National-Scale AI Deployment in UK Public Sector: Contact Center Automation and Citizen Information Retrieval

capita_/_uk_department_of_science · Government · 2025 · source Agentic: yes · promotional bias: med · source-backed · extracted with prompt 0.5.6 (schema v2)

Generated from extraction.json — do not edit by hand. Every claim cites verbatim source evidence; stated means the quote directly asserts the claim, implied means it supports it indirectly.

Problem

Domain: Public-sector citizen service automation

Capita and GDS needed AI systems that could serve millions of citizens in high-stakes government contexts while preserving trust, safety, and human oversight.

Constraints:

Capita handles more than 100,000 daily contact-center interactions.
GOV.UK content spans more than 850,000 pages.
The systems serve 67 million citizens.
Public-sector use has zero tolerance for errors in critical citizen-service workflows.

Architecture

Capita used AWS Connect and Bedrock with Claude for contact-center automation, while GDS built GOV.UK Chat as a national-scale Bedrock-backed RAG system over government content, both with safety, trust, and oversight controls.

Diagram confidence: reconstructed (reconstructed from prose).

flowchart LR
  capita-contact-center["Capita Contact Center"]
  aws-connect["AWS Connect"]
  amazon-bedrock-with-claude["Amazon Bedrock with Claude"]
  gov-uk-content-corpus["GOV.UK Content Corpus"]
  gov-uk-chat["GOV.UK Chat"]
  human-oversight["Human Oversight"]
  capita-contact-center --> aws-connect
  capita-contact-center --> amazon-bedrock-with-claude
  aws-connect --> amazon-bedrock-with-claude
  amazon-bedrock-with-claude --> aws-connect
  amazon-bedrock-with-claude --> gov-uk-chat
  gov-uk-content-corpus --> gov-uk-chat
  gov-uk-chat --> gov-uk-content-corpus
  gov-uk-chat --> amazon-bedrock-with-claude
  human-oversight --> capita-contact-center
  human-oversight --> gov-uk-chat

Drag to pan · scroll or use +/− to zoom · reconstructed from the source's described component topology

Components

Component	Role	Interacts with
`Capita Contact Center`	Handles citizen interactions and automation workflows.	`AWS Connect`, `Amazon Bedrock with Claude`
`AWS Connect`	Provides contact-center infrastructure.	`Amazon Bedrock with Claude`
`Amazon Bedrock with Claude`	Powers contact-center AI and GOV.UK Chat generation.	`AWS Connect`, `GOV.UK Chat`
`GOV.UK Content Corpus`	Stores government information pages for retrieval.	`GOV.UK Chat`
`GOV.UK Chat`	Provides citizen-facing RAG over government content.	`GOV.UK Content Corpus`, `Amazon Bedrock with Claude`
`Human Oversight`	Maintains trust and safety in public-sector workflows.	`Capita Contact Center`, `GOV.UK Chat`

Data flow

Citizen interactions enter either contact-center automation or GOV.UK Chat. Capita routes contact-center workflows through AWS Connect, Bedrock, and Claude; GDS retrieves government content from GOV.UK pages through a RAG pipeline before responding to citizens.

“Capita implemented AWS Connect and Amazon Bedrock with Claude to automate contact center operations handling 100,000+ daily interactions”summary, stated

“GDS launched GOV.UK Chat, the UK's first national-scale RAG implementation using Amazon Bedrock, providing instant access to 850,000+ pages of government content for 67 million citizens.”summary, stated

Why this architecture

The organizations used managed AWS and Bedrock systems because national-scale citizen service requires fast deployment, scalable retrieval, and explicit safety/oversight controls.

“Both organizations prioritized safety, trust, and human oversight while scaling AI solutions to handle millions of interactions with zero tolerance for errors”summary, stated

“providing instant access to 850,000+ pages of government content for 67 million citizens.”summary, stated

Models used: Claude · Amazon Bedrock · AWS Connect

Scale

Capita handles more than 100,000 daily interactions.
“handling 100,000+ daily interactions”summary, stated
Capita reported 35% productivity improvements.
“achieving 35% productivity improvements”summary, stated
Capita targets 95% automation by 2027.
“targeting 95% automation by 2027.”summary, stated
GOV.UK Chat covers more than 850,000 government pages.
“850,000+ pages of government content”summary, stated
The public-sector audience is 67 million citizens.
“67 million citizens.”summary, stated

Alternatives considered (roads not taken)

Manual-only contact-center operations

Rejected because: The scale of daily interactions required productivity gains and automation targets.

“automate contact center operations handling 100,000+ daily interactions, achieving 35% productivity improvements”summary, stated

Ungrounded general chatbot

Rejected because: Government information retrieval required access to official GOV.UK content and zero-tolerance safety posture.

“GOV.UK Chat, the UK's first national-scale RAG implementation using Amazon Bedrock”summary, stated

Patterns

`rag-retrieval` (retrieval)

GOV.UK Chat retrieves official government content before answering citizens.

“GOV.UK Chat, the UK's first national-scale RAG implementation using Amazon Bedrock”summary, stated

`human-in-the-loop-gate` (governance)

The public-sector deployments retained human oversight as part of trust and safety.

“prioritized safety, trust, and human oversight”summary, stated

`safety-policy-enforcement` (governance)

The systems were designed for public-sector zero-error tolerance and citizen trust.

“zero tolerance for errors in this high-stakes public sector environment.”summary, stated

Tradeoffs

Scale automation while retaining oversight

Chose AI automation with safety, trust, and human oversight over Fully autonomous public-sector service answers.

Rationale: High-stakes citizen workflows require oversight even when automation targets are ambitious.

“prioritized safety, trust, and human oversight while scaling AI solutions to handle millions of interactions”summary, stated

Use managed Bedrock stack

Chose Amazon Bedrock and Claude over Building a bespoke model platform from scratch.

Rationale: Managed services supported both contact-center automation and national RAG deployment.

“Capita implemented AWS Connect and Amazon Bedrock with Claude”summary, stated

Gotchas (what bit them)

Citizen-service AI has much lower tolerance for incorrect answers than ordinary support automation.

Root cause: Errors can mislead citizens about government services and obligations.

Resolution: Prioritize safety, trust, human oversight, and grounded retrieval.

“zero tolerance for errors in this high-stakes public sector environment.”summary, stated

National-scale content retrieval requires official corpus grounding.

Root cause: Government answers span hundreds of thousands of official pages.

Resolution: Use a RAG implementation over GOV.UK content rather than unconstrained generation.

“providing instant access to 850,000+ pages of government content for 67 million citizens.”summary, stated

Evaluation

The case reports productivity improvement, automation targets, content coverage, and public-scale reach rather than a full offline model benchmark.

Metrics: Productivity improvement · Automation target · Daily interactions · Government pages covered · Citizen population served

“achieving 35% productivity improvements and targeting 95% automation by 2027.”summary, stated

Outcomes

Capita achieved 35% productivity improvements.
Capita targeted 95% automation by 2027.
GOV.UK Chat made more than 850,000 pages available through national-scale RAG.

The deployments span 100,000+ daily interactions, millions of interactions, and 67 million citizens.

“Two UK government organizations, Capita and the Government Digital Service (GDS), deployed large-scale AI solutions to serve millions of citizens.”summary, stated