AI in R&D Runs on People, Trust, and a Clear Return
Part 2 of our AI in R&D series. The skills, rules, risks, and costs to plan for, and the customer payoff that makes them worth it.
In part 1, we looked at how quickly AI is changing research and development (R&D), and why data readiness decides who captures the gains. This part covers the rest of the plan: the people, the rules and risks, the cost, and the payoff customers see.
The Cambridge report behind part 1 notes that frontier AI companies are deploying more capital, as a share of US GDP, than the Manhattan Project and the Apollo Program combined. Anthropic reports that AI's share of its approved code rose from low single digits to over 80% between January 2025 and May 2026. Most R&D teams move at a different speed, with the same pressures on skills, risk, and cost.
Skills decide how far AI goes
Deloitte's State of AI in the Enterprise 2026, a survey of 3,235 leaders, names the AI skills gap as the top barrier to integration. The people doing the work feel it too. In BCG's AI at Work 2025 survey of more than 10,600 workers, only 36% felt adequately trained in AI use, and 41% worried their roles could disappear within the next decade.
In R&D, the training gap and the fear of displacement are one problem. The Cambridge authors add a warning from AI R&D itself: as humans become less involved in R&D, they could lose both the opportunities and the expertise needed to identify and fix problems. The answer is to build skills alongside the tools, keep experts reviewing the outputs that matter, and be open about which tasks change and which roles grow.
Plan for rules, security, and ethics
The EU AI Act entered into force on 1 August 2024. It reaches UK businesses that place AI systems on the EU market, or whose AI output is used in the EU. Bans on prohibited practices and AI literacy obligations have applied since 2 February 2025. After the AI Omnibus, an amendment in force since 27 July 2026, the rules for high-risk uses in sensitive areas apply from 2 December 2027, and those for high-risk AI built into regulated products from 2 August 2028. Fines for prohibited practices reach €35 million or 7% of worldwide annual turnover.
R&D teams have a useful carve-out. AI developed solely for scientific research and development sits outside the Act, as does research and testing before a system reaches the market, though testing in real-world conditions is covered. In the UK, the approach runs through existing regulators: in January 2026 the government wrote to 19 regulators asking each to publish a plan for supporting safe AI innovation.
Security needs the same attention. IBM's Cost of a Data Breach Report 2025 found that 13% of organisations reported breaches of AI models or applications, and 97% of those lacked AI access controls. In incidents involving shadow AI, meaning AI tools used without the organisation's approval, intellectual property was compromised in 40% of cases, against a 33% global average. In R&D, the intellectual property is the product, so access control comes first.
Ethics belongs in the same plan. The Stanford AI Index 2026 counts 362 documented AI incidents in 2025, up from 233 in 2024, and finds that responsible AI benchmarking is not keeping up with AI advances and deployments. Bias checks on training data, a clear record of how each model was tested, and a named owner for each decision turn good intentions into practice. We set out three habits in shipping AI features without losing trust.
Costs arrive before the return
The bill covers specialist software, compute, data work, and people. IBM's 2025 CEO Study of 2,000 CEOs found that only 25% of AI initiatives had delivered expected return on investment (ROI) over the last few years, and only 16% had scaled enterprise-wide. In Deloitte's 2025 survey of 1,854 senior executives across Europe and the Middle East, the UK included, most reported satisfactory ROI on a typical AI use case within two to four years. Only 6% reported payback in under a year.
Unit costs are falling fast: CASP cites research showing the price of running a model at a given capability milestone falling roughly 9- to 900-fold per year. Yet the Stanford AI Index 2026 reports compute costs and infrastructure spending at record levels. That makes cost a design choice: set the target KPI (key performance indicator) before the pilot, budget for the whole life of the workflow, and keep only what pays. We cover the method in how to make your AI project pay back.
Personalisation is the market-facing payoff
Customers see the return when R&D builds products around what they want. BCG's 2025 Personalization Index found that personalisation leaders achieve compound annual growth rates 10% higher than laggards, yet only 10% of companies are leaders.
AI shortens the path from customer signal to tailored offering. In BCG's view of AI-powered R&D, AI-automated trend reports built on weak-signal analysis replace extensive manual market research in ideation. That gives product teams evidence for tailored variants earlier in the cycle. We explore the customer side in customer experience optimisation.
The bottom line
AI helps R&D teams accelerate innovation cycles by improving experimentation, data analysis, and product development. As part 1 set out, BCG expects AI in R&D to cut time to market by 10-20% and R&D costs by up to 20%. Integrated well, it removes unnecessary complexity and opens new sources of competitive advantage. The challenges to plan for are data quality, security, data availability, and software costs.
Organisations that plan for all of them from day one turn AI from a pilot into part of how R&D works. If you are building that plan, we can help you map the skills you need, set the controls and the KPIs, and choose the first workflow that pays back. Book a call and we'll start with the decision that matters most to your R&D roadmap.
General information only and not legal advice. Sources. CASP, What if automating AI R&D triggers an intelligence explosion?. Deloitte, The State of AI in the Enterprise 2026. BCG, AI at Work 2025. European Commission, AI Act. IBM, Cost of a Data Breach 2025. Stanford HAI, AI Index 2026. IBM, 2025 CEO Study. Deloitte, AI ROI. BCG, Personalization Index 2025. BCG, AI-powered R&D.
