Chapter 4 · Generative AI in Practice: Where Enterprise Value Is Created
From experimentation to infrastructure — how organisations are deploying generative AI at scale.
From Pilots to Production
Generative AI has crossed a significant threshold. What began as a wave of cautious experimentation in 2023 has become, by 2025, something closer to organisational infrastructure. Innovation budgets, which still accounted for a quarter of enterprise LLM spending just a year earlier, have fallen to just 7% as AI spend migrates into permanent IT and business-unit budget lines.2 Surveys of U.S. enterprise leaders now show over 80% using generative AI at least weekly — up from 72% the prior year and 37% in 2023.3
The question organisations now face is not whether to use generative AI, but how to deploy it systematically, responsibly, and at the scale required to generate real returns. This commitment shows up in procurement behaviour: by 2025, 76% of enterprise AI use cases are purchased rather than built internally — a reversal from the near-even split of the prior year — and AI solutions convert to production contracts at 47%, almost twice the 25% rate of traditional SaaS.1
Key takeaway: Enterprises have decisively shifted from building AI in-house to buying it, and they are closing deals at twice the rate of conventional software — a sign of genuine confidence rather than cautious experimentation.
Global enterprise generative AI spend reached an estimated $37 billion in 2025, up from $11.5 billion in 2024 — a 3.2x year-over-year increase, according to Menlo Ventures' bottom-up market model.1 This is not speculative investment. It reflects genuine operational deployment across every major industry sector.
Where the Value Is Actually Being Created
Despite the breadth of generative AI discourse, enterprise value is concentrating in a surprisingly focused set of applications.
Coding is the clear standout at $4.0 billion — 55% of departmental AI spend — making it the largest single category across the entire application layer. This reflects a broader pattern: generative AI delivers the most immediate and measurable ROI in domains where the output is verifiable, the iteration cycle is fast, and the cost of error is contained.
The most used applications are also the highest rated in performance — data analysis, document summarisation, and document editing and writing — while particular functions are adopting specific use cases such as code writing for IT, employee recruitment and onboarding for HR, and legal contract generation for Legal.
The Six High-Impact Application Domains
1. Software Development and Engineering
Software development has undergone a step change in AI adoption, driven by a combination of high-quality off-the-shelf tools, rapidly improving model capabilities, and a clear ROI case. One CTO at a high-growth SaaS company reported that nearly 90% of their code is now AI-generated through tools such as Cursor and Claude Code, up from 10–15% twelve months prior.
The value chain extends beyond code generation to encompass automated testing, bug detection, documentation generation, and code review. This compresses development cycles and redistributes engineering effort toward higher-order architectural decisions.
2. Knowledge Work and Document Intelligence
Generative AI has fundamentally changed the economics of knowledge work — the drafting, summarising, translating, and synthesising of information that consumes enormous portions of professional time across every industry.
Retrieval-Augmented Generation (RAG) — a technique that grounds model responses in an organisation's own documents and data — has become the dominant architecture for enterprise knowledge applications. It enables AI systems to answer questions accurately against proprietary knowledge bases without requiring full model retraining.
3. Customer Experience and Support
AI-powered customer support has moved well beyond the scripted chatbots of the previous decade. Modern generative AI systems can handle complex, multi-turn conversations, understand emotional context, escalate intelligently to human agents, and maintain consistency across thousands of simultaneous interactions.
Key capabilities now in production deployment include: personalised response generation grounded in customer history; real-time agent assistance that suggests responses and surfaces relevant information; automated ticket classification and routing; and post-interaction summarisation that reduces agent administrative burden.
4. Content Creation and Marketing
Generative AI has restructured the economics of content production. What previously required teams of copywriters, designers, and translators can now be produced at scale with AI as a core production tool — not a novelty, but part of the standard workflow.
The practical applications range from personalised email generation at scale, to multilingual content adaptation, to the creation of product descriptions, marketing copy, and social media content. More sophisticated deployments integrate AI-generated content with real-time data feeds to produce dynamically personalised communications at the individual customer level.
