article 3 of Regulation 165/2014 requires a tachograph in every vehicle those rules cover. The UK's DVSA applies the same line to vans of 2,501 kilograms or more on international hire-and-reward journeys, with UK-only operation of those vans still exempt.

Smart tachograph version 2 and its deadlines

The second-generation smart tachograph is the one device in a fleet whose data model was written by legislators. Under Regulation 165/2014 as amended by 2020/1054 it records position at the start and end of the working day, each time accumulated driving reaches a multiple of three hours, at each border crossing and at each loading and unloading; the implementing regulation adds border detection by digital map, a GNSS anomaly event, software updates without replacing the unit, a mandatory Bluetooth interface and Galileo's authenticated navigation signal, OSNMA. Enforcement can query the unit while the vehicle is moving over a short-range link, with "exceeding maximum driving time" among the data exposed, and the device must be "designed in such a way as to ensure privacy". The driver card holds at least 56 days of activity.

The Commission's question-and-answer page consolidates the retrofit dates. The Member States granted a two-month "educational learning period" after the December 2024 deadline, but the Commission's pages show none for the August 2025 deadline, and transitional units without OSNMA could be fitted until five months after the service declaration, which EUSPA dated 24 July 2025, hence the December 2025 line that the DVSA repeats for Great Britain.

For the software, three consequences follow. The tachograph is not a tracking source, so the live map still comes from the gateway or the manufacturer modem. The compliance job is the download schedule, the infringement analysis against the table above, the 12-month archive and the 56-day roadside window, and from July 2026 the vehicle master data has to know which vans do international work. And the legal footing of the data shifted on 12 September 2025, when the EU Data Act, Regulation 2023/2854, began to apply: Article 4 gives the user of a connected product access to its data "continuously and in real-time" where technically feasible, Article 5 lets the user direct it to a third party of their choosing, Article 3 requires products placed on the market after 12 September 2026 to be designed for that access, and from 12 January 2027 cloud providers may no longer charge switching fees. A European platform built in 2026 should assume the vehicle's data belongs to the operator and the manufacturer's portal is not the only route to it.

The UK runs the same limits under assimilated law, 9 hours a day, 56 a week and 90 a fortnight for vehicles over 3.5 tonnes, with a separate domestic regime of 10 hours of driving and 11 hours of duty for the rest, and employers keep records for at least a year. Zero-emission goods vehicles of 3.5 to 4.25 tonnes are exempt from the tachograph rules in Great Britain from 1 June 2026.

What the measured studies say about driver safety programs

Fleet management software sells safety, and the independent evidence is narrower and more useful than the brochures. Read together, the studies say that monitoring changes the monitored behaviour, that feedback without a human in the loop changes little, and that the regulator's own flagship intervention did not deliver the safety gain for small carriers that its rule projected.

The ELD mandate is the natural experiment. FMCSA's 2015 rule projected 1,844 crashes avoided and 26 lives saved a year and assumed $419 a year for a telematics ELD and $166 for a USB and Bluetooth one, against $809 a year of paperwork saved per driver. The pre-mandate 2014 FMCSA study of voluntary adopters in the table was favourable; the post-mandate evidence is not. Scott, Balthrop and Miller compared small and large carriers before and after the mandate and found, in the abstract's words, that monitoring "achieved its first-order effect of reducing the frequency that drivers were in non-compliance with work hour rules", that "drivers for small firms that were most affected by the mandate also appear to have increased their frequency of unsafe driving (e.g., speeding)", and that "crash counts for small firms did not fall relative to large firms, and may have increased"; the University of Arkansas's news release on the paper puts the owner-operator increase at 11.6 percent. The likely explanation, our reading rather than a finding of either study, is selection: the carriers that chose recorders were mostly large for-hire fleets, and a device that limits hours does nothing about the pressure that made drivers exceed them. The DOT Inspector General estimated one such pressure in 2018: each 15-minute increase in average dwell time at a facility raises the expected crash rate by 6.2 percent, and ATRI's GPS data show detained trucks driving 14.6 percent faster on average than trucks that were not detained. Recording dwell time at customer sites measures a safety variable as well as a billing one.

