Generative AI Adoption by Industry: Where It Is Being Used, What Is Working, and What Comes Next

Generative AI has moved from boardroom conversation to operational infrastructure in under three years. In 2025, 88% of organisations reported using AI in at least one business function. McKinsey's State of AI research finds 71% regularly using generative AI specifically — a figure that would have seemed implausible at the start of 2023.

But aggregate statistics obscure the reality on the ground. Adoption is uneven. Some industries are running production AI systems that have measurably transformed their operations. Others are running pilots that have yet to find a business case. And within industries, the gap between the most and least advanced organisations is widening rapidly.

This post is a sector-by-sector breakdown of where generative AI is actually being used and what use cases are delivering the most value. It also looks at what the adoption data tells us about where things are heading.

The Headline Numbers

The scale of what is happening deserves context:

• $3.70 — average return for every $1 invested in generative AI, per a Microsoft-sponsored IDC study
• $109 billion — private AI investment in the United States in 2024 alone, per Stanford's AI Index
• 10 percentage points — the increase in US adult AI adoption in the last 12 months, per the Federal Reserve's Real-Time Population Survey
• 4.2x ROI — the return seen by financial services firms on generative AI investment specifically, the highest of any sector tracked

These are not the numbers of an experimental technology. Generative AI has crossed the line from novelty to economic driver.

Financial Services — The Most Advanced Adopter

Financial services has moved faster and deeper into AI than any other major sector. Cambridge Judge Business School's 2026 Global AI in Financial Services Report puts the figure at 92 percent of global banks with AI deployed in at least one core function. The great majority have also implemented AI-powered fraud detection.

What is working

Fraud detection and risk is the most mature use case. Real-time transaction monitoring using AI has cut fraud losses significantly at scale. Its models identify anomalous patterns far faster than rule-based systems could manage. The ROI is clear, the risk is manageable, and the regulatory environment has been relatively accommodating.

Credit and risk scoring is being transformed from a backwards-looking statistical exercise into a forward-looking assessment that incorporates unstructured data — news, social signals, document analysis — alongside traditional financial metrics.

Regulatory and compliance work is being automated at pace. Summarising regulatory changes, monitoring communications for compliance issues, preparing regulatory submissions and conducting first-pass contract reviews are all areas where generative AI is delivering measurable gains.

Customer service and advisory is evolving. AI-powered chat and voice assistants now handle a significant proportion of routine customer enquiries at major banks, with human agents handling complex or sensitive cases.

AI agents became the sector's story in 2026. KPMG's Q2 2026 AI Pulse Survey recorded banking-sector agent deployment jumping from 23 percent to 56 percent in a single quarter — the fastest movement of any industry tracked — against average planned AI investment of $170 million per institution.

What is still hard

Explainability in credit decisions remains a regulatory challenge. High-stakes automated advice is still heavily governed. The compliance overhead of deploying AI in regulated functions is real, and smaller institutions often lack the resources to navigate it.

Healthcare — Highest Growth Rate, Highest Stakes

Healthcare AI is growing at a compound annual rate of 36.8% — among the fastest of any sector's AI market. Growth is driven by the sheer scale of the productivity problem healthcare faces and the potential value of getting AI right in clinical settings.

What is working

Clinical documentation has emerged as the most impactful near-term use case. Clinicians spend a disproportionate amount of their time on administrative work. AI-powered ambient documentation tools listen to patient consultations and draft structured notes. They have shown dramatic time savings, with physicians in pilot programmes recovering hours of clinical time per week.

Medical imaging and diagnostics has seen accelerating adoption, with AI tools now reading radiology scans, detecting anomalies and flagging cases for urgent review in a growing number of NHS trusts and hospital networks.

Patient communication and triage is evolving rapidly. A growing share of major healthcare networks now use AI-powered chatbots to handle initial patient enquiries, freeing clinical staff for higher-complexity work.

Drug discovery and research is perhaps the most exciting frontier. Generative models are being used to propose novel molecular structures, predict protein interactions and dramatically accelerate the early stages of drug development pipelines.

What is still hard

Patient data governance is genuinely complex. Integration with legacy clinical systems remains a significant barrier. The consequences of errors in clinical AI are severe, which means validation and regulatory approval processes are — rightly — demanding.

