The blind spot - what nobody measures, and why a wrong figure costs more than a missing one
The most expensive state a figure can be in is not missing. It is wrong - and believed.
A missing figure has an advantage that is rarely noticed: everyone knows they don't have it. Decisions are then made cautiously, with a margin, with a second opinion. A wrong figure, on the other hand, carries decisions, gets copied into reports, ends up in the budget, and is never questioned again. It doesn't cost once; it costs every time someone builds on it.
This article takes stock of a series. Nineteen articles so far have dealt with how to measure people in spaces and what can go wrong along the way. This one is about the question that comes before: what does it cost not to measure - and what does it cost to measure wrongly?
Three states, three costs
| State | What happens | What it costs |
|---|---|---|
| Not measured | Decisions from experience, assumption, one-off observation | opportunities nobody sees, and mistakes nobody notices |
| Measured wrongly | Decisions from a figure that looks plausible and is not true | everything above, plus the trust the figure was given |
| Measured and labelled | Decisions from a figure whose state is visible | the measurement itself |
The middle row is the most dangerous, and it is more common than you might think. A sensor that half fails doesn't produce an empty day but a quiet one - which then enters the statistics as a slump. A display that shows zero where there is no data looks like an empty room. Both are covered in detail in The half outage and Occupancy is a balance, not a count. The rule behind them is simple: what has not been measured must look like it - as a dash, as “unknown”, as “stale”. Never as a flattering zero.
Where bad data comes from
Bad data has a small number of recurring causes, and those can be checked:
- Unreliable capture - a method that counts two people walking side by side as one.
- No audit - nobody checks whether a figure is still right.
- Duplicated, missing and obsolete data - the three faults that turn up in any data store.
Each of these has its own article in this series: the capture method in Which technology counts people how accurately?, double counts in From movement to figure, missing days in The half outage, stale displays in Occupancy is a balance, not a count.
That leads to a guideline: perhaps the only thing worse than leaving data out of decisions is relying on inaccurate data.
Two places where data fails
A common dictionary definition describes data as “factual information … used as a basis for reasoning”. That contains two conditions, and each can fail independently of the other.
| Condition | Question | Where it is decided |
|---|---|---|
| Factual | Is the figure right at all? | at the sensor: capture method, mounting, counting line |
| Used | Does it ever become a decision? | in the platform: storage, comparability, thresholds, alerts, access |
The first condition is the familiar one. Accuracy claims, technology comparisons and tenders revolve almost entirely around it. The second fails more quietly: a correct figure that nobody can retrieve, compare or understand changes nothing. It sits in a database, and the decision is still made on gut feeling.
The split is also the most precise description of how sensor and analytics platform work together. The sensor delivers the fact. The platform makes sure it gets used - and that it keeps its state while it does.
Where the blind spot is
Every sector has its own place where decisions are made without anything being measured. The series went through them one by one:
| Place | What is known | What is missing | Article |
|---|---|---|---|
| Shop | revenue and receipts | the majority who buy nothing | The visitors who don't buy |
| Counter and till | complaints | the people who join the queue and leave again | Managing queues |
| Office | leased space and bookings | the space actually used | Occupied versus paid-for space |
| Public transport | samples and ticket sales | the load on every trip | Passenger counting in public transport |
| Museum | admissions | the impact of an exhibition | Measuring museums |
| Event | the limit in the permit | proof that it was kept | Capacity as a permit condition |
The pattern is the same everywhere. What is known is whatever another system records anyway: the till, the booking, the ticket. What is unknown is what happens in the space. And that is exactly where most of the costs and most of the opportunities arise.
Why waiting is a decision too
The most common response to all this is not “no” but “later”. The investment seems high, the benefit uncertain, and things have worked without it so far. The sober view, though, is this: companies that improve their decisions with data will keep reaping the efficiency gains, while those that hold off because of the cost or doubts about the benefit risk falling behind.
The blind spot has long stopped being a technology question. Sensors count, platforms calculate, privacy can be solved - that is covered in Privacy is an architecture decision. What remains is a decision. And the next question is no longer whether to measure, but with what.
How to start small
Deciding to measure doesn't mean measuring everything at once. A start that works has four steps:
- A question, not a programme. “Do we need a second till on Saturdays?” is a better starting point than “We want to become data-driven”.
- The entrances first. A reliable visitor figure is the foundation for everything else - occupancy, comparisons, denominators.
- A baseline period. Measure for a few weeks before changing anything. Without a before, there is no after.
- The state of every figure visible. From day one, distinguish what is measured, what is estimated and what is not measured - otherwise the blind spot has only moved, not gone.
What belongs in the tender
Nineteen articles boil down to the questions every system should answer before it is bought:
- Which capture method is used, and how is its accuracy checked at your own site?
- What does a figure that hasn't been measured look like - a dash, “unknown” or zero?
- How does the system recognise a sensor that half fails, and what happens to those days?
- How is a message prevented from being counted twice, or a count from being lost?
- Which denominator is used when sites, days or periods are compared?
- What data leaves the sensor, and what of it is stored, and for how long?
- How do the figures get into your own systems, and how do third-party counts get in?
How we do it
The ANALYSIT Counting System is the second half of the split. The fact comes from the 3D sensor. ACS makes sure it gets used: storage on a one-minute grid for 730 days, aggregation in the site's own time zone, comparability across sites and periods, thresholds, alerts over five channels, reports and an open REST API.
Throughout, every figure keeps its state. What hasn't been measured renders as a dash, never as a flattering zero. A sensor that half fails is detected against the site's own history, and the excluded days are shown with the result. A display without fresh data says “stale” rather than “empty”, and a waiting time visibly carries whether it was measured or estimated.
If you want to know where the blind spot is in your operation, talk to us. One question is usually enough to find it.