Real-Time Visibility is a Latency Claim, and Most Geofences Never Fire

Real-time visibility in logistics means position and status data arriving fast enough to act on before the situation changes. That definition contains a number, and the number is almost never stated. Every visibility platform sells geofence-triggered automation, where a vehicle entering a delivery zone fires a notification, starts a dwell clock or releases the next task. Whether that automation can fire is not a software question. It is arithmetic on how often the vehicle reports its position against how long it is physically inside the fence, and the arithmetic is unforgiving. A vehicle at 40 km/h crosses a 200 meter geofence in 36 seconds. On a five-minute reporting interval, the probability that any report lands inside that fence is 12%, so 88% of arrivals are never detected at all.

Nothing about that is a defect in the platform. It is a property of the feed, and it is knowable in advance from two numbers you already have.

Why real-time is an unfalsifiable claim without a latency number

The phrase real-time survives in logistics procurement because it is never quantified. Two platforms can both claim it while delivering data an order of magnitude apart in freshness, and no line in a standard RFP separates them.

Start with what the physical layer actually supports. GPS itself is not the limitation: the US government reports that GPS-enabled smartphones are typically accurate to within a 4.9 meter radius under open sky, against a committed signal-in-space error of 2.0 meters or better with 95% probability. That is far finer than any geofence you would draw. The caveat in the same source is the operationally relevant one, since accuracy degrades near buildings, bridges and trees, which describes most of the urban delivery environment.

So position precision is rarely what breaks. Reporting frequency is. And on the regulated side the frequency is coarse by design. Under 49 CFR 395.26, an electronic logging device creates an intermediate recording when the vehicle is in motion and no duty status change or other intermediate record has occurred in the previous hour. The same rule specifies that during authorized personal use, location is recorded at single decimal point resolution, which the regulation itself describes as approximately within a 10-mile radius. A compliance-grade feed is a 60-minute, sometimes 10-mile-resolution record. It is entirely fit for its purpose, which is hours-of-service enforcement, and entirely unfit for firing a geofence.

The cost side is worth stating because it sets what a missed detection is worth. ATRI’s 2026 report puts the industry-average cost of operating a truck at $2.336 per mile in 2025, a record for the series and 3.4% above the prior year, with $1.854 of that excluding fuel. An undetected arrival does not cost a mile directly. It costs the dwell that nobody clocked, the next task that was not released, the customer message that did not fire and the exception that surfaced twenty minutes later as a phone call. Those are the compounding costs, and they are invisible precisely because the triggering event was never recorded.

How to compute whether your geofences can actually fire

1. Establish the real reporting interval for each feed, not the platform’s

The number that matters is how often position reaches your system, per source. Owned fleet on a driver app, contracted vehicles on carrier telematics and third-party carriers on an API relay will differ by an order of magnitude. Ask for the interval in seconds, in motion, and ask separately what happens when the device loses signal and backfills. A platform refreshing its own dashboard every 30 seconds from a feed that updates every 5 minutes is displaying stale data promptly.

Carrier structure makes this unavoidable rather than unlucky. The American Trucking Associations reports almost 580,000 active US motor carriers registered with FMCSA as of June 2025, of which 91.5% operate 10 or fewer trucks and 99.3% operate 100 or fewer. The long tail of your carrier base is small operators, and small operators are the least likely to run tight-interval telematics, so a blended coverage figure averages your best feeds with the ones that will fail.

2. Compute dwell time inside the geofence

Dwell is the chord across the fence divided by approach speed. For a circular fence of radius r meters and a vehicle passing through the center, that is 2r divided by speed in meters per second. A 200 meter fence at 40 km/h, or 11.11 meters per second, gives 36 seconds. Use the approach speed on the final leg, not the route average, because the two differ most exactly where stops are dense.

3. Multiply out the detection probability

If dwell is shorter than the reporting interval, the probability that at least one report lands inside the fence is dwell divided by interval. If dwell is longer, detection is certain.

Read down the last column. The pattern that surprises operations teams is the inversion: a faster vehicle is harder to detect, so the arrivals most likely to be missed are the ones on open roads, not the ones crawling through a city center.

4. Solve for the radius your interval can support

Turn the calculation around and ask what radius would give reliable detection at your current interval. For 95% detection the fence has to be large enough that dwell reaches 95% of the interval, which makes the required radius roughly 0.95 times interval times speed, divided by two.

