Expert guide · Manual vs automated
Manual vs automated people counting: accuracy, cost, and how to audit a counter (2026)
A person with a clicker undercounts by 8–25% and costs more in a week than most sensors cost outright. Here is the evidence, the 2026 wage arithmetic, the seven things a machine does that a person cannot, the two jobs manual counting still does well — and the 30-minute video audit that ends every argument about whether the counter is right.
Published by SMS Storetraffic. We sell automated counters, so we would say this. The accuracy study and the wage data are not ours; both are linked. Judge the argument on them.
Two sensors the size of a phone replace a person at the door — PEARL, $499.95
The short answer: automated counting is more accurate, far cheaper and the only method that produces data by the hour for every day of the year. Manual counting is right for exactly two jobs — a short video audit to check a sensor, and a one-off count somewhere nothing can be installed. Everything in between is a person doing badly, at $17 an hour, what a $500 device does well for years.
AccuracyManual counts with sheets or clickers undercounted by 8–25% against video in the one peer-reviewed comparison we know of. Good sensors count 95–99%, and do it the same way every day.
CostOne person at one door for a 62-hour week costs about $1,056 in wages at the U.S. median retail wage. A wireless beam counter costs $499.95 once; a 3D sensor $1,149.95. Both are cheaper than the first two weeks of counting by hand.
CoverageA sensor counts every hour you are open, every door, all year. A person counts the hours you can afford, at one door, on the days you remember.
TrustThe right way to check a sensor is not a clicker but a recorded video with the count lines overlaid, at peak, for 15–30 minutes. It settles the question with evidence either side can replay.
Terms
What counts as manual, semi-automated and automated counting
Definitions
Manual people counting is a person observing an entrance and recording each arrival by hand — a tally sheet, a mechanical or digital clicker, a phone app. Semi-automated counting uses a sensor that counts by itself but leaves the data handling to a person: a beam counter with a display that someone reads and resets. Automated people counting is a sensor that detects, counts and transmits to software with no human in the loop, producing timestamped counts for every hour the door is open.
The distinction that matters is not whether a machine is involved but whether a human attention span is. Manual and semi-automated counting both depend on someone remembering, watching and writing things down; automated counting does not. Every problem below traces back to that.
Evidence
How accurate is manual people counting?
Less than people think, and always in the same direction. The best evidence we know of is a 2007 study by the University of California’s Safe Transportation Research and Education Center, published in Transportation Research Record. Researchers counted pedestrians at San Francisco intersections three ways at once — tally sheets, clickers and video — and treated the video count as the truth. Manual counts with either sheets or clickers systematically undercounted, by 8% to 25%, and the error was worst at the beginning and end of each session, which the authors attributed to unfamiliarity and then fatigue (Diogenes, Greene-Roesel, Arnold & Ragland, TRR 2002, 2007).
That matches twenty years of watching people count at shop doors. Three scenes, all real:
The furniture-store receptionist
Responsible for the tally sheet. Two people walked in; she wrote them down. A minute later she turned to the card terminal to process a payment, and two more walked in behind her. She never noticed. Nobody who has ever staffed a front desk will find that surprising.
The two kinds of clicker face
The new counter’s eyes dart across every group — did I get that right? The experienced counter clicks with no expression at all, eyes wandering, because they have worked out that nobody will ever recount the group to prove them wrong. One is stressed and inaccurate; the other is relaxed and inaccurate.
The instruction to count less
A store manager under pressure about a “low conversion rate” told the door counter to under-count, so that when the numbers were questioned the machine — not the team — would take the blame. A human counter has interests. A sensor has none.
The library sector documents the same effect at national scale. U.S. public libraries report “visits” annually to IMLS, and where continuous counting is absent a “typical week” count may be multiplied by 52 to estimate the year. IMLS itself has acknowledged that reported visit counts can be inaccurate because of limited counting technology and unrepresentative sample weeks — a week with a snowstorm, a renovation or a summer programme carries its distortion through the whole annual figure (our IMLS guide).
And the deeper problem with manual counting is not the error rate but the hours it never covers. Even a perfect human count of Saturday 10 to 4 says nothing about Tuesday evening, and it is Tuesday evening — the hour where conversion halves because one associate is alone with a queue — that the data was supposed to find.
