Derby City Council eyes £12.25m annual savings with AI and chatbot deployments

Chatbots at the council

This initial project saw the council roll out both “phone-based” AI, in place of a conventional IVR system and switchboards, and two chatbots, Darcie and Ali. These, the council said, have handled over a million web and phone queries. The council had expected its AI tools to handle 20% of phone calls, but it said the actual figure has already reached 43%.

Derby City Council now hopes to save £4m in the 2024/25 financial year though AI, with savings forecast to rise to close to £12.25m once the system is fully installed. The bulk of the savings, some £8.9m, will come from adult social care, according to council documents.

The remainder will come from children’s services, “customer channels”, internal automation and debt management.

The cost, which covers the council and housing provider Derby Homes, is £7.4m for a three-phase, five-year project with Microsoft and integrator ICS.AI.

This includes a 12-month, fixed-cost “provisioned deployment” arrangement. “The fixed cost agreement gave the council two benefits,” said Brammall. “First, it protected us from unexpected price changes. Second, it guaranteed that the AI engine would be available and meet performance standards.”

Improving efficiency

Derby City Council’s project is one of a growing number of initiatives looking to improve efficiency and productivity in the public sector.

Research carried out by Goldsmiths University, for Microsoft, found that public sector workers could save more than 23 million hours of administration each week if AI tools were deployed across the board.

Microsoft cites staff at Buckinghamshire Council as saving 90 minutes a day by using Microsoft Copilot for 365, which it deployed to 300 staff. And Barnsley Council is deploying Copilot “at scale”, according to Microsoft. Staff across finance, HR and legal teams are using the tool. Like Derby, Barnsley is finding AI to be useful in social care.

According to Rob Anderson, chief analyst at GlobalData, Bolton Council is another large user of Copilot. “A few months ago, over 60 local authorities were paying Microsoft for the use of Copilot for Microsoft 365, usually for a few hundred users,” he said. “This number appears to be growing exponentially.”

Anderson sees local authorities deploying AI in two ways: Copilot for Microsoft 365, and through existing enterprise software packages. Derby, for example, is integrating AI into Liquidlogic, a line of business system for social care.

Councils are looking to generative AI, first and foremost, to streamline administration. As GlobalData’s Anderson puts it, they are using “intelligent cut and paste” to move information between records and systems. They are also using it to summarise information, for example in consultation exercises. But they are also integrating generative AI with tools such as machine learning and business intelligence.

“The AI programme is a crucial element of [Derby] council’s digital transformation, supporting all its operations,” said Brammall. “The principle of professional co-pilots is that AI is employed to do much of the “administration heavy lifting”, freeing up professionals to focus on what only humans can do – using their expertise to solve complex problems.”

There are risks with any AI deployment. Aside from direct security risks, such as prompt injection attacks, councils are concerned about AI hallucinations, as well as the risk of lower-level mistakes or misleading advice. Even something as simple as an inaccurate meeting summary could have serious ramifications in local government.

“The consequences of privacy and security breaches are far higher for the public sector, so, understandably, it needs to be slower and more considered in adopting AI than the private sector,” said Radek Dymacz, head of cloud architecture at Databarracks.

Derby City Council acknowledges this, and has set up an AI compliance and ethics board to monitor its roll-out. The council notes that, so far, it has had just 40 complaints out of 1.1m AI-based transactions with the public.

But public-sector bodies also need to ensure they are using the right tools for the task, including the right type of AI; this will not always be a generative AI system. “There are some interesting use cases for better-supporting customers, such as intelligent customer support or helping users to find out information,” said Mission Decisions’ Hillier. “However, we are seeing quite a few poorly implemented solutions that just frustrate the user. This doesn't help with the credibility of AI.”

Derby, though, sees itself as an early adopter – officials believe it is one of the first, if not the first, to adopt AI as broadly as it is – and are keen to share what they have learned with other public sector bodies.

“When you put forward a business case, every executive board will always say to you: ‘Who else has done this? Who can we learn from?’,” said Brammall. “Well, in our case, nobody else had done it, so we accepted that what we learned we would share.”

The council is now using AI for multi-language translation and document redaction, and is seeing benefits there. And both Derby and ICS.AI expect the use of AI to grow, provided it continues to prove its value.