Now compare against stop spacing. Urban delivery stops commonly sit 150 to 400 meters apart, so a fence much beyond 400 meters cannot tell one stop from the next. That is the real constraint. At a five-minute interval and suburban speeds, the fence you would need for reliable detection is 1.58 km, which would trigger on four or five stops at once and is therefore useless for attribution. Ping interval, geofence radius and stop spacing form one constraint system, and fixing any two determines the third.

5. Add the other five terms in the latency budget

Detection is the first term, not the whole chain. A realistic budget on a five-minute feed, where the expected wait for the next report is half the interval, runs as follows.

Against a 10-minute window before a customer notices a problem themselves, that budget consumes 6.8 minutes, or 68%, before anybody acts. Note which terms dominate. The reporting interval and your own polling cadence account for 270 of the 405 seconds, and the rule engine everyone focuses on accounts for 10.

6. Instrument missed detections as a metric

None of this improves without measurement, and the measurement is available. Compare geofence entry events against confirmed stop completions from the driver app or the proof of delivery. Every completion with no matching entry event is a missed detection. That ratio is a direct read on feed adequacy, it costs nothing to compute, and almost nobody reports it. Teams that do usually find the number is not close to the one they assumed.

What each reporting interval can actually support

The last row is the point of the table. A 60-minute feed is not a bad feed. It is a feed built for a different job, and the failure happens when it is bought for one purpose and used for another.

Five criteria for evaluating a real-time visibility claim

Interval in seconds, per feed source. Ask for the number, in motion, separately for owned fleet, contracted vehicles and third-party carriers. A single blended figure hides the sources that will actually fail.

Dashboard refresh separated from data freshness. These get conflated in demonstrations. Ask what the age of the newest position on screen is, not how often the screen redraws.

Behavior on signal loss. Ask whether gaps are backfilled with timestamps from the device or stamped on arrival at the server. Backfilled data reconstructs history correctly and cannot trigger anything, and the difference matters for any automation.

Geofence radius defaults and whether they are configurable per site. A platform with one global radius cannot be right at both a distribution center and a high street. Radius has to follow interval and local speed.

A missed-detection report. Ask to see entry events reconciled against completed stops. If the platform cannot produce that comparison, it cannot tell you whether its own triggers are firing.

What this looks like in enterprise deployments

A leading North American retailer running multimodal logistics automation across several hundred stores replaced six legacy systems and now resolves exceptions in under two hours, with 99%-plus on-time store delivery and 95%-plus route compliance. Two hours is the operationally honest number here. It is not two seconds, and it does not need to be, because a store delivery exception is actionable for hours. Matching the latency of the response to the latency of the consequence is the discipline, and a two-hour resolution against a two-hour window is a solved problem, while a two-hour resolution against a ten-minute window is not.

A Canadian grocery brand running carrier orchestration for fresh and perishable home delivery across more than 30 cities reports customer support resolution 10 to 20 times faster, alongside 33% faster deliveries and 15% lower fulfillment cost. The support number is the visibility number in disguise. Resolution speed collapses when the agent already has position, status and cause on screen instead of calling a depot, which is what a feed fresh enough to act on actually buys. The value did not come from a tighter geofence. It came from the data being current at the moment somebody asked.

Four mistakes operations make on visibility latency

Buying real-time without a number attached. If the contract does not state a reporting interval in seconds per feed source, it does not specify real-time, and both parties can be satisfied while the automation silently fails.

Setting geofence radius from the map. Radius gets drawn to look sensible against a site boundary. It has to be derived from the reporting interval and the approach speed, and where those demand a radius wider than stop spacing, the honest conclusion is that stop-level triggering is not available on that feed.

Treating the rule engine as the bottleneck. In the worked budget the exception logic takes 10 seconds of 405. Effort spent tuning rules while the feed updates every five minutes is effort spent on 2% of the problem.

Assuming a compliance feed is an operations feed. ELD data is regulated, reliable and hourly. It answers hours-of-service questions completely and arrival questions not at all, and the mistake is structural rather than technical.

How Locus approaches visibility latency and exception triggers

Locus, the world’s first Decision-Intelligent, Agentic TMS, treats visibility as an input to a decision rather than a display. The Control Tower consolidates owned fleet, contracted transporters and a network of more than 1,000 carriers into one operational picture, and the Driver Companion App provides a first-party position feed on owned and contracted movements rather than depending on a carrier relay, which is where interval control is actually won.