Why the machine wins
Seven things an automated counter does that a person cannot
We first wrote this list in 2018 after a new retail customer emailed to say their VP of Sales would audit several stores by hand between 9 and 3 and compare the totals to the dashboard — “a major black eye to the project if the numbers don’t match”. The numbers did not match, and the machine was right. It usually is, for these reasons:
It does not get bored, tired or distracted.
No daydreaming after a late night, no sick child at home, no phone in the other hand. The 2007 study’s finding that error rises with fatigue simply does not apply.
It makes no judgement calls outside its rules.
“Did I already count those two? I blanked for a second. I’ll add two.” A sensor applies the same rule to every crossing, so its errors are consistent — and consistency is what makes trends trustworthy.
It takes no breaks.
No lunch, no bathroom, no shift change, no gap while the counter helps a customer find the changing rooms. The count runs from open to close, every day, including the days nobody thought to count.
It does not talk to customers or colleagues.
A person at the door is, inevitably, a greeter, a direction-giver and a conversation. Each conversation is a gap in the count.
It has no bias and no interests.
It does not know the store is under pressure about conversion, does not round up on a good day, and cannot be asked to count a little lower.
It filters what a person cannot.
Height filtering to exclude children and carts; U-turn logic to ignore the customer who steps in and straight out; staff exclusion by a tag the employee carries. A person can attempt these rules; a sensor applies them identically 40,000 times a year.
It timestamps everything.
The value of a count is knowing when. A sensor gives you every 15 minutes of every day against transactions and staff hours; a tally sheet gives you a total for the hours someone stood there.
Arithmetic
The 2026 cost math: a person at the door vs a sensor
Use the median wage for the people who would actually be asked to count. In the United States that is a retail salesperson at $17.03 an hour (U.S. Bureau of Labor Statistics, May 2025). In Canada the federal minimum wage is $18.15 from April 2026 (Government of Canada) and Ontario’s general rate is $17.60, rising to $17.95 in October 2026. A typical store trades about 62 hours a week. Wages only — no payroll taxes, benefits, management time or the sales the counter is not making while counting.
| Method | Upfront | Per week (one door, 62 hours) | Per year | Hours covered | Accuracy |
|---|---|---|---|---|---|
| Person with a clicker, U.S. median retail wage | $0 | $1,056 | $54,900 | Only the hours paid for | Undercounts 8–25%, varies by person and hour |
| Person with a clicker, Canadian federal minimum wage | $0 | C$1,125 | C$58,500 | Only the hours paid for | Same |
| Person counting one sample week a year, extrapolated × 52 | $0 | $1,056 once | $1,056 | One week in 52 | Undercount plus whatever that week was not typical of |
| PEARL wireless beam counter Ours | $499.95 | $0 on the free plan; ~$8 on the Retail plan | ~$10 batteries; $0–$419 software | Every open hour, all year | Upwards of 98% on adult single-file traffic; consistent |
| 3D Scope II LC stereo counter Ours | $1,149.95 + cable run if needed | $0–$12 software | $0–$623 software | Every open hour, all year | 99% rated; 95% floor guaranteed 2 years |
Weekly figures: 62 h × $17.03 = $1,055.86; 62 h × C$18.15 = C$1,125.30. Software: our Retail plan is $40.95 per device per month, Real-Time $51.95; annual figures are the monthly rate × 12. Other vendors’ published prices are in the nine-system comparison.
29 h
of door-wages pays for a PEARL
$499.95 ÷ $17.03. Less than half a trading week.
68 h
of door-wages pays for a 3D Scope II
$1,149.95 ÷ $17.03. About one trading week; add a few hours for the cable run.
52×
more hours covered than a sample week
A sensor’s year of hourly data costs less than the one manual week it replaces.
In 2016 we wrote that “$2,000 is a reasonable starting rate for a good solution and its related expenses”. Ten years later the sensor costs a quarter to a half of that, the software can be free, and the wage has gone up. The case was already lopsided; it is now not a case at all. The only remaining reason to count by hand is that the alternative is unfamiliar — which, outside mid-size and large retail, it still often is.
Fair is fair
When manual counting still makes sense
Yes
Auditing a sensor — from video
The best use of a human counter is to check the machine, and the best way is to count a recording rather than a live door: you can pause, rewind and recount the moment two people walked in abreast. The protocol is below. This is the one form of manual counting that is more accurate than the sensor, because it can take as long as it needs.