Retail and E-commerce — Speed and Personalisation at Scale

The generative AI retail market was valued at over $1 billion in 2025 and is growing rapidly. Growth is driven by the industry's intense competitive pressure on personalisation, efficiency and speed-to-market.

What is working

Product content and copywriting was one of the earliest and most straightforward use cases. Generating product descriptions, category pages and marketing copy at scale — across thousands of SKUs, in multiple languages, for multiple channels — is a task that generative AI does well.

Personalised recommendations and messaging has moved beyond collaborative filtering into generative territory, with systems that compose personalised outreach and adjust promotional messaging to individual customer segments in real time.

Visual merchandising and creative is emerging as a significant use case. AI image generation is being used to create product imagery variations, lifestyle photography and seasonal campaign assets at a fraction of the cost of traditional photography.

Customer service at retail scale — handling returns, tracking enquiries, product queries — is heavily automated, with generative AI handling the long tail of customer interactions.

What is still hard

Maintaining brand voice and quality at AI generation scale requires systematic evaluation and human review at every stage. Hallucinated product specifications are a real and embarrassing risk.

Legal and Professional Services — From Experiment to Infrastructure

The American Bar Association's assessment in late 2025 was unambiguous: AI has moved "from experiment to infrastructure" in the legal profession. 26% of legal organisations are actively using generative AI, up from 14% in 2024. 95% of legal professionals expect it to become central to their workflow within five years.

What is working

Document review and due diligence is the single highest-impact use case. Reviewing thousands of documents in a data room, identifying relevant clauses, flagging anomalies and producing structured summaries can now be done in hours with AI assistance. This work previously required large teams of junior lawyers.

Legal research has been transformed by AI tools that can synthesise case law, identify relevant precedents and produce structured research summaries. Document summarisation, at 74% adoption among legal professionals, is now close to standard practice.

Contract drafting and review is evolving rapidly. Agentic systems can now review an uploaded agreement, identify clauses that deviate from standard templates, flag risk areas, and produce a redlined version — without step-by-step prompting.

What is still hard

Hallucination in legal research is a serious risk — there have been high-profile cases of AI-generated citations that do not exist. Privilege and confidentiality concerns around client data entering third-party AI systems remain live issues.

Manufacturing and Engineering — Efficiency at the Operational Level

What is working

Predictive maintenance uses AI to identify patterns in machine sensor data that precede failures, allowing maintenance to be scheduled before breakdowns occur. The ROI in reduced downtime is often compelling enough to justify substantial investment.

Quality control and defect detection — computer vision models monitoring production lines in real time — has become standard in advanced manufacturing environments.

Design and engineering assistance is an emerging use case where generative AI helps engineers explore design variations, generate documentation and identify potential failure modes earlier in the development cycle.

What is still hard

Legacy operational technology systems in factories are often decades old and not designed for AI integration. The gap between what is technically possible and what is safely deployable in regulated manufacturing environments remains significant.

Education — Early but Accelerating

Personalised tutoring is the most promising long-term use case. AI systems that adapt to an individual student's pace, learning style and knowledge gaps have shown strong results in pilot programmes. The UK government is investing in AI tutoring tools specifically aimed at supporting disadvantaged pupils.

Teacher productivity — lesson planning, resource creation, marking assistance, differentiation of materials — is already delivering time savings for teachers willing to adopt AI tools.

Assessment and feedback at scale is being explored by universities and examining bodies, though with significant caution around assessment integrity.

Non-profit and Charity — High Promise, Significant Barriers

The non-profit and charity sector presents one of the most compelling — and most complex — stories in generative AI adoption. The headline figure is striking: 82% of non-profits now use AI in some form, according to BDO's 2024 sector survey. But dig deeper and the picture is more nuanced. Only 7% say AI is embedded into their goals, budgets and performance indicators, and a mere 10% have any AI governance policy in place.

The sector is adopting AI broadly but shallowly. Sixty-five percent describe their AI use as "reactive and individual" — one-off prompts, personal experimentation — rather than operational.

What is working

Fundraising and donor engagement is the area of highest interest and most tangible impact. Organisations using AI for fundraising are reporting meaningful increases in donations through predictive analytics, personalised outreach and automated engagement strategies. AI helps gift officers prioritise their portfolios, draft personalised communications in their own voice, and identify lapsed donors most likely to re-engage. This delivers relationship-level personalisation at a scale that was previously impossible for under-resourced teams.