The coaching evidence is the most useful for product design. The FMCSA video study found 38.1 and 52.2 percent fewer safety-related events per 10,000 miles once safety managers coached drivers on their recorded events, though it had no control group and the camera vendor reduced the data. The NIOSH study had a control group and separated the mechanism: in-cab lights alone were not significantly different from no intervention, while lights plus video coaching roughly halved the odds of a risky event. The Dutch recorder study had found about 20 percent fewer accidents with feedback two decades earlier; usage-based insurance drivers cut hard braking by 21 percent in six months, while a randomised trial of telematics feedback sent by text message and app found no significant effect from feedback alone and a small one when money was attached. Hardware that warns or brakes on its own has the cleanest numbers: forward collision warning and automatic emergency braking on heavy trucks were associated with 44 and 41 percent lower rear-end crash rates in the IIHS analysis.

A score on a dashboard changes nothing by itself; a clip a supervisor reviews with the driver within the week does. The software's job is to make that coaching loop cheap and fast, keep the raw event so a disputed clip can be re-examined, and measure the fleet against its own per-mile baseline rather than a vendor's "up to" figure. For context, 5,340 people died in crashes involving large trucks in 2024, the involvement rate was 1.58 fatal crashes per 100 million large-truck miles, and the Large Truck Crash Causation Study, on crashes from 2001 to 2003, assigned the critical reason to the truck in 55 percent of cases and to the driver in 87 percent of those, with brake problems coded on 29 percent of the trucks.

Fuel, idling and maintenance: where the savings can be measured

The fuel and maintenance modules are where fleet management software returns money, and the measured quantities behind them come from national laboratories and enforcement data rather than from vendors. The Department of Energy gives about 0.8 gallons an hour for an idling heavy-duty truck and roughly 1,800 hours a year of rest-period idling for a long-haul sleeper, about 1,440 gallons by our multiplication; all US idling burns more than 6 billion gallons a year. A Class 7 bucket truck idling to run its equipment burns about 1.0 gallon an hour in the Department's conservative base case and 1.5 at high idle, a light vehicle a quarter to half a gallon an hour. NREL's 1-hertz CAN and GPS logging of 46 drayage trucks at the Port of New York and New Jersey found 4.4 hours of idling in an 8.4-hour engine-on day, about 9 percent of daily energy. At the late-September 2026 diesel price, 0.8 gallons an hour is about $5.10 an hour, our calculation, which is what an idling alert is worth when a supervisor acts on it.

Speed is the second lever, and the measured curve is steeper than the rule of thumb. Oak Ridge National Laboratory instrumented six Class 8 trucks with 56 channels at 5 hertz over about 700,000 miles and found, at an average vehicle weight of about 60,300 pounds on flat terrain, 10.02 miles per gallon at 59 mph, 8.58 at 65 and 7.98 at 70: a 20.4 percent penalty from 59 to 70 mph, our calculation, or roughly 0.19 mpg per mile per hour. EPA's SmartWay guidance uses the gentler rule of about 0.1 mpg per mile per hour, which it calls "a general rule of thumb". For light vehicles, aggressive driving lowers fuel economy by 15 to 30 percent at highway speeds and feedback devices improve it by about 3 percent on average, the sober number to set against the larger savings fleets are often promised. Tyre pressure costs about 0.2 percent of fuel economy per 1 psi of under-inflation across all tyres.

Maintenance is where the fault-code stream earns its place. In the CVSA's 2026 International Roadcheck, inspectors placed 10,350 vehicles out of service in three days, 19 percent of 54,575 inspections across North America; brake systems accounted for 24.3 percent of vehicle out-of-service violations and tyres for 20.9 percent, the same ranking as in 2025. In 2023 the national vehicle out-of-service rate in FMCSA's data was 22.6 percent of vehicle inspections, out of 3 million roadside inspections. A peer-reviewed study logged nineteen Swedish city buses at 1 hertz on more than a hundred signals for four years: they spent 11 percent of their time in or at workshops, about eleven visits a year, more than twice the planned number, and the authors showed that comparing each bus against the fleet's consensus flagged coolant, air-leak and climate deviations that ended in repairs. ATRI's 2026 update has repair and maintenance costs up 8.6 percent in a year when total cost per mile rose 3.4 percent.

The maintenance module is a scheduler with three triggers, odometer and engine-hour intervals read from the bus, calendar intervals and the DM1 fault stream, feeding work orders whose records double as the Part 396 maintenance file and the evidence a residual-value buyer asks for. The fuel module matches each fuel-card transaction to the vehicle's position and fuel level at the swipe and flags a fill larger than the tank, a purchase far from the vehicle, or two fills minutes apart. No credible public measurement shows how much fleet-card fraud that catches, so this article quotes none.