Yes
A one-off count where nothing can be installed
A single-day pop-up, a street festival gate, a temporary exhibition with no power and no mounting point. Count by hand, accept the undercount, write down the hours covered. (Even here a battery beam on a pair of stands often works; a trade-show exhibitor used a 3D sensor on a 13-foot pole to measure its booth’s stop rate — case study.)
Sometimes
Classifying what a sensor cannot
If the question is “how many visitors were tour groups” or “how many customers asked for a fitting”, a person is the only instrument. Use the sensor for the count and the person for the classification, over a sample period, and do not confuse the two.
No
Routine traffic measurement, occupancy control or annual reporting
Anything that needs every hour, every door, or a defensible annual figure. This is where manual counting fails on accuracy, coverage and cost at once, and where a library’s “typical week × 52” becomes a number nobody can stand behind.
The middle ground
Tally counters with a display, and occupancy apps
Between a clicker and a connected sensor sit two things that are better than a person and worse than automation.
A beam counter with an LCD — the sub-$100 devices on Amazon and eBay — counts far more accurately than a person, but you are the software. Someone has to read the number at close, write it down, reset the device, type it into a spreadsheet and build the formulas; by the time you want hourly data or a comparison with sales, the routine has collapsed. We have tested several: some count only in the scenarios the box does not mention (a beam that the door itself breaks when it opens, sometimes twice), support is thin, and most owners quietly stop after a few months. They are a fair way to discover that counting is useful; they are not a way to keep doing it.
An occupancy app on a phone is the right tool when a person must be at the door anyway — capacity-limited events, a controlled entrance during a busy period. Plus and minus buttons, a maximum threshold, a live percentage the whole team can see. Ours is free on iOS and Android. It removes the arithmetic, not the attention problem, so it belongs on the door only for the hours a person is there regardless.
Settling the argument
How to audit an automated people counter
You should expect an automated counter to be accurate, which is exactly why validating it is reasonable — and why doing it badly is so common. The VP who audits from a chair with a clicker between 9 and 3 is comparing the machine to the less reliable instrument, and when the two disagree, the machine takes the blame. Do it this way instead:
Agree the rules before you count.
Are children counted? Staff? Someone who steps in and straight out? A person standing in the doorway? Two counters can both be “98% accurate” against different definitions. Write down which rules the sensor is configured for and audit against those.
Record video with the count lines overlaid.
A good sensor does this itself: the 3D Scope II LC records a clip with the lines and the running count drawn on the frame. Otherwise place a camera looking straight down at the line. Never audit from a side angle where people overlap.
Choose a peak window of 15–30 minutes.
Peak, because that is when groups, carts and abreast crossings happen and when a counter earns its keep. Our auditing policy uses 30-minute recordings; an hour is more than enough. If the window turns out quiet, record another.
Count the video, pausing and rewinding.
Two people, ideally, each counting independently, then reconciled. The whole point of video is that the human count can take as long as it needs; a moment of doubt is replayed, not guessed.
Compare against the same window, to the minute.
Pull the sensor’s count for exactly that period, in the same direction (IN, OUT or both) as you counted. Time-zone and boundary mistakes account for a remarkable share of “the counter is wrong” tickets.
Judge against the guaranteed floor, then fix rather than argue.
Inside the guaranteed range — 95% for our 3D Scope II, for two years — the counter is doing its job. Outside it, the cause is almost always a placement or calibration issue: a display on the line, a door in the zone, a height setting. That is a minor adjustment, made remotely, and the recording proves whether it worked.
What a fair audit avoidsCounting live with a clicker while the sensor counts video. Counting from a vantage point where people overlap. Counting a couple as one “shopper” while the sensor counts two people. Forgetting that staff are excluded, or that children are. Comparing a 9-to-3 clicker total against a 9-to-3 dashboard total that includes the greeter’s 300 crossings. Each of these has produced a furious phone call to our support desk; none was a counting error.
The first week
“The counter must be wrong”: the objections everyone raises after installing one
Everyone is a mix of excited and afraid to learn their real conversion rate, and most people are disappointed by the first number. The next stage, reliably, is a list of reasons the count must be too high. The reasons are real; the conclusion does not follow.
“My customers shop in couples and families, so my conversion rate is much higher than this.”
