Exception triage
Ranks flagged sessions by how much they matter — a five-minute fence miss and a full duplicated shift are not the same problem — so review time goes where the money is.
Zepth Node · Workforce
Attendance & Time Tracking
Attendance you can trust — verified at the boundary, by the face, on the site, not on an honour system.
Last updated
Zepth Node module
Attendance & Time
AI agent built into the module
Geofenced clock-inOn-device face verificationMulti-site field attendanceWork-hours policies
2.7 billion
deskless workers worldwide — roughly 80% of the global workforce
Emergence Capital, The State of Technology for Deskless Workers
The share of workers whose day never touches a desk is the reason attendance is a field problem before it is an HR one. Software written for the seated minority tends to assume a browser, a fixed location and a stable connection — three assumptions a site crew breaks before lunch.
1.5–5%
of gross payroll lost to time theft each year
American Payroll Association
A range, not a point estimate, and it varies enormously by how attendance is captured. The figure is worth quoting for its floor: even the low end is a larger number than most attendance systems cost.
2.2%
of gross payroll lost specifically to buddy punching
Nucleus Research
Buddy punching is the failure mode an honour-system register cannot detect by design — the record is correct in form and false in fact. Verification at the moment of capture is the only place it can be caught.
6–9 min
of payroll staff time to process a single manual timesheet
American Payroll Association
Collection, entry, verification and chasing the missing ones. The cost scales with headcount and pay frequency, which is why it stays invisible until the workforce grows.
Overview
A paper register and a WhatsApp check-in cannot tell you who was really there. They record an assertion, and an assertion is what you end up paying. Zepth Node captures attendance where the work happens, verifies it at the moment of capture, and sends clean hours into payroll without a re-keying step in between.
Every session is its own record: where it was opened, whether the device sat inside the site boundary, whether the face matched, and who approved the correction if one was needed. The hours that reach payroll are the hours somebody can account for.
Two questions, one record
Attendance answers two questions that are not the same question. Who is here, and who worked how long. The first is answered in seconds or not at all. The second is answered within a payroll cycle, and has to survive being disputed.
A system that answers only the second is a timesheet. A system that answers only the first is a headcount. Node keeps both off one set of verified sessions, because the moment they are maintained separately they disagree — and the disagreement always surfaces at the worst time, in front of the person whose pay is short.
Why verification has to happen at capture
Every attendance dispute is a dispute about a record that was created without evidence. Once the punch exists as a bare timestamp, no amount of downstream review can establish who made it or where they stood. The evidence has to be attached at the moment of capture, or it does not exist.
That is what the geofence and the face check are for. Neither is surveillance and neither decides anything: together they turn a claim into a record with provenance, so the argument a month later is about a fact rather than a memory.
Location, at capture. The session records whether the device was inside the site boundary. Outside is not blocked — it is flagged, because a legitimate reason usually exists and a human should hear it.
Identity, at capture. The face check runs on the device, under consent, and attaches a match signal to the session. It is evidence for a reviewer, not a gate on someone’s wages.
Correction, on the record. Regularization creates a new, approved entry rather than overwriting the original. What was captured, what was changed and who decided all remain visible.
The handoff into payroll
Most payroll errors are not arithmetic. EY put the error rate on a traditional payroll process near 20%, with time and attendance the most common category — and each correction costing an average of $291 to put right. Those errors overwhelmingly enter at a boundary: hours leave one system as a report and arrive in another as typing.
Node removes the boundary. Verified sessions, approved leave and computed overtime are already on the record payroll runs from, so there is no export, no re-keying, and no window in which the two systems can disagree.
The value
Why it matters
Clean, verified hours flow straight into payroll — no register to reconcile at month end.
A clock-in from outside the boundary is flagged for a human, not silently paid.
Field crews moving between sites are captured per site, per session, on the same record.
Corrections are requested, approved and logged — so the audit trail survives the argument.
Capabilities
What you can do
01
Geofenced clock-in
Define the site boundary once. Employees clock in inside it; a punch from outside the fence is captured and flagged for review rather than quietly counted.
