Seven findings
- Punches arrive in seconds, unless the terminal is offline. 94.6% reached the server within a minute and 90% within 11.1 seconds. The 4.6% that took more than 15 minutes were not slow, they were stored: they arrived together in 1,370 catch-up bursts when the terminal reconnected.
- Outages are common and concentrated. On 13.8% of terminal-days a punch arrived over 15 minutes late. 82 of 280 terminals never had such a day in four weeks; 29 had eight or more, and five terminals produced 42% of every late punch.
- Most outages are short; one in ten runs past a day and a half. On a typical outage day the longest delay was 5 hours 15 minutes. On one outage day in ten it was more than 40.2 hours.
- Face leads, fingerprint holds, card is growing. Face was 53.3% of terminal punches and 61.9% of staff punches. Fingerprint is still 41.3%. Card rose from 1.3% of punches in May and June to 4.8% in September.
- Terminal clocks are rarely wrong, and then by hours. 2 of 281 terminals had a clock off by more than a few minutes: one 2.5 hours fast, one 11 hours behind. None of the 260 eSSL terminals did.
- Staff rarely press the out key. 22.9% of staff working days had a single scan, and the terminal's out key was used on 9.0%. Half of all staff days had four or more scans.
- Load is modest. The median terminal recorded 104 punches on a working day, and even the busiest terminal's busiest minute was 12 scans, one every five seconds.
What to do with this before you run payroll
- Do not close the month the moment it ends. Nine in ten outage days were resolved within 40.2 hours, so wait two days after month-end, or first check that every terminal has reported since. AionHRMS shows a terminal as offline after ten minutes of silence, with the last time it was heard from.
- Fix the few terminals that cause most of the trouble. Late punches cluster on a handful of machines. Start with what is local to them, the power supply and the network, before doubting the whole system.
- Do not rely on the out key. Pair the first scan of a shift with the last, and decide in writing what a single-scan day means. AionHRMS counts it as present and prints "No out punch", leaving duration at zero rather than inventing one.
- Check the terminal's clock after a power cut or a firmware change. A clock hours off moves every late mark and can put a punch on the wrong day.
- Choose the capture method for your workforce, not the trend. Face and fingerprint both stay common. Cards and PINs are the methods a colleague can borrow.
For the step-by-step checks on a terminal that has gone quiet, see biometric machine showing offline.
How long a punch takes to reach the server
Every punch carries two times: the one the terminal stamps when a person scans, and the one our server records when the punch arrives. The gap between them is the delivery delay. Across 9,62,730 punches from 280 terminals, the median delay was 4.3 seconds. 94.6% took under a minute; the chart shows the rest.
- 1 to 15 minutes0.85%
- 15 to 60 minutes0.66%
- 1 to 6 hours1.43%
- 6 to 24 hours1.36%
- 1 to 3 days0.65%
- More than 3 days0.48%
Late punches arrive in bursts. When a terminal comes back online it sends everything it stored: the typical burst was 6 punches, one in ten bursts carried more than 85, and the largest carried 652. A burst that size can be several days of a terminal's attendance appearing at once, which is why a register that looked short in the evening can be complete the next morning.
How often a terminal goes offline
We call a terminal-day an outage day when at least one punch from that day arrived more than 15 minutes late. Of 6,901 terminal-days, 953 were outage days: 13.8%. The spread between terminals is the real finding.
- No outage day82 terminals (29%)
- 1 to 2 days98 terminals (35%)
- 3 to 7 days71 terminals (25%)
- 8 or more days29 terminals (10%)
198 terminals sent at least one late punch, but the five worst produced 42% of all of them, and three terminals produced 82% of the punches that were more than three days late. We did not measure causes; when a few terminals account for most of the delay, the place to start is what is local to them.
- Monday16.6%
- Tuesday14.6%
- Wednesday13.7%
- Thursday14.7%
- Friday12.2%
- Saturday13.0%
- Sunday11.0%
Mondays had the most outage days (16.6%) and Sundays the fewest (11.0%). We did not test why.
Face, fingerprint or card
Each punch records how the person was verified. Face leads, but fingerprint is far from finished, and staff use face more than members and visitors do.
- Face
- Fingerprint
- Card (RFID)
All terminal punches · 9,68,417 punches
Face 53.3% · Fingerprint 41.3% · Card (RFID) 5.4%
Staff · 1,91,607 punches
Face 61.9% · Fingerprint 34.9% · Card (RFID) 3.2%
Members and visitors · 7,76,810 punches
Face 51.2% · Fingerprint 42.9% · Card (RFID) 5.9%
By terminal, 132 recorded only fingerprint scans (with or without cards), 89 recorded only face, 44 recorded both, and 16 recorded other combinations. Across 2026 the split barely moved, except for card:
| Month | Face | Fingerprint | Card | Punches sampled |
|---|---|---|---|---|
| February | 53.9% | 44.0% | 2.1% | 36,425 |
| March | 52.3% | 45.7% | 2.0% | 50,729 |
| April | 49.8% | 48.5% | 1.7% | 48,034 |
| May | 54.5% | 44.2% | 1.3% | 68,118 |
| June | 55.5% | 43.3% | 1.3% | 72,756 |
| July | 54.8% | 42.8% | 2.3% | 86,302 |
| August | 51.4% | 44.9% | 3.7% | 86,209 |
| September | 54.0% | 41.2% | 4.8% | 1,03,996 |
For what each method is good at, see fingerprint vs face recognition attendance.
