New technologies demand new approaches. This is especially true with the emergence of a general-purpose technology (GPT) such as generative artificial intelligence (Gen AI). What defines a GPT, and what are the implications of Gen AI for process innovation? How should decision-making evolve in the context of Gen AI‑driven innovation?

The emergence of a GPT is a rare and transformative event. The steam engine powered the Industrial Revolution, serving as a centralized energy source on the factory floor. Electricity followed, making power ubiquitous and enabling new production methods and factory layouts. The invention of the semiconductor revolutionized workflows and organizational structures, enabling electronic computing and modern communication via the internet.

Gen AI became broadly accessible to enterprises after the release of ChatGPT in November 2022, when cloud service providers integrated large language models and related tools into their platforms. What can we learn from previous GPTs to guide the adoption of Gen AI?

Lessons from the Adoptions of Previous General-Purpose Technologies

Technology scholars identify three defining traits of a GPT:¹

• It is pervasive across applications.
• It is capable of continuous technical improvement.
• It drives co‑innovation in the application domain.

GPT adoption unfolds over time. As Gen AI advances, insurers will continue to co‑invent, expanding its use and generating revenue for further development – a feedback loop arises of downstream innovation and upstream reinvestment. The steam engine evolved over two centuries, with applications expanding steadily. Electricity advanced through innovations such as alternating current and the electric motor, and battery technology continues to evolve. Electronic computing has progressed through milestones such as the invention of graphics processing units by NVIDIA. In less than two years since ChatGPT’s debut, reasoning models have emerged, enabling Gen AI to generate new data² and solve complex mathematical problems³.

Three Dimensions of Process Innovation

Gen AI‑driven process innovation occurs across three dimensions, as illustrated below: task-level improvement, workflow redesign, and organizational redesign. These dimensions differ in the co‑innovation effort required and the economic value delivered. As the technology matures, adoption will shift from enhancing existing tasks to enabling new ones within redesigned workflows – some of which will be coordinated and executed by AI agents in novel organizational structures.

Process Innovation Dynamics of Gen AI

Decision-Making and Real Options

This dynamic interplay between technological progress and co‑innovation informs how decisions should be made. Gen AI adoption is iterative, involving continuous learning and deployment in new workflows and organizational models. The pace of this feedback loop has accelerated from steam to electricity to semiconductors – and is expected to be even faster with Gen AI, placing the insurance industry on a path of rapid learning and continuous innovation.

Each adoption decision carries risk: an error of omission (failing to pursue a valuable project) or an error of commission (pursuing a project that should have been avoided). Not pursuing a project is fully irreversible, while executing one often involves sunk costs – expenditures that cannot be recovered if the project is abandoned or altered.

This creates value in waiting. Delaying a decision preserves the option to act later with better information. This option value reduces the traditional net present value (NPV) of a project. However, waiting also has a cost: missed opportunities and reduced future flexibility. This cost of waiting increases the project’s value.⁴

Real options theory emphasizes learning and reversibility. It favors technologies that enable reversible decisions and expand future choices – both factors that support low-cost complementary innovation and early adoption. While real options are difficult to quantify in complex environments, their practical value lies in guiding decision-making: prioritize technologies that allow reversibility and broaden future opportunities.

The J‑Curve in Measured Productivity

Adopting a GPT requires complementary investments. For Gen AI, foundational investments include cloud infrastructure, data architecture, and organizational capabilities in data and AI engineering. At the workflow level, co‑innovation requires building competence in human‑AI collaboration and in coordinating workflows with AI agents.

These investments create intangible assets that initially function as outputs – products of the organization’s efforts. Over time, they become inputs to production. Traditional productivity metrics often fail to capture these assets, underestimating productivity during their formation and overestimating it once they are deployed.⁵

For insurers, a rough productivity measure is the expense ratio. While intangible assets are being developed, they increase expenses without contributing to output. Once deployed, they enhance output without adding to expenses. This dynamic can create a J‑Curve in measured productivity: initial decline followed by long‑term gains, as illustrated below. Recognizing this effect is essential for accurately assessing Gen AI’s impact on productivity in both the short and long term.

J‑Curve in Measured Productivity

Endnotes

1. See Timothy F. Bresnahan and Manuel Trajtenberg, “General Purpose Technologies: ‘Engines of Growth’?,” Journal of Econometrics 65 (1), 83‑108, 1995, https://www.sciencedirect.com/science/article/pii/030440769401598T. Working Paper (August 1992) at https://www.nber.org/system/files/working_papers/w4148/w4148.pdf.
2. An example of an AI system capable of generating new data is Google Deep Mind’s AlphaGo, which invented new moves in the game of Go. See Google Deep Mind, AlphaGo, https://deepmind.google/research/projects/alphago/.
3. See, for instance, Ben Cohen, “The High‑Schoolers Who Just Beat the World’s Smartest AI Models,” The Wall Street Journal, July 27, 2025, https://www.wsj.com/tech/ai/imo-gold-math-olympiad-google-deepmind-openai-2450095e.
4. See, for instance, Keith J. Leslie and Max P. Michaels, “The real power of real options,” The McKinsey Quarterly, 1997, Number 3, 97‑108, https://www.mckinsey.com/capabilities/strategy-and-corporate-finance/our-insights/the-real-power-of-real-options#/.
5. See Erik Brynjolfsson, Daniel Rock, and Chad Syverson, “The Productivity J‑Curve: How Intangibles Complement General Purpose Technologies,” American Economic Journal: Macroeconomics 13(1), 333-372, 2021, Working Paper (January 2020) at https://www.nber.org/papers/w25148.

Endnotes last accessed on 22 August 2025.