DVLA adopts natural language processing for call handling


SOURCE: UKAUTHORITY.COM
APR 16, 2026

Mark Chillingworth @mchllingworth

Get UKAuthority News

3D illustration of human face and mouth and sound waves

Image source: istock.com/peterschreiber.media

The Driver and Vehicle Licensing Agency (DVLA) has overhauled its contact centre operations with the introduction of a natural language voice recognition system, reporting a 50% reduction in the time customers spend navigating phone menus

The new Interactive Voice Response (IVR) system replaces traditional touchtone menus, where callers previously navigated up to four levels of press button options. Under the new format, approximately 900,000 monthly callers are asked to describe their enquiry in their own words at the start of the call.

Natural Language Processing (NLP) technology, powered by Google DialogFlow and integrated via the Storm platform from customer service technology firm Content Guru, identifies the caller's intent so it can route them to the appropriate department. The system handles enquiries across several areas, including vehicle licensing, driving licences, and medical updates.

The DVLA stated that the transition has simplified the customer journey, reducing the average time spent in the IVR by 90 seconds. The agency also noted that the improved routing accuracy has helped automate 20,000 transfers every month and contributed to a 5% increase in customer satisfaction scores.

The technology is designed to provide automated responses for general enquiries or send links to digital services via SMS. However, the agency confirmed that the system does not access personal records or make significant decisions. If the software cannot confidently interpret a caller's request, or if record access is required, the call is automatically directed to a human advisor.

To develop the system, the DVLA analysed approximately 100,000 test calls to understand the vocabulary used by customers. The agency’s in-house administration team now monitors the system daily, reviewing logs of uncategorised intents to refine routing accuracy.

Data processed by the NLP engine is limited to the verbal intent provided by the caller. These logs are stored for 90 days before being deleted in accordance with security policies. A Data Protection Impact Assessment was most recently updated in January 2025 to ensure ongoing compliance.

While the system uses machine learning to parse language, the DVLA emphasised that it does not employ generative AI. All responses remain pre-programmed by staff to ensure the information provided is factual and controlled. Callers also retain the option to ask for an advisor at any point during the interaction.