The Economics of Facial Recognition

Facial recognition is sold at two very different prices depending on what you are asking it to do. This is a look at why, what the market actually charges, and what it means if all you need to know is who turned up for their shift.

What the market charges

Biometric identity verification is normally sold per check. You send a face, you get an answer, you pay for the answer. The published rates as at September 2026 look roughly like this:

These vendors all publish in US dollars. The figures above are converted at roughly £0.74 to the dollar as at September 2026, so treat them as indicative rather than quotable. They are also per verification, on identity verification platforms. Not subscriptions, and not the same thing we sell.

Those prices are not wrong

It is worth being clear about this, because the easy version of this article is to hold up a dollar figure next to £2 and declare victory. That would be dishonest and anyone in the industry would see it in about four seconds.

When a KYC provider charges a pound a check, you are not paying a pound for a face match. You are paying for document forensics across ten thousand document types, liveness and deepfake detection, sanctions and PEP screening, a human review queue for the edge cases, a tamper-proof audit trail, and a vendor who will stand behind the result when a regulator asks. If you are onboarding a customer into a regulated financial product, opening a gambling account, or running a right to work check, that is exactly what you need and that is a fair price for it. We would tell you to go and buy it.

We are not in that market and we are not trying to be.

The job we are actually doing

Alpaka Presence answers one question: is this the person whose staff record this is?

The people being checked are already known to you. They are enrolled, they are on your payroll, they were on the rota, and they are standing at the door of a building you own. There is no document to authenticate, no watchlist to screen, no identity to establish. It has already been established. All that is left is to confirm the same person came back today, and to write that to a timesheet.

That is a genuinely smaller problem, and it deserves a smaller price. This is what we mean when we say we bring the technology downstream: the same underlying capability, applied to a task with a lower burden of proof, at a cost that matches the task rather than the category it came from.

The volume problem

The other reason per check pricing does not fit is the shape of usage.

Take a care home with 50 staff. Everyone clocks in at the start, out for break, back from break, and out at the end. Four checks per person per shift. Twenty shifts a month each.

That is 4,000 checks a month. Around 48,000 a year, from one site, with fifty people.

Metered at 59p that is £2,360 a month. At £1.37 it is £5,480. Those numbers are not a fair comparison to anything, because nobody sells KYC to a care home for clocking in, and we are not suggesting they do. They are there to make a narrower point: workforce verification is high frequency by nature, and any pricing model built around occasional, one-off checks breaks when you point it at a rota.

On Alpaka, 50 people on the Presence plan is £100 a month. Not per check, not per site, not per terminal. If those fifty people clock in four times a day, or six, or twice, the bill is £100. There is no meter, so there is nothing to manage, nothing to forecast, and nothing that gets more expensive because your team started taking their breaks properly.

What the technology actually costs

Here is the part the industry does not usually say out loud. The raw compute for a face match is close to free.

Amazon's Rekognition prices face comparison at around a tenth of a penny per image in its first tier, and its dedicated liveness check at just over a penny. Those 4,000 monthly clock-ins, run as raw face comparisons, cost under £3.

So the gap between a tenth of a penny and a pound is not the technology. It is the compliance apparatus, the document estate, the human reviewers, and the liability wrapped around it. All of which is real, all of which has to be paid for, and almost none of which is relevant to a support worker starting a shift.

We are not undercutting anyone. We are just not carrying that overhead, and we are not charging you for it.

We should also be straight that we are not alone in offering face-based clocking at a low price. Some time tracking tools include it cheaply or free, and that is fine - the check was never where the value was going to come from.

Verification is the cheap part

If facial recognition were the whole product, the economics would be interesting and not much else. The check is the beginning of the process, not the end of it.

Every clock-in writes to a timesheet. Every timesheet reconciles against the rota you already built, at the pay grades you already configured. Exceptions surface as exceptions: the late start, the missed clock-out, the shift nobody covered. Absence and leave sit in the same record, so the hours that were not worked are accounted for alongside the hours that were. Compliance and training expiries hang off the same person, so the qualification that lapsed last Tuesday is visible before someone works a shift they should not have.

At the end of the period you export it. Sage, Wagestream and LevelFT, and a range of other payroll-ready formats, on every plan.

That is the actual saving, and it is not measured in pence per check. It is measured in the days a manager currently spends chasing paper timesheets, decoding handwriting, arguing about a Tuesday in March, and re-keying the result into a payroll system. Payroll stops being a job. It becomes a download of data that assembled itself while people got on with their work.

A per-check identity verification platform will never do this, and it is not supposed to. It verifies and hands back a result. What happens next is your problem.

Two things worth saying about consent

Biometric clocking is not a decision to take casually, and we would rather flag this than let you find out later.

Facial recognition is never the only way to clock in on Alpaka. A member of staff who does not want to use it can use a PIN or a QR badge on the same device, and they are paid from the same record. That matters practically, and it matters for your consent position under UK GDPR, because consent that carries no genuine alternative is not consent.

And if you use biometric data to identify people, you need a Data Protection Impact Assessment. That obligation is yours as the controller, not ours. We do not currently ship a template for it, but if you get in touch we will tell you plainly where that stands and what we can offer.

On our side, the biometric template and the employee's identity are held in separate systems. Alpaka's database holds a reference, not a face. Full detail is on our security and data page.