Mark T. Boyer: The Integration Edge – Leading Fintech & AI Innovation

AI is changing the economics of creativity, research and development

Executive teams are increasingly treating AI as an R&D force multiplier: faster literature review, quicker prototyping, more simulations, more drafts, more options. Evidence reviews (including OECD’s synthesis of experimental research) suggest AI can boost productivity and creativity—especially when paired with human expertise and oversight—but also warn that outcomes depend heavily on who is using the tool, on what tasks, and how much they trust it.1

From idea search to experiment design and simulation

That last point matters because organizations tend to adopt AI first where it “obviously” saves time: summarization, drafting, search, and first-pass analysis. Those uses can be meaningful—St. Louis Fed research using survey data and modeling estimates self-reported time savings from generative AI could translate to a ~1.1% increase in aggregate productivity, and implies workers can be ~33% more productive in each hour they use generative AI.2 But speed is not the same as sustained advantage. The durable edge comes from what the organization does with the time returned: more experiments, better decisions, and stronger learning loops.

In R&D-heavy environments, the “value chain” is broad: problem definition → concept generation → feasibility and risk analysis → design and simulation → testing → documentation and compliance → iteration. Consulting research emphasizes the opportunity if AI is integrated into the operating model, not bolted on. BCG frames the impact as material—expectations of 10–20% time-to-market reduction and up to 20% lower R&D costs—and argues AI agents will increasingly cover more steps of the process.3

That’s the promise. Now the pragmatic reality: AI adds leverage when it compresses search and coordination (finding prior art, locating internal knowledge, drafting documentation, running first-pass analysis). It can also improve option generation (more design alternatives, more hypotheses). But it does not automatically improve the organization’s capacity for judgment, validation, and accountability. Those are learned capabilities, and they’re fragile if you automate the wrong things too early.

Banking is a leading indicator of automation with accountability

Banking has three characteristics that make it a useful preview of where AI is headed elsewhere: high documentation load, heavy internal knowledge, and strict risk constraints. Regulators are explicitly observing that generative AI use cases in banks have been largely internal-facing—employee efficiency, coding, call-center agent assistance, employee knowledge base support, and document creation and summarization—alongside a call for strong governance and risk management.4 That is exactly the pattern you’d expect in a sector optimizing for reliability and auditability.

Spotlight: nCino’s agent-driven workflow vision

nCino’s nSight 2025 announcement is a clear example of a banking technology leader positioning AI as routine-work replacement and workflow redesign. In its press release, CEO Sean Desmond describes “pivoting…R&D capacity toward deploying agents to reimagine and automate every task,” explicitly aiming to maximize efficiency so bankers can focus on clients.5

What’s notable is the emphasis on measurable workflow outcomes. nCino cites examples such as onboarding time reduced from months to days, one institution cutting document processing time by 74%, small-business loan decisions accelerated by 62%, and mortgage document validation reducing documentation completion time by 47% (with inquiries down 68%).6 These are the kinds of improvements that, if real and repeatable, materially change cost-to-serve and speed-to-yes.

The deeper question—especially for long-term creativity and R&D—is whether institutions also redesign roles so that humans still build judgment (credit reasoning, exception handling, and contextual risk decisions), instead of becoming passive recipients of AI output.

Bank of America: internal knowledge turned into instant answers

Bank of America’s AskGPS is a clean internal-knowledge use case: an in-house genAI assistant for Global Payments Solutions trained on 3,200+ internal documents, intended to save “tens of thousands of employee hours annually.”7 The press release also explains the operational shift: work that previously might take an hour and require calls across regions can now be done “almost instantly.”8

From an R&D lens, this matters because internal knowledge retrieval is often the bottleneck to progress. When knowledge work becomes “find, interpret, apply,” AI can collapse the find step. But the interpret/apply steps must remain human-owned—especially where accountability is real.

Citi and Wells Fargo: scaling internal tools and agentic workflows

Citi offers a scale signal: CEO Jane Fraser said AI use freed up 100,000 hours of weekly capacity for software developers, and nearly 180,000 employees had access to internal AI tools across 83 countries.9 Wells Fargo, meanwhile, describes expanding its relationship with Google Cloud to deploy agentic AI tools across the bank, intended to improve customer experience, automate routine tasks, and “unlock new levels of innovation,” with an emphasis on responsible AI.10

These examples show the next frontier: not just copilots, but agent-like systems that execute multi-step work. That increases the upside—and also increases the risk of over-automation, weak auditability, and expertise decay if humans become out-of-the-loop supervisors without deep understanding.