Terminal clocks
A slow clock looks exactly like a delay, so we flagged a terminal only when even its fastest punches arrived more than 15 minutes "late" (its clock is behind), or when any punch arrived before it was made (its clock is ahead). 2 of 281 terminals qualified, one 2.5 hours fast and one 11 hours behind, both on the Realtime protocol (2 of 18); none of the 260 eSSL terminals. Both were left out of the delay and outage figures above. Rare, but an 11-hour error turns a 9:00 arrival into an evening punch on the register.
How busy a terminal gets
On a day it was in use, the median terminal recorded 104 punches; one in ten recorded more than 303, and the busiest recorded 812. In its single busiest minute of the four weeks, the median terminal took 4 scans, nine in ten took 6 or fewer, and the busiest took 12, one scan every five seconds. If a queue forms at shift change, a second terminal at the same entrance shortens it more than a faster one.
How staff actually scan
This section covers only the 1,787 staff in the data, across 29,645 working days. They work at fitness and wellness centres, where split shifts and early starts are normal (41.6% of first scans were before 8:00), so read it as a frontline-shift benchmark rather than an office one.
- One scan22.9%
- Two scans16.5%
- Three scans10.1%
- Four or more50.5%
The terminal's own out key was pressed on 9.0% of staff-days, and 29.8% of staff-days ran more than ten hours from first scan to last, which at a fitness centre usually means a morning and an evening session. An attendance system that waits for people to press "out" would have no out-time on nine days in ten; one that pairs the first and last scan of each shift, and supports two shifts a day, does not have that problem. That is how AionHRMS turns scans into a day.
Method and limits
- Data
- Every punch received by the AIONDATA attendance platform with a timestamp from Monday 14 September to Sunday 11 October 2026, India Standard Time: 10,05,483 in all, of which 9,68,421 came from biometric terminals (33,587 were manual entries and are not studied). Read-only aggregate queries on the production database on 12 October 2026.
- Who is in it
- 281 terminals (260 eSSL on iClock/ADMS, 18 Realtime, 3 TIMY) at 269 workplaces: 268 fitness and wellness centres that run on AdviceFit, AIONDATA's software for fitness businesses, and 1 company on AionHRMS. 68,666 people scanned, 1,787 of them staff.
- Exclusions
- 4 punches older than their terminal's registration (log uploads on first connection, not outages), and the terminal whose clock was 11 hours behind, from the delay and outage figures. A re-scan by the same person within five minutes is dropped by the platform before it is stored, so repeat scans are not measured. Terminal models are not reported: the model field is a form default, not data the terminal sends.
- Sampling
- Everything is a full count of the four weeks except the monthly table, which uses a repeatable 10% block sample of 2026 terminal punches. January is left out because the database holds only its last days.
- Limits
- One month, one platform, mostly one industry. The study sees only punches that arrived, not any a terminal failed to store. With 18 Realtime and 3 TIMY terminals against 260 eSSL, brand-to-brand comparisons would not be fair, so we do not make them.
- Privacy
- Only aggregates left the database. No workplace, person or terminal is named or can be picked out.
Download and reuse
Every number on this page is in the CSV, with the base each percentage is calculated on. You may reuse it under CC BY 4.0, with a link to this page. Cite as: AionHRMS (AIONDATA Private Limited), "Biometric attendance machines in the field", 12 October 2026, aionhrms.com/biometric-attendance-field-data-india.
Questions this data answers
How often does a biometric attendance machine go offline?
In our data, on 13.8% of terminal-days (about one day in seven) at least one punch reached the server more than 15 minutes after it was made, which means the terminal had lost its connection for a while. It was uneven: 82 of 280 terminals had no such day in four weeks, while 29 had eight or more.
Are punches lost when the machine is offline?
A terminal keeps punches in its own memory while it is offline and sends them when the connection returns. In these four weeks, late punches arrived in 1,370 catch-up bursts, typically 6 punches at once and up to 652. This study can only see punches that arrived; it cannot count any a terminal failed to store.
Is face recognition replacing fingerprint attendance in India?
It leads but has not replaced it. Of 9,68,421 terminal punches, 53.3% were face, 41.3% fingerprint and 5.4% card. Face stayed between 50% and 56% of punches every month from February to September 2026, while card rose from 1.3% to 4.8%.
How many people can one biometric machine handle?
The median terminal recorded 104 punches on a working day and the busiest recorded 812. In its busiest minute of the four weeks, the median terminal took 4 scans and the busiest took 12, one scan every five seconds. If a queue forms at shift change, a second terminal at the same entrance shortens it more than a faster one.
Do employees punch out on biometric machines?
Not reliably. On 22.9% of 29,645 staff working days the terminal recorded a single scan, and the terminal's out key was used on only 9.0%. Attendance software should treat the first scan as in and the last as out, and have a written rule for days with one scan.