← Cases
United KingdomPublic sector2025 – ongoing

Department for Work and Pensions (DWP)

Conversational Platform telephone virtual assistant

DWP phone assistant handles about 1 million calls a month and raised right-first-time routing by 21%

WorkingCompany's own claim

Problem

Callers to DWP had to be sorted by reason for calling, then sent to self-service or the right adviser.

Approach

A voice assistant from Omilia asks callers why they are calling and routes them to self-service options or a human adviser. DWP trained the language model on 80,000 caller utterances recorded over one week. Speech recognition is tuned monthly and the language model updated each quarter.

Results

21%Increase in right-first-time routing of callers (baseline not stated)Company's own claim[1]
9%Increase in calls deflected to self-service (baseline not stated)Company's own claim[1]
97%Speech recognition success rateCompany's own claim[1]
1 millionCalls handled per month (approximate)Company's own claim[1]

Takeaway

The assistant used callers' own descriptions to route them to self-service or an adviser. DWP did not disclose the baseline for its reported routing improvements or any cost savings.

Categories

Business function
Customer service
Innovation type
Process innovation
Evidence
Company's own claim

Sources

  1. DWP: Conversational Platform - Algorithmic Transparency Record
    GOV.UK (Department for Work and Pensions)16 Dec 2025Government publication

Source summaries

[1]GOV.UK (Department for Work and Pensions)16 Dec 2025

DWP described the phone assistant, its supplier and how often it is retrained. It reported call volume, recognition accuracy and routing changes.

Full details are in the original sources.

Added 25 Oct 2026 · Updated 25 Oct 2026
Open case page ↗