02
On-device face verification
A face check confirms the right person is clocking in. It runs on the device, is gated by the employee’s consent, and produces a risk signal a human reviews — never a silent lockout.
03
Multi-site field attendance
Built for crews who move. Several sites in one day, each session captured separately with its own location, so hours can be costed to the right place.
04
Work-hours policies
Shifts, expected hours, break rules and overtime thresholds defined once and applied automatically to every session.
05
Regularization & approvals
A missed or mistaken punch is corrected by request, not by editing history. The employee raises it, the manager decides, and both sit in the audit trail.
06
Live presence
Who is on the clock, and where, read from live sessions rather than a roll-call in a group chat.
The workflow
How it actually runs
- 1
Define the boundary and the policy
Draw the geofence for each site, set shifts, expected hours and overtime thresholds, and publish the regularization window.
- 2
Capture the session
The employee clocks in on their device. Location and, with consent, a face check are attached to the session as it is created.
- 3
Flag the exceptions
Punches outside the fence, verification misses and impossible shifts are surfaced with their evidence for a manager to review.
- 4
Correct on the record
Missed or wrong punches are regularized by request and approval. The original entry stays visible beneath the correction.
- 5
Release the hours
Approved sessions become the hours payroll runs from — no export, no re-keying, no reconciliation step.
AI that does the work
How AI changes Attendance & Time management.
Pattern detection
Surfaces repetition a single session cannot show: the same pair of employees always clocking within seconds of each other, corrections clustered on one approver, a site whose exception rate moved.
Drafted, never decided
Every finding arrives as a draft with its evidence attached, for a manager to accept or dismiss. No agent adjusts hours and no agent touches pay.
The engineer’s judgment stays in charge; the AI removes the latency and the blind spots.
Best practices
- Set the geofence to the working boundary, not the property line — a fence drawn too generously verifies nothing.
- Publish the regularization window with the policy. A correction route nobody knows about becomes an argument at payroll.
- Review flagged punches weekly, not at month end. A flag investigated four weeks late is a flag nobody can explain.
- Treat face verification as a signal for a human, never an automatic denial of pay. A false negative that locks someone out of their wages is a worse failure than the one it prevents.
Dashboards & reporting
Dashboards cover live presence by site, hours by employee and project, exception rates by team and approver, and the reconciliation between captured hours and the hours that reached payroll. Every figure drills through to the sessions behind it.
Live dashboards
Drill-down & filters
Export to Excel / PDF
FAQ
Common questions
What happens when someone clocks in outside the geofence?
The session is captured and flagged, not rejected. There is usually a real reason — an early delivery at another gate, a site boundary redrawn last week — and a manager reviews the flag with the location evidence attached. Blocking the punch outright would only push the correction into a WhatsApp message nobody can audit.
Is face verification stored as a biometric record?
The check runs on the device and is gated by the employee’s explicit consent. What reaches the record is a match signal against the session, which a human reviews. Consent can be withdrawn, and the disclosure and the consent both sit on the record.
Can one person work across several sites in a day?
Yes. Each site visit is its own session with its own location and duration, so hours can be costed to the right project rather than lumped into a single daily total.
How are missed punches handled?
Through regularization. The employee raises a correction request, the manager approves or rejects it, and the approved entry sits alongside the original rather than replacing it. Both remain in the audit trail.
Does attendance data reach payroll automatically?
Yes — approved sessions, leave and overtime are already on the record payroll runs from. There is no export step and no re-keying, which is where time and attendance errors usually enter.
Does this work without a connection on site?
Sessions are captured on the device and synced when the connection returns, with the original capture time and location preserved. A dead spot delays the record; it does not rewrite it.
Sources
- Emergence Capital — The State of Technology for Deskless Workers
- EY — Cost and Risks Due to Payroll Errors (2022)
- American Payroll Association — time theft and manual timesheet processing
- Nucleus Research — buddy punching as a share of gross payroll
Related modules
Zepth is the construction project delivery platform — it runs construction, procurement and asset management on one record, and does the work: reading the drawings, reviewing the submittals, matching the invoices and flagging the risks, with a human sign-off on anything consequential.
See it on your project.
A short, tailored walkthrough on your real workflow — no generic demo.