Because the platform is built to act, the latency that matters is measured end to end rather than at the dashboard. Geofence-based exception triggers, automated re-sequencing and customer ETA updates run against the live plan, and the route planning system re-optimizes in roughly two minutes against more than 250 real-world operating constraints, so a detected deviation can be answered with a new plan inside the window where a new plan still helps. Explainability and Traceability record the trigger, context, reasoning, action and outcome for each decision, which is what makes a missed-detection audit possible: entry events can be reconciled against completions rather than assumed.

Two boundaries belong in this conversation. Locus is not a telematics provider and does not set the reporting interval on a third-party carrier’s hardware. Where a carrier supplies an hourly feed, no platform can manufacture a 30-second one, and the honest response is to expose the feed’s age per shipment so the operation knows which movements support stop-level automation and which do not. Second, tightening an interval has costs in device battery, data volume and driver-app behavior that belong in the decision. The right interval is the one your automations actually require, computed from the tables above, rather than the fastest one available.

Locus supports more than 360 enterprise customers across 30-plus countries, with over 1.5 billion deliveries optimized, more than $320 million in documented client logistics savings and 99.99% uptime. It has been recognized by Gartner for seven consecutive years, featured in the 2026 Hype Cycle for Supply Chain Execution and Logistics Technologies, named a Leader in TMS by QKS Group (SPARK Matrix), and ranked #1 in Route Planning on G2’s 2026 Best Software Awards.

In October 2025, Ingka Investments, the investment arm of Ingka Group, the world’s largest IKEA retailer, acquired Locus. Locus continues to operate independently.

So what does real-time visibility actually require? A reporting interval short enough that your geofences fire, which is a number you can compute rather than a claim you accept. Detection probability is dwell divided by interval, so a vehicle at 40 km/h crossing a 200 meter fence in 36 seconds is detected 12% of the time on a five-minute feed and 100% of the time on a 30-second one, and the required radius for reliable detection at five minutes is 1.58 km, wider than the 150 to 400 meter spacing between urban stops and therefore useless for attribution. Detection is also only the first of six terms: a worked budget consumes 6.8 minutes of a 10-minute window before anyone acts, with the feed interval and your own polling accounting for two-thirds of it. Locus addresses this by holding a first-party position feed through the Driver Companion App, consolidating owned, contracted and 1,000-plus carrier movements in the Control Tower, re-optimizing in roughly two minutes so a detected deviation can be answered while it still matters, and recording each decision so entry events can be reconciled against completions. Request a Locus visibility assessment to compute the detection rate your current feeds support.

FAQs

What does real-time visibility actually mean in logistics? It means position and status data arriving fast enough to act on before the situation changes, which makes it a claim about latency. Without a stated reporting interval in seconds the phrase is unfalsifiable, and two platforms can both claim it while delivering data an order of magnitude apart in freshness.

How do I calculate whether my geofences will fire? Divide the time a vehicle spends inside the fence by the reporting interval. Dwell is twice the radius divided by approach speed, so a 200 meter fence at 40 km/h gives 36 seconds. Against a five-minute interval that is a 12% detection probability, and against 30 seconds it is certain.

Why do faster vehicles get missed more often? Because dwell inside the fence falls as speed rises. At a 200 meter radius, a vehicle at 15 km/h is inside for 96 seconds while one at 60 km/h is inside for 24, so on the same feed the slower approach is detected four times as often. Highway-speed arrivals fail hardest.

Can I just make the geofence bigger? Only up to a point. Reliable detection at a five-minute interval and 40 km/h needs a radius of about 1.58 km, but urban stops sit 150 to 400 meters apart, so a fence that large triggers on several stops at once and cannot attribute the event. Interval, radius and stop spacing are one constraint system.

Is ELD data good enough for arrival detection? No, and it was never meant to be. Under 49 CFR 395.26 an ELD records an intermediate location every 60 minutes in motion, and during authorized personal use at a resolution the rule describes as roughly a 10-mile radius. That is complete for hours-of-service enforcement and unusable for stop-level triggering.

What is the rest of the latency budget? Five more terms after detection: carrier platform ingest, your poll of their API, rule evaluation, notification send and the customer opening the message. A worked example totals 6.8 minutes, of which the feed interval and your polling cadence are 270 of 405 seconds, while rule evaluation is 10.

How do I measure missed detections? Reconcile geofence entry events against confirmed stop completions from the driver app or proof of delivery. Any completion without a matching entry is a missed detection, and the ratio is a direct read on whether your feed supports the automation built on it.