InnerWeb
Skip to main content

Guide: how to read accuracy claims from a supplier

The phrase "average accuracy below one metre" sounds unambiguous, but it describes a single statistic from a single measurement. An IT engineer who is to compare offers should know what this number hides, which additional values to request and how to verify it in their own hall. Below is a step-by-step guide.

< 1 m
average BLE accuracy from many measurement points
Source: InnerWeb facts
95
facilities in the city project with accuracy below 1.5 m
Source: InnerWeb city project
10–20
control points in your own test
Source: InnerWeb recommendation

What the word average means

The average accuracy is the sum of the deviations from the control points divided by their number. For InnerWeb, the BLE location algorithm inside buildings reaches an average accuracy below 1 m across many measurement points. Such a number tells how the system behaves typically, but not how much it was wrong in the worst cases.

The average hides rare large deviations, because a few points with a 6 m error among a hundred points with a 0.5 m error still produce an average that looks good. For this reason, when you deal with a location supplier, ask for the median and the 95th percentile of the error.

Median and 95th percentile: two numbers instead of one

The median is the value below which half of the measurements fall, that is, a typical error that is resistant to extremes. The 95th percentile is the value below which 95 percent of the measurements fall, that is, an error that occurs once in every twenty readings. In applications for the safety of people, the 95th percentile is the one that matters, because an SOS alarm does not choose the moment when the largest error occurs.

So ask for three numbers: the average, the median and the 95th percentile, and additionally the maximum value with a description of the point where it occurred.

Example: the same average, different 95th percentile (m)

Example: the same average, different 95th percentile (m)Chart shows two suppliers with the same average and a different 95th percentile of the error.01234mSupplier A (Average): 0.9 m0.9 mSupplier B (Average): 0.9 m0.9 mAverageSupplier A (Median): 0.7 m0.7 mSupplier B (Median): 0.5 m0.5 mMedianSupplier A (95th percentile): 1.6 m1.6 mSupplier B (95th percentile): 3.8 m3.8 m95th percentile
  • Supplier A
  • Supplier B
Source: Illustrative example with assumed values, not a measurement result

Where and under what conditions it was measured

Ask about the area and type of the facility, the density of transmitters and the number of control points. A practical example: in a city project, 11,000 transmitters reached an accuracy below 1.5 m in 95 facilities with a combined area of 120,000 m². Such a description gives the scale, the number of facilities and the conditions on which the result is based, not just one number.

A measurement in an empty hall with a dense placement of transmitters does not transfer to a plant with metal racks and a density of 1 transmitter per 100–200 m². Ask for a description of the equipment and occupancy at the time of measurement.

It is good when the supplier can point to references from facilities of a similar character and agrees to a conversation with an end user, not only to a slide presentation.

Accuracy in the plane and on a floor are different questions

The question about the position on the plane of a hall does not settle whether the system will correctly indicate the storey. A position on a floor is not given in metres without a source: the function documentation rules out such claims, so treat an offer that contains such numbers without a description of the measurement with caution. For multi-storey buildings, ask separately about the correctness of floor recognition and about the accuracy within a floor.

How to verify the figure in your own hall

Prepare 10 to 20 control points with known coordinates: entrances, corners, passages, cabins, places next to machines. Walk past them with the wearable device at several times of day, record the indication on the map and calculate the distance from the real point. From the distances, calculate the average, the median and the 95th percentile. Also record the number of transmitters heard at each point.

If the results differ significantly from the declaration, ask the supplier about the density of transmitters at the point and about the positions recorded on the map: an incorrect transmitter position spoils the result more than the variability of the signal does.

Evidence for an audit and a tender

Write down the result of the test in a protocol: date, map, list of points, table of deviations, the three statistics and remarks on the conditions. Such a protocol is evidence for the auditor and the basis for a contractual clause, for example a threshold for the 95th percentile in critical zones. It is better to base the contract clause on an acceptance measurement than on a general wording from the offer.

Typical pitfalls in offers

The first pitfall is an accuracy given without a statistic: the word "accuracy" may mean the average, the median or the error at the best point. The second is a measurement in one room described as the result for the whole plant. The third is combining accuracy with a transmitter density assumption that is not part of the scope of supply. The fourth is a laboratory figure without a description of occupancy by people and equipment. For each offer, record in a table which statistic, under what conditions and at what density it refers to, and fill in the empty cells with a question to the supplier before comparing prices.

Keep the protocol together with the map in the version on which the measurement was made, because after a later correction of the transmitter positions the result will have to be repeated and compared with the previous one.

Video

Frequently asked questions

Is an average below one metre enough?

No, not as the only number. Add the median, the 95th percentile and a description of the measurement conditions.

How many control points are enough?

For a hall, 10 to 20 points give readable statistics; in critical zones, add points more densely.

Must the supplier provide the raw measurement data?

It is worth asking: from the raw deviations you can calculate the medians and percentiles yourself instead of relying on a single number from a brochure.

What if the test result is worse than the declaration?

Check the positions of the transmitters on the map and their density. An error in recording the position is a more frequent cause than the device itself.

Let's talk about location in your plant

We will show how a system based on many BLE transmitters behaves in a hall similar to yours. Book a conversation with the InnerWeb team.

Let's talk