Derby Council AI tools still resolving half of calls three years on

TL;DR:

• Derby City Council says its AI tools Darcie and Ali resolve about 50% of public queries without human intervention — unchanged from when they launched in 2023.
• The technology, which replaced four full-time agency roles, has saved £200,000 a year on the original £168,000 outlay, with the wider council AI programme credited with £12 million of savings, half from adult care.
• For other UK councils piloting AI, Derby’s flat performance curve is a useful corrective to the assumption that AI tools improve simply by remaining in service.

Derby City Council has published updated figures showing its public-facing AI tools Darcie and Ali continue to resolve around half of incoming queries without human assistance — the same rate they were achieving when the council became the first UK local authority to replace its main switchboard with an automated system in 2023. Darcie handles customer service centre calls; Ali takes housing enquiries for Derby Homes.

What the numbers actually show

The initial deployment cost £168,000 and replaced four full-time agency positions, generating reported annual savings of £200,000. Last year the council brought in a “major upgrade” after complaints that the tools struggled with local Derbyshire vocabulary — including “duck” (dear or love) and “mardy” (moaning). Even after the upgrade, the resolution rate has not climbed.

The council reports that 77% of respondents who gave feedback described the tools as positive, but the static resolution figure raises a question other councils piloting similar systems should ask: at what point does the residual 50% — the citizens the AI cannot help — become a service-quality problem rather than a cost-saving win? Derby Conservatives leader Steve Hassall has previously warned the system can feel like a “digital barrier” to vulnerable users.

The £12 million claim

The wider council AI programme is credited with £12 million in savings, with half coming from adult social care. That figure deserves scrutiny: the Community Care sector is currently raising significant legal-literacy concerns about AI use in social care assessments, particularly where decisions produce legal effects under the Mental Capacity Act 2005 or trigger UK GDPR Article 22 protections. Cost savings achieved through AI-assisted social care work that has not yet been legally tested could prove fragile.

Looking forward

For UK councils watching Derby as a proof point, the lesson is mixed. Three years of operational data show genuine, sustained cost reduction — but the absence of improvement in resolution rates suggests AI is plateauing at what it can handle in this configuration. Councils planning similar deployments should budget for human-handled exception flows that will not disappear, and treat AI savings in clinically or legally sensitive areas with greater care than savings on routine switchboard work.