Video telematics and driver privacy: the rules a platform has to design for

Video telematics has moved from evidence after a crash to inference during the drive, and that shift brings the platform into the territory of privacy regulators who have written specific rules for it. Current dash cameras run their models on the device: Samsara's dual-facing camera records the road at 1080p and the cabin at 720p, keeps 33 to 87 hours of trip video locally and uploads clips of critical events automatically; Motive's 2026 camera carries a Qualcomm QCS6490 processor and up to 160 hours of recording, with retrievals of up to 15 minutes per request; Lytx's recorder uploads events daily from 256 gigabytes of local storage. Continuous video stays in the vehicle and events leave it, so the platform's part is the clip queue, the review workflow and the retention schedule.

The European Union has written the strictest design pattern into the vehicle itself. The General Safety Regulation requires a driver drowsiness and attention warning on every new vehicle registered from 7 July 2024 and an advanced driver distraction warning from 7 July 2026, and its Article 6 states that these systems "shall be designed in such a way that those systems do not continuously record nor retain any data other than what is necessary", that the data "shall not be accessible or made available to third parties at any time and shall be immediately deleted after processing". The delegated rules add that drowsiness detection must work without biometric information, including facial recognition, and that the distraction warning must come once the driver's gaze has been off the road for at most 3.5 seconds at 50 km/h and above. A fleet camera that continuously records and uploads the cabin does, as a commercial product, what the regulator refused to let the manufacturer's safety system do, and a European platform should expect works councils and data protection authorities to notice.

The table draws on primary texts. CNIL's guidance on geolocation of employees' vehicles also records a EUR 10,000 fine against an employer that refused a driver access to his own location data, and CNIL's simplified procedure sanctioned continuous vehicle geolocation again in 2023 and 2025. The UK row rests on the ICO's guidance on monitoring workers; the US state row on New York's Civil Rights Law 52-c, California's Penal Code 637.7, which accepts the consent of the vehicle's owner, lessor or lessee, and AB 1331, signed on 30 September 2026, which in its enacted form prohibits only surveillance in workplace bathrooms; and the German row on the Works Constitution Act and the Federal Labour Court's July 2024 ruling. The EDPB's connected-vehicle guidelines set the baseline every authority borrows from, location only "if doing so is absolutely necessary" and "not by default and continuously when the car is started", and the board's May 2026 summary adds that reading data from a vehicle usually requires prior consent under the ePrivacy rules.

These controls are cheap to build early and expensive to retrofit, and the most structural is separating the coarse regulated record from the precise operational feed, so that an inspector's request and a customer's "where is my delivery" are answered from different tables. The hub logistics practice describes how we scope them into the discovery phase rather than after a works council objects.

EV fleets: what changes in the software

An electric fleet changes the data model of fleet management software before it changes anything else, because the two quantities that matter most, state of charge and charging, come from places the diesel pipeline never touched. For a battery-electric vehicle, battery state is not part of the emissions-driven OBD-II data stream, which carries only a hybrid's battery pack remaining charge. SAE published J1979-3, a diagnostic standard for zero-emission propulsion, in October 2023, and California's 13 CCR 1962.5 would have required 40 percent of model year 2026 and all model year 2027 zero-emission vehicles to expose speed, odometer, pack state of charge and state of health, cell voltages and charge rate on the standard connector. Whether that still binds is unsettled: Congress disapproved the federal waiver for the regulation it belongs to in Public Law 119-16 on 12 June 2025, and whether the data requirement survives that disapproval is a legal question this article does not try to answer. In 2026 a platform gets state of charge from the manufacturer's feed or from a device vendor's model-specific decoding.

Charging brings a second protocol family. Depot chargers speak OCPP to a management backend; the Open Charge Alliance states that the third edition of OCPP 2.0.1 was approved as IEC 63584 in 2024, that version 2.1 arrived in 2025 with bidirectional power support, and that 1.6 and 2.0.1 are not compatible, so a depot with both generations runs two integrations. The vehicle-to-charger conversation is ISO 15118-20, the second-generation network and application layer, which defines the messages for bidirectional power transfer and for wireless charging. The EU's infrastructure regulation requires ad hoc payment with a widely used payment instrument at public points deployed from 13 April 2024 and heavy-duty pools of at least 3,600 kW every 60 km on the core network by 2030; the Battery Regulation requires state-of-health data in the battery management system from 18 August 2024 and a battery passport from 18 February 2027. The platform therefore joins vehicle telemetry, charger sessions, tariff windows and battery health in one schedule, and routes on remaining energy, not remaining fuel.

