RAISE: Scaling AI POV
C · Weak signal Short vendor point-of-view brochure from Capgemini's agentic AI practice with no original data; its few figures are borrowed from a 2024 BCG study and an earlier Capgemini survey.
Summary based on the full text·Original: about 12 minutes to read
In short
"From islands of value to compounding advantage" is a 20-page point of view from Capgemini's RAISE agentic AI offering, written by Craig Suckling, Global Head of Agentic Scale. It claims 2026 is the year enterprise AI moves from scattered pilots to industrial, governed deployment of agents. It lists seven barriers to scaling and proposes a four-step playbook. The document contains no new research. It is a positioning paper for Capgemini's services.
Key numbers
- 5% to 14%rise in enterprise AI success rates cited from BCG (January 2024)
- 1.7xaverage ROI on first AI use-case deployments, cited from BCG
- 27%of organizations fully trust autonomous AI decisions, cited from Capgemini (February 2026)
- fewer than 20%of organizations report high data maturity
As reported by the source.
English·PDF·opens capgemini.com
What it reports
- It asserts that model capability doubles every six months and inference costs fall 10x every nine months. No source is given for either claim.
- It cites BCG (January 2024): AI success rates rose from 5% to 14%, with 1.7x average ROI on first use-case deployments.
- It cites Capgemini's February 2026 agentic AI survey: only 27% of organizations fully trust autonomous AI decisions.
- It states that fewer than 20% of organizations report high data maturity.
- It describes agents spreading in three forms: custom-built agents, co-pilots in existing workflows, and agents built into SaaS platforms.
- The seven barriers it names are process redesign, talent and culture, data readiness, trust and governance, regulation and sovereignty, AI security, and total cost of ownership.
- Its playbook is "firm vision, focused progress, fast, flexible scale" plus a flywheel. It promotes "microtransformations", "governance as code" and FinOps-style token cost tracking.
How they got it
There is no stated method. The document combines practice opinion with three external or earlier figures. The time series the authors describe (2024 "broad and shallow", 2025 "narrow and deep", 2026 "broad and deep") is a framing device, not measured data.
Limits
- The figures are second-hand. The BCG source is from January 2024 and predates agentic deployments.
- The capability-doubling and cost-collapse claims are unsourced.
- The author sells the services the paper recommends, so treat it as a sales argument.
Why it matters for companies
The paper is worth a skim as a checklist of what a large integrator will pitch: agent control planes, semantic data layers, governance built into pipelines, and model tiering for cost. Two practical points hold up regardless of the sales angle. Pilot economics often break at scale once inference, evaluation and oversight costs appear. And putting agents into processes designed for humans makes their existing problems run faster.
Summary based on the full text. Read the original for the full argument and data.Open PDF on capgemini.com
- Tags
- AI agents, Operating model