Cross-industry proof points beyond banking

Banking isn’t alone. Across industries where new technology relies on creativity and discovery—design engineering, clinical care, marketing science, and digital product development—the same pattern emerges: AI speeds iteration, but the organization must protect validation and learning.

Manufacturing: smart factories and generative design

Deloitte’s 2025 smart manufacturing survey reports moderate but meaningful adoption: 29% using AI/ML at the facility or network level, 24% deploying generative AI at that scale, and 38% piloting generative AI.11 That adoption rate suggests manufacturing leaders are moving, but carefully—often due to IP, security, and integration concerns.

At the micro level, generative design can compress engineering cycles. An ASSEMBLY Magazine feature on Eaton reports outcomes such as minimizing the weight of a heat exchanger by 80%, lowering design time for a high-speed gear by 65%, and reducing design time for an automated lighting fixture by 87%.12 Those are big numbers—but they also raise a leadership question: what happens to the next generation of engineers if AI handles too much of the “routine” iteration that builds intuition about constraints?

Healthcare: ambient documentation and clinician capacity

In healthcare, “routine” paperwork is often the enemy of time and attention. UConn Health’s early adoption story of ambient AI documentation includes a straightforward operational benefit: clinicians report improved patient engagement and reduced documentation burden, and one physician estimates the tool saves up to 30 minutes per day finishing notes.13

That time savings can be reinvested into care and clinical reasoning—but only if clinical teams treat AI notes as draft artifacts requiring clinician judgment, not as authoritative records. The highest-value parts of a clinical note are precisely the parts that reflect reasoning (“assessment and plan”), and the article highlights clinicians using the saved time to focus on that thinking.14

Retail and consumer goods: faster content cycles and measurable conversion

Retail offers a hard-numbers example from Lowe’s: in its Q3 2025 earnings call transcript, the company stated its virtual assistants answer nearly one million questions per month, and when customers engage with the tool online “the conversion rate more than doubles.”15

Consumer and fashion brands are also using generative AI to shrink campaign timelines. Reuters reports Zalando reduced imagery production time to 3–4 days from 6–8 weeks and cut costs by 90%, with its VP noting, “We are using AI to be able to be reactive.”16 Reuters also reported that around 70% of Zalando’s editorial campaign images were AI-generated in a recent quarter.17

For packaged goods marketing, Reuters reports Mondelez aims to cut marketing content production costs by 30–50% using a genAI tool it has invested $40M+ in—while keeping humans in review and establishing rules prohibiting offensive stereotypes and other problematic content.18 This is an important “control case”: faster creative production with explicit guardrails.

Media and games: AI agents in production—with IP and ownership risk

In game development, Reuters reports a Google Cloud survey showing 87% of videogame developers are using AI agents; 44% use agents to optimize/process content (text, voice, code, audio, video); and 63% are concerned about data ownership.19 This is creativity under cost pressure: AI agents increase throughput, but IP and ownership remain strategic risks.

On the creator economy side, Adobe reports 86% of creators use creative generative AI; 81% say it helps them create content they otherwise couldn’t; and 69% are concerned about their content being used to train AI without permission.20 Adobe’s framing is telling: “creative decisions remain firmly in the creators’ hands.”21 The market is moving toward AI as production infrastructure, but trust and control are becoming differentiators.

The learning paradox in AI-enabled work

Removing routine tasks can free creativity, but routine tasks are also how people learn—how they build tacit knowledge, pattern recognition, and judgment.

Routine work as cognitive training data for humans

A Microsoft Research/CHI 2025 paper frames a classic automation risk: when automation mechanizes routine tasks and leaves exceptions to humans, it can “deprive the user of the routine opportunities to practice their judgement and strengthen their cognitive musculature,” leaving them unprepared when exceptions arise.22 That maps directly to knowledge work and R&D: exceptions are where breakthroughs live, and also where failures become expensive.

The OECD’s evidence review makes a similar point in different terms: overreliance can “negatively affect critical thinking and hinder cognitive development over time,” especially if users fail to critically analyze AI output, and it notes concerns about long-term skill retention when users rely on AI-generated answers without engaging underlying concepts.23

Designing for productive struggle and deep understanding

Education research often uses the phrase “productive struggle.” An OECD education analysis warns that “productive struggle [is] essential for learning,” and it argues that removing that struggle can lead to more superficial understanding and even “metacognitive laziness.”24

In executive terms: if AI always supplies the next step, employees may get facts without feel—answers without intuition. Early gains can be strong because AI helps people identify and patch gaps quickly. But long-term creativity depends on deeper mental models—built through repeated reasoning, not just receiving outputs.