True, some of them do, and the counter counts people, not shopping parties. This does not change the decision. Your conversion rate is a baseline: what matters is whether it moves from 12% to 13% after you change staffing or layout, and a consistent people-count measures that perfectly. “Group counting” features that merge people who arrive close together raise the number but rest on a proximity rule that cannot know who is shopping with whom; we advise against them.
“It counts my staff, who go in and out all day.”
Then exclude them properly rather than mentally. Options in order of accuracy: a separate staff door that is not counted; a 3D sensor with ultra-wideband staff exclusion, where employees carry a small tag the sensor ignores without identifying anyone; or a consistent daily deduction. Jewellers, boutiques and showrooms with few customers and many staff crossings need the tag-based method for the conversion rate to mean anything.
“It counts delivery drivers, the postman and staff’s partners.”
A handful of crossings a day against hundreds of customers — a small, consistent error that does not change any decision built on the trend. If deliveries come through the customer door at a predictable time, note it; if they are numerous, give them a back door.
“It counts children and strollers.”
Only if you asked it to. Beam counters mounted at 54 inches do not see children or carts; overhead 3D sensors filter by measured height and can report adults and children separately. Libraries and museums usually want children counted; shops usually do not. Set the rule, then audit against it.
“We had 25 people ‘in the store’ at closing time according to the occupancy display.”
That is arithmetic, not a fault. Occupancy is IN minus OUT, and every counting error persists; a sensor that is 95% accurate at a door seeing 100 people an hour drifts by around five an hour. Occupancy systems need the most accurate sensor on every entrance, a daily reset and a correction button. Traffic analysis does not have this problem, because each hour stands alone.
Practical
Making the switch from manual to automated counting
Match the sensor to the door, not the budget.
Standard outward-opening or open door with mostly single-file traffic: a wireless infrared beam (ours is the PEARL, $499.95, no cable). Busy, wide, glass-fronted or automatic entrance: an overhead 3D sensor (ours is the 3D Scope II LC, one PoE cable). The technology guide covers every option, including the ones we do not sell.
Decide the counting rules once.
Adults only or everyone; IN only or both directions; staff excluded how. Write them down. Every later comparison depends on the rules staying fixed.
Run both methods for one week, then stop the manual one.
If you have historical manual data, one overlapping week tells you the ratio between the two so old and new figures can be compared. Expect the sensor to count more.
Audit once from video, then trust the trend.
Do the 30-minute audit above in the first month while calibration support is free. After that, watch for the alerts a good system sends when a sensor goes offline or is blocked — a silent sensor is the real risk, not a slightly wrong one.
Put transactions next to traffic and look at it weekly.
Conversion by hour and by day is where the money is. Free software tiers show hourly traffic; retail tiers pull transactions from the point of sale, or take a daily figure typed in by hand, and show conversion and the key selling periods. A counter nobody looks at has an ROI of zero.
Common questions
Manual vs automated counting: questions people ask
How accurate is manual people counting with a clicker?
In the one peer-reviewed comparison we know of, manual counts with tally sheets or clickers systematically undercounted pedestrians by 8–25% against video of the same crossings, with error highest at the start and end of each session (Diogenes et al., Transportation Research Record, 2007). At a shop door the causes are fatigue, distraction, conversations, breaks and, occasionally, an interest in the result. Good automated counters count 95–99% and apply the same rules every day.
Is it cheaper to count people manually or with a sensor?
A sensor, by a wide margin. One person at one door for a 62-hour week costs about $1,056 in wages at the U.S. median retail wage of $17.03 an hour (BLS, May 2025), or about C$1,125 at Canada’s federal minimum wage. A wireless beam counter costs $499.95 once with software from $0; a 3D stereo counter $1,149.95. Both are paid for by less than two weeks of counting by hand, and they count every hour of the year.
What is the difference between manual and automated people counting?
Manual counting is a person recording arrivals by hand with a sheet, clicker or app; it covers only the hours paid for and undercounts. Semi-automated counting uses a sensor with a display that a person reads and resets. Automated counting is a sensor that detects, counts and sends timestamped totals to software with nobody in the loop, for every hour a door is open, and can filter children, U-turns and staff by rule.
When does manual people counting still make sense?