NREL's evaluation of twelve battery-electric buses at Foothill Transit found 2.17 kWh per mile and 90 percent availability against 93 percent for the natural-gas fleet, with the two charging heads available 98 and 99 percent of the time; NREL's modelled battery-electric drayage tractor on the measured New York duty cycle needs about 2.4 kWh per mile and 417 kWh a day. NACFE tracked 22 battery-electric trucks at ten depots over Geotab telematics and found them connected to a charger for 4.3 hours on average per session, and quotes Geotab's analysis of 1.3 million vehicles that 81 percent of medium-duty trucks drive less than 150 miles a day. Public charging is the weak link: a field test of 655 fast-charging ports in the Bay Area found 73.3 percent functional against operators' reported uptime of 95 to 98 percent, so a fleet platform needs its own charger-reliability data rather than the network's. The European fleet is still small: the Commission's Alternative Fuels Observatory counted 36,765 battery-electric trucks in classes N2 and N3 in the EU in 2026 and 2,671 public charging points, our sum of its mixed-use and heavy-duty-only figures, at locations with at least one 350 kW unit in the third quarter, while the heavy-duty CO2 standards require manufacturers' fleets to cut emissions 45 percent by 2030 and 90 percent by 2040 against 2019. California's Advanced Clean Fleets rule now applies only to state and local government fleets after the 2025 amendments were filed on 10 September 2026.

What fleet management software costs to buy

Published fleet management software plan prices run from about $2.50 to $55 per vehicle per month, and about $65 to $67 with a camera add-on, the spread is almost entirely explained by what is included, and the enterprise vendors publish nothing at all. We read every pricing page on 3 October 2026; the figures below are list prices for the plan named, monthly equivalents where billing is annual, and exclude taxes and installation.

Sources: Simply Fleet, Fleetio, AUTOsist, Bouncie, One Step GPS, Quartix, Force by Mojio, Linxup, Azuga, Garmin, Konexial; Samsara, Motive, Geotab, Verizon Connect, Webfleet, Fleet Complete and Netradyne publish no list price.

The enterprise tier is quote-only, and the only primary prices come from public bodies that must publish what they pay. A Samsara order form issued to the City of Cayce, South Carolina, on 12 November 2025 and put to its council on 2 December 2025 priced 135 vehicle gateways on a 36-month term at $216.72 per gateway per year, $18.06 a month by our division with hardware included, on a public-sector licence without Wi-Fi or ELD, plus $504 a year for each forward-facing camera licence. Oklahoma's statewide contract lists Geotab's Pro+ plan at $25.00 a month list and $17.00 contract price with the device included and installation at $150 list; the 2022 NASPO schedule had the bundle at $20.50 a month with a 12-month minimum and a $135 cancellation fee.

Contract length matters more than the monthly rate. Samsara's annual report says its contracts "are typically for an initial subscription term of three to five years", that it prices "on a per asset, per application basis", so a vehicle with telematics and video counts as two subscriptions, and that about 98 percent of its revenue is subscription bundling the device, the cellular connection, the applications and the warranty. Motive's registration statement, filed in December 2025 and withdrawn in September 2026, says its subscriptions "typically span three years, and are generally noncancellable with automatic renewal". Dividing Motive's disclosed $501 million of annual recurring revenue at September 2025 by its "more than one million" vehicles and assets gives an upper bound of about $42 a month per asset, our calculation; Powerfleet's $359.8 million of services revenue across "nearly three million" subscribers, a figure from its results release, is about $10 a month, our calculation, and neither number is a market average, because the units and periods differ.

Samsara reported $1,889.9 million of annual recurring revenue at 31 January 2026 and $2,124.7 million at 1 August 2026, revenue of $1,618.6 million for the fiscal year at a 77 percent gross margin and about 14 percent of revenue from outside the United States. Motive disclosed revenue of $370.5 million for 2024 and nearly 100,000 customers, and said in a September 2026 release that recurring revenue had passed $600 million. Powerfleet reported $443.8 million for the year to March 2026; Bridgestone counts about 860,000 vehicles under Webfleet and 400,000 under Azuga; Michelin's connected-fleet unit covers more than a million vehicles. Research-firm market sizes in circulation differ several-fold for the same year depending on the firm and are not used here.

Build versus buy: the arithmetic for 50, 500 and 5,000 vehicles

The build-versus-buy decision for fleet management software is settled by three numbers that are now public, the per-vehicle subscription, the per-vehicle running cost of a self-built pipeline, and the cost of the team, and for most fleets the team settles it. Take the subscription at $25 a vehicle a month, the list price of Linxup's tracking plan and Azuga's entry tier and equal to the list price in Geotab's Oklahoma contract; the cloud and SIM cost of an own pipeline from the cadence model above; a one-time device at $69 to $149, the Linxup tracker price and the Force tracker's list price before its sale discount, since no white-label device price was readable; and the team at our own published rate card, $6,000 to $9,500 per engineer-month, which is the publisher's price and should be read as such. Five engineers for a year to reach a compliant first release, and two to run it, are assumptions, not measurements.

