As AI takes on a larger role within the supply chain industry, the concept of “garbage in, garbage out” should be at front of mind for anyone looking to expand their visibility or risk management capabilities. If either your data or processing methods are unreliable, any outputs will be equally so. Below are some important considerations to ensure you are getting the most out of your IoT devices.

Are Your Geofences Too Precise?

Once you have GPS location data reporting to your platform on a regular basis, you will probably look to begin using geofences to more easily understand where your shipments are or even to send notifications. It’s important to understand the limitations inherent to your devices when setting up a process like this. Modern IoT trackers aim to locate your shipments with an error range of a few hundred meters up to a few kilometers. To put this in perspective, a typical 3PL warehouse in the United States is around 16 square kilometers- about 125 meters wide. If you take the time to draw precise geofences around the perimeter of a waypoint or destination, you may be missing out on more than half of your at-location pings. In the below example, only 34% of pings at this warehouse were captured by a very precise geofence across many shipments.

Devices often promise location data with margins of error exceeding 1000 meters.

If your device’s configured reporting rate is forgiving enough, you may be notified on the next ping 15 minutes later. However, many devices are configured to rates of a ping every 2 to 4 hours (or even more). This often results in an average of 10 or more hours between pings when connectivity issues are factored in. It could be upwards of half a day before you are actually notified of its arrival. By this time, the device may have already been offloaded and turned off. So much for real-time visibility!

Consider your use case carefully and ask a few questions when deciding how to construct geofences:

• How accurate are my devices, and how often are they reporting? Can I afford to incorrectly classify a ping or two?
• How big is the location? Is it a warehouse that is only a few hundred meters wide, or is it a 10-kilometer-spanning international airport?
• Are there other waypoints nearby to avoid overlap with? Watch out for warehouses neighboring your tarmacs.

If your reporting configurations are less frequent and you value being notified sooner rather than later, a more generous geofence is probably a safer bet. For the above stop, a simple circular geofence of 5 to 10 kilometers is probably appropriate. If your use case is expanded, such as measuring verified route compliance, performing risk optimizations, or simply data analysis at scale, improperly labeling your IoT data can severely hinder your results.

Battery Life — Data Granularity Trade-off

As a data scientist working in logistics, the golden egg of IoT data utility is a frequent and reliable reporting configuration. A 15 minute ping rate means virtually every stop made and temperature excursion experienced will be accounted for. Even with connectivity limitations, you are unlikely to experience beyond an hour without visibility. Automating assessment of route compliance and precise dwell-time analyses become infinitely more tractable. On the other hand, a 4 hour ping rate with gaps that become half a day can turn headaches into impossibilities. Your shipments blip across countries and you are fortunate to get a single ping at each waypoint leaving you with no understanding of how long the shipment spent where. If you’ve also sent passive loggers, you’ll understand temperature history but with no geographical context.

It is difficult to overstate the benefit that frequent device reporting can have when it comes to operationalizing that data. If you are looking to turn visibility into actionable decisions both for live shipments and high level strategies, a strong data foundation is the single most important component. If you are looking to enjoy the fruits of the exponential growth in GenAI but don’t know how to start, begin by ensuring your processes are producing the highest quality data.

With all of this in mind, data quality shouldn’t outweigh data completeness. If your devices are running out of battery 2/3 of the way through the journey, you’ve now turned your last mile or segment into a costly blind spot. Unfortunately, as you can see in the charts above and below, there is often a direct correlation between ping rate and battery longevity. The first two lanes had devices configured at a 30 minute to 1 hour ping rate, while the latter two were configured to between 2 and 4 hours.

The problem is even more complex considering ping rate reliability often decreases as battery life does:

When performance degradation occurs, it is typically around three quarters of the way through a device’s battery life. However, some devices are less prone to these types of issues than others, so consider investigating the ping rate and battery life trade-off when performing pilots with device manufacturers. While ping rate does have an effect on battery loss rate, the vast majority of these shipments were arriving with a minimum of a quarter of their battery life remaining after journeys up to a month or more. If your lead times are shorter than this, it would likely be a valuable endeavor to optimize your ping rate configurations on those lanes. A 4 hour ping rate on a 5 day journey is throwing away insights you have already invested in.

One approach is to first understand your specific device’s battery performance at various reporting configurations and aim to have your devices retain at least a quarter of their total battery life by the end of the journey while accounting for the typical performance of your logistics providers. For best insights and modeling results, a good starting point would be to aim for 30 minutes to one hour in between pings and no more than two.

