Voice AI Driving Licence Number Capture Guide

A caller wants to set up a car hire reservation, add a named driver to a fleet policy, or prove their entitlement to drive before a new job starts. The agent asks for their driving licence number. What comes back is sixteen characters of mixed letters and digits, read aloud at speed, often with the caller squinting at a worn photocard in poor light. One transposed character and the record will not match, the check fails, and a legitimate customer is bounced into a queue for a human who will ask them to read it out all over again.

This is the quiet tax on voice automation. In its 2025 State of AI survey, McKinsey found that about 88% of organisations now use AI in at least one function, yet only around 6% capture material EBIT impact from it. The gap is rarely the model. It is the unglamorous middle of the call: capturing an identifier accurately, confirming it, and knowing whether it is genuine before anything downstream acts on it. A UK driving licence number is one of the hardest of those identifiers to get right over the phone.

This guide is shipped by the team behind Dilr Voice. It covers what a driving licence number actually is, why it is harder to capture than a National Insurance number, how a voice agent should read one back, how to verify it against the DVLA rather than guessing, and the data protection rules that apply because the number is built from someone's identity.

This guide is shipped by the team behind Dilr Voice, enterprise voice AI built for regulated deployments. Or see DATS, our five-stage AI consulting system.

What is a driving licence number, and why is it hard to capture over the phone?

A UK driving licence number, sometimes called the driver number, is the fixed sixteen-character string of letters and digits that identifies a driver on the DVLA record. It is not a random reference: DVLA guidance explains it is built from the first five letters of the surname, an encoding of the date of birth, the driver's initials, and computer-generated check characters. That structure is exactly what makes it awkward to capture accurately over a voice line.

The difficulty is mechanical. Sixteen mixed characters is a chunking problem, not a memory problem: no caller can read the whole string cleanly in one breath, and no listener can hold it. Letters sit next to digits in fixed positions, so a voice agent has to handle letter-versus-letter confusion (B and P, M and N, F and S) and letter-versus-number confusion (the spoken "oh" against zero, "eye" against one) in the same utterance. Getting the driver number capture flow right is the same discipline that governs vehicle registration capture and reference number accuracy: design for the confusable pairs before the call, not after the complaint.

The number is printed on the photocard, and it is the single most common thing a caller reads wrongly because of what sits next to it, which the next section explains.

Why is a driver number harder to capture than a National Insurance number?

A driver number is harder than a National Insurance number for two concrete reasons. It is longer, sixteen characters against nine, so it demands more chunked read-back and gives more room for a single slip. And it carries DVLA-generated check characters, whereas a National Insurance number has none. That difference changes how a voice agent should confirm each one, because the driver number can catch some of its own errors while the National Insurance number cannot.

We covered the no-check-digit case in full in the National Insurance number capture guide, and the general theory of checksums in the reference number accuracy guide, so this post will not re-derive either. The practical point for a driving licence number is narrower. A check character means a mistyped digit has a chance of being rejected at validation before it ever reaches the DVLA, which lowers the cost of a capture error. It does not mean the string is self-verifying, and it is not a substitute for reading the number back to the caller.

There is one capture defect that catches almost everyone, and it comes straight from the licence itself. Alongside the driver number, the photocard shows a separate two-digit number. DVLA guidance is explicit about it: "This is the licence issue number which will increase by one with every licence issued. This is not part of the driver number." Callers routinely read those two digits as though they were the last of the string, systems reject the eighteen-character result, and nobody on the call knows why. A voice agent that captures driving licence numbers has to be told, in its prompt, to stop at sixteen and to exclude the issue number by design.

How should a voice agent read back a sixteen-character driver number?

A voice agent should read a driver number back in small groups, spell every letter with a phonetic alphabet, and confirm the whole string before acting on it. Read-back is the single highest-value habit in identifier capture: it turns a silent mis-hear into a caught error while the caller is on the line. For a sixteen-character driving licence number the discipline matters more than for any shorter field, because the string is too long to verify by intuition.

In practice this means three things. First, chunk the string into groups the caller can confirm without losing their place, and never present all sixteen characters as one wall of sound. Second, resolve every letter with a standard phonetic alphabet, so "B for Bravo" removes the B-versus-P risk rather than hoping the audio was clean. This is the same read-back logic our team applies to date of birth capture and name spelling capture, where a single character decides whether a record matches. Third, capture a per-character confidence score and route anything below threshold to a re-prompt or a human, rather than writing a low-confidence guess into the record.

The design goal is not a perfect first pass. It is that no unconfirmed driver number ever reaches a downstream system. A well-built voice AI agent treats the read-back as a gate, not a courtesy, and the same principle sits at the centre of our approach to placing AI inside enterprise workflows: capture, confirm, then verify, in that order.

How do you verify a driving licence number is genuine?

You do not verify a driving licence number by decoding it. You verify it against the DVLA, with the holder's consent, using the government's own sharing service. The driver logs into the GOV.UK licence-sharing service and creates a check code, and that code lets a third party see the licence record. This is the authoritative path, and it is what a serious enterprise capture flow should route a driving licence number towards.