For two jobs: auditing a sensor by counting a recorded video, where a person can pause and rewind and is therefore more accurate than any live method; and a one-off count where nothing can be installed, such as a single-day event with no power or mounting point. It also helps for classifying visitors in ways a sensor cannot, over a sample period. It is the wrong tool for routine traffic measurement, occupancy control or annual reporting.
How do I check if my people counter is accurate?
Record 15–30 minutes of video at a peak period with the count line visible — good sensors record this themselves with the lines and running count overlaid — then count the video, pausing and rewinding, and compare with the sensor’s total for exactly the same window and direction. Agree the counting rules first (children, staff, U-turns). Judge the result against the vendor’s guaranteed floor; a shortfall is almost always a placement or calibration fix.
Why does my people counter show more visitors than my staff counted?
Usually because the staff count is low, not because the sensor is high. Manual counts miss 8–25% of crossings through fatigue, distraction and breaks. Check three other things before blaming the sensor: whether the comparison covers exactly the same minutes and direction, whether staff and children are excluded in one count and not the other, and whether a display or door is sitting in the count zone. Then settle it with a video audit.
Are cheap tally counters with a display worth it?
As a way to discover that counting is useful, yes; as a way to keep doing it, rarely. They count more accurately than a person, but someone must read, reset and transcribe the number every day and build the spreadsheet, and most owners stop within months. They also work only in the scenarios the box does not mention. A connected counter with a free software tier costs a few hundred dollars more and removes the routine entirely.
Can a people counter exclude employees from the count?
Yes, three ways. A separate staff door that is not counted; an overhead 3D sensor with ultra-wideband staff exclusion, where employees carry a small tag the sensor ignores without identifying any individual; or a consistent daily deduction. For stores with few customers and many staff crossings, such as jewellers and showrooms, tag-based exclusion is what makes the conversion rate meaningful.
How long does an automated people counter take to pay for itself?
Against the wages of a person counting, under half a trading week for a $499.95 wireless beam counter and about one week for a $1,149.95 3D sensor at $17.03 an hour. Against the sales it helps create, take your weekly traffic, conversion rate and average sale and ask what one extra sale per hundred visitors is worth per year; for most stores with a few hundred visitors a week that alone repays the system within months, provided someone acts on the data.
Do libraries have to count visitors automatically?
No. U.S. public libraries report annual visits to the IMLS Public Libraries Survey, and where continuous counts are absent a typical-week count may be multiplied by 52. IMLS itself has noted that such figures can be inaccurate because of limited counting technology and unrepresentative sample weeks. An automated counter is not required, but it is the difference between an estimate and a defensible annual figure for boards and funders.
Since 1972
Replace the clicker
PEARL — the counter for a standard door Ours
Two battery sensors, peel-and-stick, on your 2.4 GHz Wi-Fi. Counts in and out every hour you are open, alerts you if it is blocked or the batteries run low, and shows the numbers in a free app. $499.95, free express shipping to the U.S. and Canada, two-year warranty, 30-day money-back. For busy, wide or glass entrances, the 3D Scope II LC at $1,149.95 comes with a video audit you can run yourself and a 95% accuracy floor guaranteed for two years.
Check our work
Sources
Accuracy
- Diogenes, M. C., Greene-Roesel, R., Arnold, L. S. & Ragland, D. R. (2007), Pedestrian Counting Methods at Intersections: A Comparative Study, Transportation Research Record 2002, pp. 26–30, UC Berkeley SafeTREC — manual sheet and clicker counts undercounted by 8–25% vs video
- SMS Storetraffic — counter auditing policy (30-minute video recordings; free within the calibration warranty)
- SMS Storetraffic — IMLS visitor counting requirements (typical-week sampling and IMLS’s accuracy caveat)
Wages
- U.S. Bureau of Labor Statistics — Retail Sales Workers, median $17.03/hour, May 2025
- Government of Canada — federal minimum wage $18.15 from April 1, 2026
- Littler — Canadian minimum wage increases in 2026 (Ontario $17.60 → $17.95)
Prices
How this guide was made
This page updates “The Differences Between Manual and Automated People Counting”, which we published in 2018 after a customer’s hand audit went the way hand audits go. The stories are real and the seven reasons are unchanged; the accuracy evidence, the wage figures and the audit protocol are new. SMS Storetraffic sells automated counters and says so throughout. Corrections via our contact page.