Grant writing and reporting is one of the most time-consuming burdens charities face. Generative AI is being used to draft grant applications, tailor proposals to specific funders, and produce impact reports — dramatically reducing the hours required while maintaining quality. For smaller organisations with no dedicated fundraising staff, the time saving is material enough to change what is possible.

Content and communications — social media, email campaigns, website copy, volunteer recruitment materials — is an area where even the smallest charity can immediately benefit. Generating a first draft in seconds rather than hours is a meaningful productivity gain for teams where everyone is stretched.

Service delivery innovation is where the most ambitious use cases are emerging. Spring ACT's Sophia chatbot, for example, assists survivors of domestic violence in more than 140 countries, providing 24/7 anonymous support in dozens of languages — a scale of reach that no human team could replicate.

What is still hard

The barriers in the sector are real. 60% of non-profits say they lack the in-house expertise to assess AI tools, and only 4% have AI-specific training budgets. Smaller organisations face the sharpest constraints: limited funds, limited technical capacity, and genuine uncertainty about where to start.

There is also a values tension that the sector takes seriously. Donor communications that feel automated risk undermining the authenticity that distinguishes a charity's relationship with its supporters. 63% of fundraisers are uncomfortable using generative AI for donor communications — even as 82% are comfortable using AI for donor research. The distinction matters: AI as a research and preparation tool feels different from AI as a voice.

Google.org's Generative AI Accelerator — a six-month programme for selected non-profits providing up to $2 million in equivalent support, Google Cloud credits and technical training — is one example of the funded routes now available. For UK charities, the AI Skills Hub and Charity Digital both publish accessible resources for organisations beginning their AI journey.

The sector's values need not block AI adoption. They can guide it instead.

The Pattern Across All Sectors

Looking across industries, several consistent patterns emerge:

The productivity wedge is real. Every sector is finding that generative AI dramatically accelerates the production of first drafts — of documents, code, analysis, images, communications. It does not replace human judgement here. It eliminates the blank-page problem and compresses the time from brief to reviewable output.

High-volume, routine tasks are the fastest wins. Fraud detection, document review, product description generation, clinical note-writing — the common thread is high volume, relative consistency and clear quality criteria. These are the use cases delivering ROI at scale today.

Agentic applications are the next frontier — and now measurable. The shift from AI-as-assistant to AI-as-agent is in production: KPMG's Q2 2026 AI Pulse puts agent deployment at 53 percent of large US enterprises, with multi-agent orchestration doubling to 18 percent in a quarter. The governance behind it lags badly, a gap we examine in The Agent Governance Gap.

The gap between leaders and laggards is widening. Organisations that adopted early are compounding their advantages: PwC's 2026 AI performance study found three-quarters of AI's economic gains are being captured by just 20 percent of companies. Those still evaluating are behind, and the gap widens with each quarter.

What Should Your Organisation Do?

Identify your highest-volume, most consistent tasks. These are your fastest wins. Where do your teams spend significant time on work that is repetitive, document-heavy or information-intensive? Start there.

Do not wait for the perfect use case. The organisations that have learned the most about AI are those that shipped something — even imperfect — into production. Learning from real usage is worth more than any pilot programme.

Invest in evaluation. The difference between useful AI and embarrassing AI is usually the quality of your evaluation and human review processes. Build these before you scale.

Think about your data. The most powerful AI applications connect models to your organisation's own data. The organisations that have done the work of making their data accessible will extract dramatically more value from AI than those that have not.

Frequently asked questions

How widely is generative AI used in 2026?

In 2025, 88 percent of organisations reported using AI in at least one business function, and by 2026, 71 percent are regularly using generative AI specifically. Reported returns average $3.70 for every $1 invested.

Which industries have adopted generative AI most?

Financial services is the most advanced — 92 percent of global banks have AI in at least one core function. Healthcare AI is one of the fastest-growing markets at 36.8 percent a year, and legal services has moved, in the American Bar Association's words, from experiment to infrastructure.

Which use cases deliver the most value?

High-volume, consistent tasks with clear quality criteria — fraud detection, document review and due diligence, clinical documentation, product-content generation and internal knowledge retrieval.

What should organisations do first?

Identify your highest-volume, most repetitive tasks; ship something into production rather than waiting for the perfect use case; invest in evaluation and human review; and make your own data accessible, since the most valuable applications connect models to it.

New posts land here first. Sign up for email updates.