Common Pitfalls in IoT Data

Sufficient ping rates and appropriate geofences will form a strong basis for visibility. However, there are other considerations that can further improve your data foundation and build it toward enabling, supporting, and eventually owning strategic decisions.

Incomplete Shipment Coverage

Though it may be a difficult process change to implement, ensuring the device reports once at each of the origin and destination will minimize the amount of assumptions that any system must make when it comes to understanding lane data. If there are no pings for the last 8 hours of a trip, there are limited options as far as imputing the remaining portion and any of these is likely to skip potential stops or underestimate the total duration of a shipment’s journey. Additionally, many instances of overcooling and other temperature issues occur at either the origin or destination of a journey. Lacking this data provides an incomplete profile of a temperature-sensitive lane, impeding RCA from the start. Additionally, devices not being turned off once they’ve reached the destination can make preprocessing more involved.

Though it has already been addressed, ping rate is also key here. Identifying waypoints such as air and sea ports, customs, logistics handoffs and more depends on a reliably reporting device. A single ping is not enough to establish a waypoint as there is no indication of speed. Two pings can only establish a dwell time as precise as the reporting interval (times three), and that assumes no missed pings, which are more common at many warehouses. A rigorous data preprocessing engine and strong statistical modeling can work with these limitations. However, to guarantee the best insights possible, aim for your data to represent every portion of the journey — including the origin and destination.

Automating Airport and Transit Mode Inference

In order to best understand the routes your shipments are taking, you will need at minimum pre-configured geofences. Given the prevalence of unapproved routes, unplanned stops, and shipments forced to reroute, geofences configured specifically for expected travel will always fail to provide complete visibility. A thorough visibility or risk mitigation system should always understand your most problematic shipments as well as the OTIF ones.

Once these waypoints are known, the mode of transport can also bring a new layer of context, exhibiting varying temperature regulation limitations, lead times, and data availability. Certain segments (movement between one waypoint and another) can exhibit multiple modes of transport, often a combination of truck or sea and air. Even simple rules such as “movement between airports must be flight” fail. Often, the locations of logistics hubs and nearby tarmacs are difficult to distinguish and nearby telemetry data overlaps. Many shipments have been observed traveling between airports via truck, such as with neighboring locations like Dubai (IATA DXB) and Al Maktoum (DWC).

Speed Calculations

Inferring transit modality is a complex problem that must take advantage of all available information. One might consider what the stops are (airports, seaports, etc), how far are apart they are, or travel times from device data. This proxy for device speed is a powerful indicator, given cargo planes fly around 15 times faster than a truck, traffic considered. However, speed calculations are delicate problems. Speed as reported by devices can be suspect and is often altogether lacked. Here are some general concerns when processing device data, especially for speed calculations:

1. Slow ping rates can easily exaggerate transit times and dwell times.

If your device with a 2 hour reporting rate pings at each of two stops, a 30 minute drive: the transit time is only known to be less than 2 hours, and the dwell time is known to be less than 4. When you consider skipped intervals and rates north of 4 hours, the problem becomes significant, offering little confidence in the execution of your contracted lead times.

2. Fast ping rates with telemetry deviations create unexpected values.

If your device pings every 5 minutes, and in two consecutive pings it reports both 500 m east and west of its true location, it has gone from a full stop to a slow-moving traffic pace (12 km/h.) In worse cases, such as if your data originates in multiple devices within the same truck, you may have short intervals of time and report air speeds or impossible values. These can easily become outliers that confuse any analysis.

3. You won’t know when it’s (completely) stopped.

Due to the deviations in location referenced above, you will likely never see perfectly stationary data. Use thresholds to understand when your data is actually stopped, and don’t let these cases skew your understanding of travel speed for that segment!

As a final concern for inferring transit modes, speed can be impossible to calculate for air segments due to their tendency (and in some cases legal requirement) to suppress the recording of location data. A gap in the reporting, especially with slow ping rates as mentioned above, can be difficult to understand. If you miss the first ping at the destination of a 2 hour flight, you may be lead to believe it was a 10 hour + drive without further study.

Know Your Devices

Consider these things when choosing a device manufacturer.

• What reporting rates are their devices capable of, and for how long?
• What precision is expected, for both location and temperature if applicable?

Understanding the limitations of the device will allow you to configure more appropriate geofences and perform data analysis more confidently. If you are investing in an active IoT use case for your business, these simple investigations and process changes could multiply the value you extract.