91探花

Governments believe text-based and image-based AI models can speed up services and cut paperwork. The Capgemini Research Institute鈥檚 2025 survey of 350 agencies shows that 64% have started pilots or early rollouts of generative AI. Defense, health and security offices take the lead as 82%, 75% and 70% of them test ideas already, thanks to bigger budgets and pressing missions.

Agentic AI is next on the agenda. Capgemini reports that 90% of all respondents hope to explore, pilot or install autonomous agents within 3 years, expecting these systems to handle multi-step tasks such as drafting letters or routing complaints.

So enthusiasm runs high, but the same study warns that progress slows once projects move past the lab. Only 21% of agencies have put a live service in citizens鈥� hands.

How Far Has Adoption Moved Today?

Capgemini splits agencies into 4 groups. Roughly 1/3 earns the tag 鈥渇rontrunner鈥�, because they pair modern tools with staff habits that treat data as routine currency. These frontrunners harness insights for everyday action and report higher returns.

On the other hand, 67% lag on one or both fronts. They may buy new software yet keep records in silos, or they train analysts yet leave servers dated.

Infrastructure troubles are at the top with the problem areas. 77% of executives tell Capgemini their current hardware cannot scale to AI workloads. Budgets squeeze next, as 65% lack cash for major upgrades in one go.

Governance also bites, with over half saying rules for privacy, quality checks and audit are still patchy, so project teams hesitate to feed sensitive material into new models.

What Blocks The Road To Ready Data?

Capgemini lists 7 parts of data maturity, from 鈥渋dentifying records鈥� to 鈥渁ctivating insights鈥�. Fewer than 25% mark themselves highly mature on any part.

Only 10% run cloud platforms that scale on demand, while 12% weave insights into daily routines.

Skill gaps add strain, just 29% run job-based training for most workers, and only 38% teach frontline staff how to turn charts into clear stories for managers.

Culture slows change as well, since many offices reward risk-free routines; staff fear mistakes with personal records. Sweden鈥檚 public employment service told Capgemini that leaders must set clear goals and discuss data use openly every week to build trust.

Data sharing falls behind too, with 35% having rolled out or finished a sharing scheme, but only 8% reached full coverage. Capgemini links the shortfall to trust issues between departments.

Energy use rounds out the list. 3/4 of agencies worry about the power draw of large models, yet only 59% act to curb it.

Why Does The UK Lead Europe In Trials?

A separate Capgemini memo focused on Britain shows 75% of UK agencies experimenting with generative AI, ahead of neighbours on the continent. Early cloud moves and a ministerial need for 鈥淎I ready鈥� services help.

The same note says 90% of British departments also eye agentic AI pilots within three years. Chatbots for tax queries and case sorting tools for social payments top pilot lists.

Data protection and sovereignty fears sit at 78%. Cost pressure follows at 68%, while 75% flag the power and water cost of big models. Only 21% feel they already hold enough clean data to train new systems, so secure sharing sits high on Whitehall鈥檚 to-do list.

Capgemini also spots new leadership: Chief Data Officers work in 60% of UK bodies, and Chief AI Officers appear in 27%, with more hires planned.