This is not an argument against automation. It’s an argument for sequencing and role design: automate the clerical friction aggressively, but preserve (and even formalize) the human reasoning loop as part of the workflow.

Building an AI-enabled R&D operating model that preserves depth

If you want AI to improve innovation over multiple years, not just quarters, design around two loops:

A two-loop system: execution and learning

  • Execution loop: AI accelerates retrieval, drafting, prototyping, and first-pass analysis.
  • Learning loop: humans must explain, validate, and generalize—turning AI outputs into durable knowledge.

Practical mechanisms that reinforce the learning loop:

  • Require a short “reasoning note” for decisions in high-risk workflows (credit exceptions, clinical decisions, model changes).
  • Use AI as a critic (“What assumptions am I missing?” “What would falsify this?”) rather than a pure author.
  • Build escalation playbooks where humans practice exception-handling—so anomaly response isn’t a cold start.

Guardrails that protect novelty and auditability

Borrow from regulated industries: the OCC explicitly emphasizes governance and risk management in AI implementation.25 And central banks are openly discussing ranges of outcomes and risks as AI spreads through the economy.26

Guardrails that tend to work in practice:

  • Curate internal sources (the “truth set”) and measure retrieval quality.
  • Make provenance visible: what policies, documents, and data sources were used.
  • Implement human review thresholds (e.g., anything customer-facing or safety-sensitive).
  • Create “red team” routines for hallucination, bias, and policy violations.

Metrics that detect expertise erosion early

Most AI programs measure throughput: hours saved, tickets closed, drafts produced. Add capability metrics:

  • Exception readiness: time-to-resolution on rare but high-impact issues.
  • Quality drift: rework rates, audit findings, post-release defects.
  • Reasoning depth: proportion of decisions with documented rationale; peer review outcomes.
  • Novelty index: diversity of proposed solutions, not just volume of ideas.

If those metrics degrade while productivity rises, you’re “buying speed with expertise.”

Challenges and pitfalls

AI in R&D and creative discovery fails most often in predictable ways:

  • Governance and privacy: internal knowledge assistants are only as safe as their data controls and entitlements.27
  • Explainability and validation: fast answers without traceability increase risk in regulated contexts.
  • Bias and “creative sameness”: uncritical reuse of AI output can converge thinking; over-trust reduces critical engagement.28
  • Over-automation: the most subtle risk—teams lose practice, so performance collapses in edge cases.29
  • Change management: adoption without training produces shallow usage; the OECD notes skills and how AI is used matter for long-run outcomes.30

What pragmatic leaders do next

  1. Pick three workflows, not three tools. Choose workflows with measurable cycle-time pain and clear owners (e.g., payments exception handling, loan onboarding docs, clinical note drafting, product spec Q&A).
  2. Stand up lightweight oversight. One cross-functional group to set policy, measure outcomes, and stop unsafe use cases early.31
  3. Train for intuition, not just prompting. The OECD notes skills constraints can slow adoption and that most jobs don’t require advanced AI-specific skills—but do require foundational capabilities and thoughtful integration.32 Build training that strengthens domain judgment, verification, and reasoning, not just “how to ask.”

The strategic goal isn’t to have the most AI. It’s to have the most learning per unit time—because that’s what compounds into better R&D, better creativity, and better decisions.