5. Financial Analysis and Decision Support
Financial agents can help private markets and commercial lending teams run financial analysis — users submit model spreadsheets and deal-related documents with a scenario description in plain language, and the agent extracts relevant data and computes financial metrics across modelling assumptions.
Beyond document analysis, generative AI is being applied to risk assessment, regulatory compliance review, fraud detection pattern analysis, and the generation of financial narratives from structured data — the kind of work that previously required significant analyst time. In mortgage underwriting, for example, United Wholesale Mortgage reports more than doubling underwriter productivity within nine months of deploying generative AI, cutting loan close times across its network of 50,000 brokers.4
6. Supply Chain and Operations
A Capgemini study of global supply chain executives reveals that 68% of supply chain organisations have implemented AI-enabled traceability and visibility solutions, resulting in a 22% boost in efficiency.
Operational applications of generative AI include predictive maintenance (using historical data to anticipate equipment failures before they occur), supply chain optimisation through natural language interfaces, and the automation of procurement workflows and vendor communications.
The Multi-Model Reality
One of the most significant developments in enterprise AI deployment is the shift away from a single-model strategy. In a 2025 survey of enterprise CIOs, 37% reported using five or more models — up from 29% the prior year.2
This reflects a maturing understanding that different models have different strengths:
| Capability | Leading Model Preference |
|---|---|
| Fine-grained code completion | Claude (Anthropic) |
| System design and architecture | Gemini (Google) |
| Writing and content generation | Claude (Anthropic) |
| Complex question answering | GPT series (OpenAI) |
| Cost-sensitive, high-volume tasks | Open-weight models |
| On-premise, data-sensitive workloads | Self-hosted open-weight |
This fragmentation by use case is not a problem to be solved — it is a rational response to a market in which no single model dominates across all dimensions. A countervailing pressure is emerging, however: as agentic workflows demand multi-step prompt engineering and custom guardrails, switching between models is becoming meaningfully harder, with enterprises reporting that re-tuning a single workflow for a different model can require significant engineering investment.2
The Persistent Gap: From Use to Value
Despite the scale of deployment, a significant gap persists between AI usage and measurable organisational value. Only 16% of enterprise deployments qualify as true agents — systems where an LLM plans and executes actions, observes feedback, and adapts its behaviour — while most are still built around fixed-sequence or routing-based workflows wrapped around a single model call.
Wharton's longitudinal survey of ~800 U.S. enterprise decision-makers adds a further layer: 72% now formally measure Gen AI ROI and three in four already report positive returns, yet 43% simultaneously worry that sustained AI use will erode employee skill proficiency — particularly among junior workers still building foundational capabilities.3
Key takeaway: Enterprises have moved from guessing whether AI delivers value to measuring it — but success is creating a new risk: over-reliance that quietly hollows out the human skills the organisation will still need.
The human side remains the bottleneck and a key potential accelerant — morale, change management, and cross-functional coordination remain persistent barriers.
The organisations capturing the most value from generative AI are not necessarily those with the most sophisticated technology. They are those that have invested as heavily in the human and organisational dimensions of adoption as in the technical ones. This theme — the gap between technical capability and organisational readiness — runs through every subsequent part of this overview.
The technology is ready. The question every organisation must answer is whether it is.
References
- Menlo Ventures (2025). State of Generative AI in the Enterprise 2025. Menlo Ventures.
- Wang, S., Xu, S., Kahl, J. and Erten, T. (2025). How 100 Enterprise CIOs Are Building and Buying Gen AI in 2025. Andreessen Horowitz.
- Wharton Human-AI Research & GBK Collective (2025). Accountable Acceleration: Gen AI Fast-Tracks Into the Enterprise. Wharton Human-AI Research & GBK Collective.
- Renner, M. and Chaban, M. (2025). Real-world gen AI use cases from the world's leading organizations [living document]. Google Cloud Blog. Available at: https://cloud.google.com/transform/101-real-world-generative-ai-use-cases-from-industry-leaders
- Stack AI (2025). Enterprise AI Adoption: State of Generative AI in 2026. Stack AI.
- Capgemini Research Institute (2025). AI in Supply Chain: Traceability and Visibility Report. Capgemini.
Building agentic AI and wondering why alignment is harder than the technology? Get in touch