A worked example of fleet management software costs at 50, 500 and 5,000 vehicles

Worked example, our calculation with the assumptions stated. At 50 vehicles the sustain team alone costs about 10 to 15 times the subscription, and at 500 the sustain team plus the pipeline, $160,200 to $271,500 a year, costs more than the $150,000 subscription; building is not a financial decision at either size. At 5,000 vehicles the subscription is $1.5 million a year and an own platform costs $306,000 to $663,000 a year after the first year's build and devices, so the build pays back in the second or third year on licence cost alone, before the value of owning the data is counted. The comparison ignores what the subscription buys that a build must reproduce, a certified ELD, a hardware supply chain, a cellular contract at scale and a support desk, along with the integration surface that our supply chain software cost guide shows drives most logistics budgets; it also ignores that enterprise contract prices are below list.

There is a third option, and it is where most custom fleet software is actually built. The device vendors sell their platforms as data sources: Samsara's annual report counts more than 350 third-party integrations in its marketplace, Geotab publishes a developer SDK, Fleetio exposes an API from its Professional plan, and the manufacturers now sell the feed without a device. A marketplace, a 3PL operation or a specialised rental fleet buys the telematics, including the ELD, and builds the layer that makes it a business: customer-facing tracking, dispatch rules, billing, and analytics that join vehicle data to orders and margins. That is the shape of the real-time platforms we have built for logistics operators, HaulBreeze, the auto-transport platform Conveya and the on-demand car-service marketplace Pikk-Up, and of most custom software development engagements in this sector.

Building the full stack is the right call in four situations: the fleet platform is the product, sold to other fleets or embedded in a vehicle-rental, leasing or insurance business, where a 3-to-5-year vendor contract would sit between you and your customers; the vehicles are not the vendors' vehicles, such as forklifts, construction equipment, vessels, rail wagons or robots; the data has to stay in a jurisdiction or on premises for a regulatory or works-council reason the vendors' terms cannot meet; or the fleet is large enough that the arithmetic above closes and the organisation already runs data infrastructure. Even then the certified ELD or tachograph remains a bought component; nobody should write an Appendix A implementation to save a subscription.

What changed in fleet management software in 2025 and 2026

Regulation and the vendor market moved more in these two years than in the five before them, and a platform specified in 2024 needs several corrections. In the United States, beyond the device removals, the vetting overhaul and the electronic inspection rule described above, FMCSA dropped the requirement to carry the device's user manual in the cab from 22 July 2026 after more than 3,000 drivers were cited for its absence in 2024, denied a request on 30 September 2026 to let drivers return to paper logs, and scheduled the long-awaited ELD specification proposal for November 2026 in the regulatory agenda. On the data side, T-Mobile retired its 2G network on 3 August 2026 after the US 3G shutdowns of 2022, Germany's regulator has published 2G shutdown dates in 2028 that will strand older devices, and SAE revised the J1939 and J1979-2 standards in 2026 ahead of the model year 2027 switch to diagnostics over UDS.

In the vendor market, Motive withdrew its IPO on 10 September 2026; Trimble sold its telematics business to Platform Science in February 2025; CalAmp went through a prepackaged Chapter 11 and emerged in July 2024 under new ownership; and Powerfleet absorbed MiX Telematics and Fleet Complete. A vendor's ownership and device list can change inside a three-year contract, and a platform that owns its own copy of the data survives that. Diesel moved too: the late-September 2026 price of $6.382 a gallon was $2.63 above a year earlier, about 70 percent higher by our calculation and about 74 percent above the 2025 average of $3.66, so every idling and speed figure in this article is worth that much more.

Where fleet management software goes from here

Fleet management software is converging on three sources of truth that it did not have five years ago: the manufacturer's own data feed at operational cadence, a camera that infers rather than records, and a regulated record that is authenticated by satellite. The winning platform will treat all four data paths as inputs to one event log, keep the raw samples behind every score, and build the privacy controls that CNIL, the ICO and a German works council will ask for. For almost every fleet the right answer is still to buy the telematics and the certified logging device and build only the layer that is the business. If you are scoping a fleet platform in 2026, start with the data path and the compliance record, and let the dashboard come last.