Footnotes

  1. Flavio Calvino, Jelmer Reijerink, and Lea Samek, The Effects of Generative AI on Productivity, Innovation and Entrepreneurship, OECD Artificial Intelligence Papers (OECD, June 2025), https://www.oecd.org/content/dam/oecd/en/publications/reports/2025/06/the-effects-of-generative-ai-on-productivity-innovation-and-entrepreneurship_da1d085d/b21df222-en.pdf. ↩︎
  2. Yuliya Zhestkova, Sevgi K. S. Yildirim, et al., “The Impact of Generative AI on Work Productivity,” Federal Reserve Bank of St. Louis—On the Economy, February 27, 2025, https://www.stlouisfed.org/on-the-economy/2025/feb/impact-generative-ai-work-productivity. ↩︎
  3. Boston Consulting Group, Executive Perspectives: AI-Powered R&D (BCG, February 14, 2025), https://www.bcg.com/assets/2025/executive-perspectives-ai-powered-r-and-d-14feb.pdf. ↩︎
  4. Office of the Comptroller of the Currency, Semiannual Risk Perspective: Fall 2025 (Washington, DC: Office of the Comptroller of the Currency, December 2025), https://www.occ.gov/publications-and-resources/publications/semiannual-risk-perspective/files/pub-semiannual-risk-perspective-fall-2025.pdf.  ↩︎
  5. nCino, “nCino Unveils Transformative AI-Powered Banking Solutions at nSight 2025,” press release, May 20, 2025, https://www.ncino.com/news/ncino-new-ai-powered-banking-solutions-nsight.  ↩︎
  6. nCino, “nCino Unveils Transformative AI-Powered Banking Solutions at nSight 2025,” press release, May 20, 2025, https://www.ncino.com/news/ncino-new-ai-powered-banking-solutions-nsight. ↩︎
  7. Bank of America, “BofA’s New GenAI Assistant Transforms Global Payments Solutions,” press release, September 30, 2025, https://newsroom.bankofamerica.com/content/newsroom/press-releases/2025/09/bofa-s-new-genai-assistant-transforms-global-payments-solutions.html.  ↩︎
  8. Bank of America, “BofA’s New GenAI Assistant Transforms Global Payments Solutions,” press release, September 30, 2025, https://newsroom.bankofamerica.com/content/newsroom/press-releases/2025/09/bofa-s-new-genai-assistant-transforms-global-payments-solutions.html. ↩︎
  9. Reuters, “Citigroup’s AI Usage Frees Up 100,000 Hours for Developers a Week,” October 14, 2025, https://www.reuters.com/business/citigroups-ai-usage-frees-up-100000-hours-developers-week-2025-10-14/.  ↩︎
  10. Wells Fargo, “Wells Fargo Announces Expansion of Strategic Relationship with Google Cloud,” newsroom release, August 5, 2025, https://newsroom.wf.com/news-releases/news-details/2025/Wells-Fargo-announces-expansion-of-strategic-relationship-with-Google-Cloud/default.aspx. ↩︎
  11. Deloitte, “2025 Smart Manufacturing Survey,” Deloitte Insights, June 3, 2025, https://www.deloitte.com/us/en/insights/industry/manufacturing/2025-smart-manufacturing-survey.html. ↩︎
  12. Alex Wallace, “Generative AI Slashes Design Time,” ASSEMBLY, February 27, 2025.  https://www.assemblymag.com/articles/99091-generative-ai-slashes-design-time ↩︎
  13. Chris DeFrancesco, “Early Adopters Embracing AI Transcription Tool,” UConn Today, September 5, 2025, https://today.uconn.edu/2025/09/early-adopters-embracing-ai-transcription-tool/.  ↩︎
  14. Chris DeFrancesco, “Early Adopters Embracing AI Transcription Tool,” UConn Today, September 5, 2025, https://today.uconn.edu/2025/09/early-adopters-embracing-ai-transcription-tool/. ↩︎
  15. Lowe’s Companies, Inc., “Q3 2025 Earnings Call Transcript,” November 19, 2025, PDF, https://corporate.lowes.com/sites/lowes-corp/files/low-usq-transcript-2025-11-19.pdf. ↩︎
  16. Helen Reid, “Zalando Uses AI to Speed Up Marketing Campaigns, Cut Costs,” Reuters, May 7, 2025, https://www.reuters.com/business/media-telecom/zalando-uses-ai-speed-up-marketing-campaigns-cut-costs-2025-05-07/. ↩︎
  17. Helen Reid, “Zalando Uses AI to Speed Up Marketing Campaigns, Cut Costs,” Reuters, May 7, 2025, https://www.reuters.com/business/media-telecom/zalando-uses-ai-speed-up-marketing-campaigns-cut-costs-2025-05-07/. ↩︎
  18. Jessica DiNapoli, “Oreo-maker Mondelez to Use New Generative AI Tool to Slash Marketing Costs,” Reuters, October 24, 2025, https://www.reuters.com/business/media-telecom/oreo-maker-mondelez-use-new-generative-ai-tool-slash-marketing-costs-2025-10-24/. ↩︎
  19. Zaheer Kachwala, “Nearly 90% of Videogame Developers Use AI Agents, Google Study Shows,” Reuters, August 18, 2025, https://www.reuters.com/business/nearly-90-videogame-developers-use-ai-agents-google-study-shows-2025-08-18/. ↩︎