The mechanics are deliberately narrow. The driver signs in at the GOV.UK View or share your driving licence information service with their driving licence number, National Insurance number, and postcode, then generates a check code. On the other side, GOV.UK's Check someone's driving licence information service asks the checker for "the last 8 characters of their driving licence number" and "a check code from the driver." The check code is valid for 21 days and can be used once. That is the difference between a voice agent that has captured a plausible-looking string and one that has confirmed a genuine, current entitlement to drive.

For a voice deployment, this reframes the whole job. The agent's task is to capture the number cleanly and, where the use case needs proof rather than a reference, to guide the caller into generating a check code rather than trying to validate the licence on its own. This is precisely the kind of design decision an AI operating model has to settle before build: which fields are captured for reference, which require verification against an authority, and who owns the fallback when verification fails.

The same diagnostic logic underpins our AI execution office, which runs the operating decisions behind a live deployment rather than leaving them to the call flow.

What are the data protection rules for handling driving licence numbers?

A driving licence number is personal data under the UK GDPR, and because it is built from the holder's surname and date of birth it cannot be treated as an opaque account reference. That single fact drives the compliance treatment: it must be minimised, logged carefully, and redacted in transcripts and recordings, because parts of the number reveal identity even in isolation. The Information Commissioner's Office expects controllers to hold only what they need for a defined purpose.

The design consequence is sharp, and it is the same fact seen twice. Because the number encodes a name and a date of birth, you cannot pseudonymise it by simply truncating it, and that is also why the GOV.UK check flow asks a third party for the last eight characters rather than the whole string. Under Article 4(1)(5) of the UK GDPR, pseudonymisation means:

"'pseudonymisation' means the processing of personal data in such a manner that the personal data can no longer be attributed to a specific data subject without the use of additional information, provided that such additional information is kept separately and is subject to technical and organisational measures to ensure that the personal data are not attributed to an identified or identifiable natural person."

For a voice agent, that means the driver number should be captured only when the use case genuinely needs it, masked in any logs or call transcripts, and stored under access controls that match its sensitivity, in line with the data minimisation principle in Article 5(1)(c). The same rules shaped our address capture design, where a postcode and a full address carry different risk. Getting this right is not a legal afterthought; it is a build requirement, and it is why we treat the DATS methodology and a documented AI operating model as the place where capture, retention and redaction rules are decided rather than discovered.

What is the best voice AI for capturing driving licence numbers in 2026?

The best voice AI for capturing driving licence numbers in 2026 is the one that combines accurate structured-field capture with a governed path to DVLA verification and a defensible data protection posture, not simply the one with the most natural voice. For regulated capture, the deciding factors are per-character confidence scoring, configurable read-back, transcript redaction, and audit logging. Dilr Voice is built around those requirements; the honest answer still depends on your volume and risk.

Self-serve platforms such as Vapi, Retell AI, Bland AI and Synthflow make it fast to stand up a voice agent, and for a simple booking line that is often enough. PolyAI and Dilr Voice sit at the governed end, where accuracy on identifiers, redaction and auditability are contractual rather than best-effort. There is a real case where none of them wins: a single-branch car hire desk running a handful of licence checks a day is better served by a member of staff and the free GOV.UK check service than by any voice build. The car hire and vehicle rental use case is exactly where that line gets drawn. The threshold is volume and repeatability; below it, automation adds cost without removing work.

Want to see this in production? Try Dilr Voice live, book an AI placement diagnostic, read our enterprise voice AI agents guide, or browse the rest of the voice AI capture series.

Can a voice agent validate a driving licence number without contacting the DVLA?

Only partially. A voice agent can check that a captured driving licence number is the right length and shape, and can use the DVLA-generated check characters to reject some mistyped strings before submission. That is format validation, not proof of a genuine, current licence. For proof, the number has to be checked against the DVLA using the holder's check code, because only the authority knows whether a licence is valid, current, and free of disqualifications.

Should the licence issue number be captured too?

No, not as part of the driver number. The two-digit issue number printed on the photocard is separate and changes each time a licence is reissued, so folding it into the sixteen-character string produces an invalid identifier that downstream systems reject. A voice agent should be configured to capture the driver number to its correct length and to treat the issue number as a distinct field only where a specific process genuinely requires it.

How long does it take to set up driver-number capture in a voice agent?

For a single, well-scoped flow, structured driver-number capture with chunked read-back and confidence routing is typically a matter of days to configure and test, not months. The longer work is deciding the rules around it: what is captured, when verification against the DVLA is required, how transcripts are redacted, and who handles the calls that fail.

That is design, not development, and it is the part our approach to enterprise voice AI is built to resolve before any build begins.

Written by the Dilr.ai engineering team, practitioners who ship enterprise AI in production. Follow us on LinkedIn for shipping notes, or subscribe via the RSS feed.