Despite the hype, most generative‑AI pilots fail to generate profit and loss (P&L) impact. Drawing on the MIT Project NANDA report, the economic theories of William Janeway and my own consulting work, this section explores why and how to change that.
The productivity J‑curve
Janeway’s innovation‑economy framework treats general‑purpose technologies (GPTs)—steam, electricity, semiconductors and today’s generative AI—as requiring learning complements before they deliver productivity. Early adoption is costly and often looks wasteful, but the “visible waste” is tuition for building:
- Memory and knowledge management – Agents and models need context windows, retrieval pipelines and feedback loops to learn from experience.
- Standards and protocols – The Model Context Protocol (MCP) and Agent‑to‑Agent (A2A) frameworks propose deterministic message types and provenance tags so agents can interoperate safely.
- Governance and integration – Embedding GPTs into existing workflows requires domain experts working alongside ML engineers; strong leadership and iterative feedback trump raw IQ.
Pilot‑to‑Production realities
Project NANDA surveyed enterprises and found that only about 5 % of GenAI pilots reach production, and even fewer show measurable P&L improvements. Key takeaways:
- Define outcome metrics – Focus on cycle time, throughput and quality rather than vanity metrics. Contract with vendors based on outcomes.
- Start with narrow “learning wedges” – Pick small, high‑value tasks that teach you about your data, workflows and user needs. Use shadow AI to explore before committing.
- Iterate and scale cautiously – After two or three consecutive improvements, expand scope. Most ROI comes from back‑office processes, not flashy customer‑facing features.
The role of the state and finance
Janeway emphasises that state investment seeds research (e.g., funding the Manhattan Project or NANDA), speculative finance builds infrastructure, and entrepreneurial firms integrate technology into products. Today’s build‑out of AI data centers and supercomputer clusters looks extravagant, but it lays the groundwork for future capabilities. Organisations should understand where they sit in this triangle and align incentives accordingly.
This page will host essays synthesising NANDA findings, case studies from my consulting practice and macro‑economic commentary. I’ll provide downloadable strategy decks and link to further reading on Janeway and the innovation economy.
therenaissancequant.com 