  20. Adobe, “[Media Alert] Inaugural Adobe Creators’ Toolkit Report,” October 28, 2025, https://news.adobe.com/news/2025/10/adobe-max-2025-creators-survey. ↩︎
  21. Adobe, “[Media Alert] Inaugural Adobe Creators’ Toolkit Report,” October 28, 2025, https://news.adobe.com/news/2025/10/adobe-max-2025-creators-survey. ↩︎
  22. Hao-Ping Lee et al., “The Impact of Generative AI on Critical Thinking: Self-Reported Reductions in Cognitive Effort and Confidence Effects From a Survey of Knowledge Workers,” paper presented at CHI ’25 (ACM CHI Conference on Human Factors in Computing Systems), April 26–May 1, 2025, https://www.microsoft.com/en-us/research/wp-content/uploads/2025/01/lee_2025_ai_critical_thinking_survey.pdf. ↩︎
  23. Flavio Calvino, Jelmer Reijerink, and Lea Samek, The Effects of Generative AI on Productivity, Innovation and Entrepreneurship, OECD Artificial Intelligence Papers (OECD, June 2025), https://www.oecd.org/content/dam/oecd/en/publications/reports/2025/06/the-effects-of-generative-ai-on-productivity-innovation-and-entrepreneurship_da1d085d/b21df222-en.pdf. ↩︎
  24. OECD, “Use of GenAI in Education: What Role Should Policy Play?” OECD Education and Skills Today (blog), January 19, 2026, https://www.oecd.org/en/blogs/2026/01/how-to-effectively-use-generative-ai-in-education.html. ↩︎
  25. Office of the Comptroller of the Currency, Semiannual Risk Perspective: Fall 2025 (Washington, DC: Office of the Comptroller of the Currency, December 2025), https://www.occ.gov/publications-and-resources/publications/semiannual-risk-perspective/files/pub-semiannual-risk-perspective-fall-2025.pdf. ↩︎
  26. Michael S. Barr, “Artificial Intelligence and Innovation,” speech, Board of Governors of the Federal Reserve System, November 11, 2025, https://www.federalreserve.gov/newsevents/speech/barr20251111a.htm. ↩︎
  27. Office of the Comptroller of the Currency, Semiannual Risk Perspective: Fall 2025 (Washington, DC: Office of the Comptroller of the Currency, December 2025), https://www.occ.gov/publications-and-resources/publications/semiannual-risk-perspective/files/pub-semiannual-risk-perspective-fall-2025.pdf. ↩︎
  28. Hao-Ping Lee et al., “The Impact of Generative AI on Critical Thinking: Self-Reported Reductions in Cognitive Effort and Confidence Effects From a Survey of Knowledge Workers,” paper presented at CHI ’25 (ACM CHI Conference on Human Factors in Computing Systems), April 26–May 1, 2025, https://www.microsoft.com/en-us/research/wp-content/uploads/2025/01/lee_2025_ai_critical_thinking_survey.pdf. ↩︎
  29. Hao-Ping Lee et al., “The Impact of Generative AI on Critical Thinking: Self-Reported Reductions in Cognitive Effort and Confidence Effects From a Survey of Knowledge Workers,” paper presented at CHI ’25 (ACM CHI Conference on Human Factors in Computing Systems), April 26–May 1, 2025, https://www.microsoft.com/en-us/research/wp-content/uploads/2025/01/lee_2025_ai_critical_thinking_survey.pdf. ↩︎
  30. Flavio Calvino, Jelmer Reijerink, and Lea Samek, The Effects of Generative AI on Productivity, Innovation and Entrepreneurship, OECD Artificial Intelligence Papers (OECD, June 2025), https://www.oecd.org/content/dam/oecd/en/publications/reports/2025/06/the-effects-of-generative-ai-on-productivity-innovation-and-entrepreneurship_da1d085d/b21df222-en.pdf. ↩︎
  31. Office of the Comptroller of the Currency, Semiannual Risk Perspective: Fall 2025 (Washington, DC: Office of the Comptroller of the Currency, December 2025), https://www.occ.gov/publications-and-resources/publications/semiannual-risk-perspective/files/pub-semiannual-risk-perspective-fall-2025.pdf. ↩︎
  32. OECD, “AI Skills: The Gap Is Smaller Than We Think,” OECD Education and Skills Today (blog), February 2, 2026, https://www.oecd.org/en/blogs/2026/01/making-ai-work-why-investing-in-skills-matters.html. ↩︎

Discover more from Mark T. Boyer: The Integration Edge – Leading Fintech & AI Innovation

Subscribe to get the latest posts sent to your email.

Leave a Reply

Discover more from Mark T. Boyer: The Integration Edge - Leading Fintech & AI Innovation

Subscribe now to keep reading and get access to the full archive.

Continue reading

Discover more from Mark T. Boyer: The Integration Edge - Leading Fintech & AI Innovation

Subscribe now to keep reading and get access to the full archive.

Continue reading