Whitehall AI savings lie in the back office, not chatbots

TL;DR:

• Analysis from consultancy Baringa argues UK government AI savings are more likely to come from back-office work than from automating citizen-facing contact.
• Just 14% of surveyed people prefer AI to human interaction, and only 0.07% want fully automated services; the top request is for AI to escalate to a human intelligently.
• Departments face Spending Review demands to find at least 5% in savings by 2028–29, with efficiency plans banking on AI to help.

With departments under pressure to do more with less, automating the public’s contact with government looks like a tempting quick win. New analysis argues it is usually the wrong place to start. Drawing on a survey of more than 1,000 customers and 37,000 service reviews, Baringa found that automating citizen interactions tends to deliver fewer benefits than promised — while the real gains sit in work the public never sees.

Where the savings actually are

The public appetite for AI on the front line is thin. Two in five people find AI in services frustrating, rising to half of the over-45s, and 82% say speaking to a human matters for resolving an issue. Even the government’s own GOV.UK Chat trial, at 90% accuracy, still leaves nearly one in ten answers short and cannot see a user’s account details.

The stronger case is behind the counter. AI can triage requests, flag missing evidence, summarise cases and analyse the root causes of repeat contact — the “failure demand” that drives cost when people call back because their issue was not resolved first time. The government’s Consult tool has run across 26 consultations and saved over £500,000, while machine learning is being turned on the £55bn–£81bn lost annually to fraud and error.

The caution matters because Whitehall’s targets assume AI delivers. Departments must find at least 5% in savings by 2028–29 and cut admin budgets by 16% in real terms, with published plans identifying almost £14bn of annual efficiency gains. That echoes private-sector findings this week that weak governance, not low spending, is why AI projects fail — pilots often succeed only because a motivated team papers over gaps that a full rollout exposes.

Looking forward

The practical steer for departments is to begin where AI improves the end-to-end journey and can show a baseline, intended gains and safeguards — and to resist the standalone chatbot as a headline-friendly but shallow fix. Meeting the efficiency targets will depend less on visible automation than on the unglamorous back-office work that actually removes cost.