The metrics catalogue - every figure with its definition and prerequisite
A metric without a definition is an opinion with decimal places. “Dwell time”, “occupancy” or “capture rate” carry the same name at two vendors and often mean something different - and anyone who doesn't know that compares two figures that were never comparable.
This page is the reference section of the series. It doesn't explain anything a second time; for each metric it says in one line what it measures, what it requires and which article covers it in full. Whether you're writing a tender, checking a quote or can't make sense of a figure in a report, this is the place to start.
Three rules that apply to every metric
- Every figure has a denominator. A share without saying what it refers to can't be read. Why that matters for demographics and site comparisons is covered in What a counter can tell apart and Comparing sites.
- What hasn't been measured looks that way. A dash or “unknown”, never a flattering zero. Why is explained in The half outage.
- Measured and estimated are two states. An estimated figure visibly carries that origin. All three states - measured, estimated, not measured - are set out in Managing queues.
Counting and occupancy
| Metric | What it measures | What it requires | In full |
|---|---|---|---|
| Visitors | entries across a counting line, stored per minute | a counting line with a correctly named counting logic on the sensor | From movement to figure |
| Resolution | minute to year; the longer the period, the coarser the finest grid | none | Occupancy is a balance, not a count |
| Live occupancy | entries minus exits since the nightly reset, never below zero | a counting line that counts both directions | Occupancy is a balance, not a count |
| Utilisation and traffic light | occupancy divided by maximum occupancy; amber from 80, red from 100 per cent by default | a maintained maximum occupancy per site | Occupancy is a balance, not a count |
| Peak hour, day, week, month | the maximum of the plotted series, descriptive | none | Occupancy is a balance, not a count |
| Best time to visit | the quietest of the next twelve hours, from a weighted weekday average | about four weeks of history | Occupancy is a balance, not a count |
Zones and behaviour
| Metric | What it measures | What it requires | In full |
|---|---|---|---|
| Average dwell time per zone | total time spent divided by the number of visits, measured by the sensor on leaving the zone | zones on the sensor, zone analytics enabled | Measuring dwell time |
| Visits per zone | completed visits, i.e. exits | as above | Measuring dwell time |
| Four behaviour stages | per visit: under 5 seconds transit, under 30 seconds browser, under 2 minutes shopper, above that committed | as above | Measuring dwell time |
| Engagement rate | share of visits with at least 30 seconds in the zone | as above | Measuring dwell time |
| Capture rate | zone visits divided by site visitors; can exceed 100 per cent | as above | Measuring dwell time |
| Minimum sample | a zone is only rated once it has at least 5 visits and 5 per cent of all zone visits | applies to every zone comparison | Measuring dwell time |
| Live zone occupancy | entries minus exits per zone; “stale” after 3 minutes without data | capacity maintained per zone | Measuring dwell time |
| Paths | the most frequent zone sequences, entry and exit zones, share of paths with at most one zone | sensor analytics enabled | The visitors who don't buy |
Demographics, objects and waiting
| Metric | What it measures | What it requires | In full |
|---|---|---|---|
| Gender distribution | share of the gender-classified population, not of all visitors | add-on function on the sensor, mounting height 2.40 to 4.00 m | What a counter can tell apart |
| Age distribution | seven groups from 0-17 to 65 and over, as a distribution | add-on function and an age counter on the sensor | What a counter can tell apart |
| Objects | bicycles, prams, wheelchairs, shopping trolleys as sightings, not entries | add-on function, mounting height up to 3.50 m | What a counter can tell apart |
| Groups | an index of group presence per 100 visitors, can exceed 100 | group counting, part of the sensor firmware | What a counter can tell apart |
| Gaze direction and attention rate | direction profile with circular mean; viewers divided by visitors | add-on function, mounting height 2.40 to 4.00 m | What a counter can tell apart |
| Walking speed | calculated from position changes, three classes with boundaries at 0.5, 1.5 and 2.5 m/s | sensor analytics enabled | What a counter can tell apart |
| Body height and child share | five classes, boundaries changeable retrospectively | sensor analytics enabled | What a counter can tell apart |
| Waiting time | measured from observed waiting episodes or estimated from headcount and service time - with its origin visible | waiting and service zones configured | Managing queues |
| Throughput | exits from the queue zone per minute | queue analytics enabled | Queues in operation |
Statistics, forecasting and context
| Metric | What it measures | What it requires | In full |
|---|---|---|---|
| Correlation | Pearson r with an exact p-value; significance is the corrected q-value below 0.05 | at least 7 days with sensor data | When a deviation is really a signal |
| Anomaly | robust deviation from the median, from an absolute value of 2.5 or outside the interquartile fence | at least 14 measurable days per metric | When a deviation is really a signal |
| Visitor forecast | Holt-Winters with weekly seasonality, with an uncertainty band | at least 14 observed days | When a deviation is really a signal |
| Key drivers | which factors explain visitor numbers, as a ranking | 14 days computable, 28 usable, 60 solid | When a deviation is really a signal |
| Opening-hours analysis | each hour's contribution to the daily average | maintained opening hours, at least 3 days per weekday | When a deviation is really a signal |
| Weather correlation | the relationship between visitor numbers and temperature or rain | your own weather service key, coordinates, 7 usable days | Context: weather, holidays and staffing |
| Visitors per staff member | ratio of totals over the period, not the mean of daily values | headcount recorded per shift | Context: weather, holidays and staffing |
| Staffing need per hour | recommendation from the 75th percentile of footfall per weekday and hour | staff planning enabled | The visitors who don't buy |
Multiple sites and alerts
| Metric | What it measures | What it requires | In full |
|---|---|---|---|
| Coverage | days with data divided by calendar days | cluster analytics enabled | Comparing sites |
| Two denominators in a cluster | the tile counts calendar days; rating and ranking count reporting days | deliberately different | Comparing sites |
| Deviation from the cluster | site average against cluster average in per cent; percentiles from four sites | at least four reporting sites for percentiles | Comparing sites |
| Occupancy alert | fires when capacity is exceeded or a minimum staffing level is undercut | alert configuration per site | Alerting |
| Trend alert | deviation against the previous week, month or same week last year | at least 5 days with data per window, at least 50 visitors in the baseline | Alerting |
| Excluded days | days on which a sensor only partly delivered are shown with the result | applies to correlation, opening hours and weather | The half outage |
| API and webhooks | REST access with scope-limited keys, signed events with retries | API enabled | Open interfaces |
How to use this catalogue in a tender
The tables above can be carried straight into a requirements list. Three questions turn every row into a requirement you can check:
- How is the metric defined? Not the name but the formula and its denominator.
- What does it require? Licence, mounting height, configuration and a minimum number of days. Why that has to be settled before mounting is covered in Sensor planning.
- What does the system show while the requirement isn't met? A named empty state, or a figure that only looks like one. Which analysis holds from which day is covered in Rolling out a counting system.
How we do it
Every metric in the ANALYSIT Counting System has a formula you can look up and a named prerequisite, and both are in this overview. Where nothing was measured, ACS shows a dash rather than a flattering zero; below every minimum threshold a named empty state appears, and an estimated waiting time visibly carries ESTIMATED.
Every figure we show you is either measured and backed by its formula, or explicitly marked as your assumption. If you want to check a metric in your tender or in a quote, talk to us. We'll put the formula next to it.