A contract cleaning company with a small guarding division. Four hundred staff, 60 client sites across two provinces, shifts running from 05:00 to 22:00 and a night crew in three office parks. Head office is eleven people, one of whom ran payroll.
Attendance arrived as photographs of paper registers, sent by supervisors on WhatsApp, usually late. The payroll clerk retyped 400 rows every fortnight. Payday produced a queue of wage disputes, and clients periodically queried hours they were billed for. Neither argument could be settled with evidence.
Distributed workforces are the hardest environment for this kind of system, and the reason is that none of the usual assumptions hold.
Why a distributed workforce is different
- There is no turnstile, and the premises are not yours. You are installing on a client's property, under their rules, and they can withdraw permission.
- Connectivity varies from good to nothing, sometimes at the same site depending on where the device is mounted.
- Not everybody has a smartphone, and building a system that assumes they do excludes exactly the staff whose hours are most disputed.
- The same hours feed two different arguments. They pay the worker and they bill the client. When those two figures come from different sources, both are eventually wrong.
- Buddy clocking is the default failure, which is why biometrics get proposed, and biometrics bring their own duties under POPIA.
- Supervisors are the system. They allocate, they cover absences, they mediate disputes. Any design that treats them as a data-entry step will fail.
The wage query queue on payday is not an admin problem. It is the sound of a business with no agreed record of who worked, and it costs more in trust than it does in hours.
One workforce, three clocking methods
The most important decision was refusing to standardise. Sites were tiered, and each tier got the method that actually fits it.
| Site tier | Profile | Method | Why |
|---|---|---|---|
| Tier 1 (9 sites) | 20+ staff, permanent client premises, power and connectivity | Wall-mounted biometric reader with battery backup | High headcount justifies the hardware; ends buddy clocking outright |
| Tier 2 (26 sites) | 4–20 staff, stable but no space for a device | Supervisor's phone app: staff clocked in with a photo, geofenced | One device per site, no client infrastructure required |
| Tier 3 (25 sites) | 1–3 staff, often a single cleaner | Fixed QR code on site, scanned by the worker's own phone, or an offline PIN fallback | Cheap, no supervisor present, works where nothing else does |
| All tiers | Exceptions | Supervisor override with a reason code, reviewed daily | Real life happens. Unrecorded exceptions become disputes. |
Every method feeds the same record. The device varies; the data does not. That is what makes payroll and client billing come from one place.
Biometrics carry duties, not just convenience
Under POPIA, biometric information is special personal information, which means you need a lawful basis for processing it, a clear purpose, real security around the template, and honest communication with staff. Practically: explain what is stored and what is not, store templates rather than images, restrict access, set a retention period, and give staff a non-biometric route if they object. Doing this properly is a fortnight of work. Skipping it is a liability that surfaces at the worst possible moment.
Twelve weeks, in order
| Weeks | Phase | What actually happened |
|---|---|---|
| 1–2 | Site survey | All 60 sites assessed for power, connectivity, client permission and headcount. Nine needed a client conversation before anything could be mounted. |
| 2 | Rules, written down | Rounding, grace periods, overtime, Sunday and public holiday rates, and what counts as a no-show. Previously custom rather than policy. |
| 3–5 | Configure | Rosters per site, pay rules coded once, exception workflow, payroll export mapping, client billing report. |
| 4–5 | Staff communication | Briefings in the languages spoken on site, a one-page explainer, and the POPIA notice for biometric sites. Before any device appeared. |
| 6–7 | Pilot, one region | Eight sites across all three tiers. Two weeks of parallel paper running. |
| 8–11 | Roll out by region | Rather than by size. A region at a time meant one trip covered installation, enrolment and training. |
| 12 | First clean payroll | Run from the system, with the paper register available and, in the end, not needed. |
Supervisors were the entire project
Head office wanted a system. Staff wanted their correct wages. Supervisors, who had been quietly absorbing the difference between those two things for years, were the group with the most to lose from accuracy and the most to gain from it, depending on the supervisor.
What worked: bringing them in during the rules phase, not after. Several rules, particularly around grace periods and travel between sites, were changed because a supervisor explained why the proposed version was unworkable. They also got the exception tool early, so their first experience of the system was being given authority rather than having it removed.
What was necessary: an unambiguous message from the managing director that hours worked would be paid, that the system existed to prove hours rather than to trim them, and that anyone finding themselves paid less than they expected in the first month should raise it and be paid the difference while it was investigated. Three people did. All three were resolved within a week, two in the worker's favour, and that fortnight decided whether the rollout was trusted.
What it cost
| Line | Once-off | Notes |
|---|---|---|
| 9 biometric terminals, installed | R147,600 | Including backup power and mounting on client premises |
| 26 supervisor devices | R93,450 | Mid-range phones, company owned |
| Configuration and pay rules | R134,200 | Rosters, overtime, exceptions, client billing report |
| Payroll integration | R57,400 | Export mapped and reconciled over two parallel runs |
| Rollout, travel and enrolment | R118,900 | Sixty sites across two provinces. The line most often underestimated. |
| Communication and training | R43,700 | Multilingual briefings, printed explainers, POPIA notices |
| Total once-off | R595,250 | Over twelve weeks |
| Running cost | R16,740 / month | Per-employee licensing, device data, support |
For scale: per-employee licensing typically runs R18 to R45 a month. A single-site business with one terminal spends well under R50,000. Travel and enrolment scale with sites, not headcount, which is why distributed rollouts cost more than their size suggests.
What changed, measured
- Payroll preparation went from three days to four hours, and the payroll clerk stopped retyping anything.
- Wage queries fell by 78%, and the ones that remain are resolved from a record rather than an argument.
- Client billing disputes became evidence-based. Two contracts were adjusted upward when the data showed consistent cover above the contracted hours, which had been absorbed for years.
- Overtime spend dropped by 12%, not by cutting hours but by making unplanned overtime visible to the person who could plan around it.
- Chronic understaffing at four sites became undeniable, which changed the staffing model rather than continuing to lean on the same three willing people.
- Attendance records became defensible, which matters the first time a dismissal or a client dispute reaches a forum where "we are sure he was late" is not evidence.
What we would do differently
We would write the pay rules before shortlisting products. Two of the three shortlisted systems could not express one of this company's overtime arrangements without a workaround, and we found that out in week five.
We would roll out strictly by geography from the start. The original plan sequenced by site size, which meant three trips to the same town.
And we would not put a biometric terminal at the two sites where connectivity was marginal. They were the only sites that generated support calls for the first two months, and a phone-based method would have worked immediately.
If payday is an argument
Start by writing the rules down. Rounding, grace, overtime, travel between sites, what counts as present. Most businesses discover they have three versions of these, and no system can resolve a policy question you have not answered. Then tier your sites and pick the method per tier. One device standard across a distributed workforce is the most expensive kind of tidiness. The same principle applies wherever a workforce outgrows a spreadsheet.
