Skip to content
back to the archive page
#AI

The GenAI Divide: State of AI in Business 2025 (Category AI)

#AI #Engineering #Software #Metrics #Leadership #ML

Yesterday I read an interesting report on AI adoption in industry by the MIT NANDA team: Aditya Challapally, Chris Pease, Ramesh Raskar and Pradyumna Chari, published in July 2025. NANDA stands for Networked Agents and Decentralized Architecture, an MIT research initiative developing infrastructure for distributed intelligence in autonomous agent systems. The research ran from January to June 2025, and this is a preliminary report on generative AI adoption in business.

Across 26 pages, the authors describe a gap between widespread interest in GenAI and actual business results. They call it the “GenAI Divide.” Their figures tell the following story:

1. A gulf in outcomes Despite $30–40 billion in global investment, roughly 95% of organisations saw no measurable return from GenAI projects, while the remaining 5% made money. The authors attribute this to differing implementation approaches rather than model quality or regulation. That is encouraging, because it can be changed.

2. Much experimentation, little ROI More than 80% of companies experimented with tools such as ChatGPT and GitHub Copilot, and almost 40% deployed them for employees. These tools mainly improved individual productivity, with little effect on key organisational financial measures. GenAI is being tried in workflows, but has not yet transformed business at scale.

3. Corporate pilots fail to scale The enterprise-project funnel is 60% → 20% → 5%: 60% evaluated custom or vendor GenAI solutions, 20% reached a pilot and 5% reached production. Most pilots remained stuck. The main reasons were inflexible, “brittle” workflows, systems unable to incorporate feedback and context, and weak integration into daily operations.

4. The GenAI Divide The study identifies four patterns separating successful and unsuccessful organisations:

  • Limited sector-wide impact: only 2 of 8 sectors — technology and media & telecom — were undergoing structural change from AI.
  • Large corporations ran many pilots but lagged in scaling them.
  • Budgets favoured visible front-office functions such as sales and marketing over back-office processes that often offered better returns.
  • Projects with external technology partners were twice as successful as internal development.

The authors identify inability to learn as the main scaling barrier: most systems do not retain context, adapt or improve through use. A small group crossed this divide by tailoring systems to their processes, integrating AI tightly into everyday work and measuring business outcomes rather than technical model metrics from the outset.

The research covered the first half of 2025 and used several data sources:

  1. A systematic review of 300+ publicly announced AI initiatives to identify trends and project outcomes.
  2. 52 structured interviews with representatives of organisations adopting GenAI, providing qualitative insights from practice.
  3. Questionnaires from 153 senior executives, including C-level leaders, collected at four major industry conferences.

Combining case analysis, interviews and surveys allowed the authors to examine GenAI adoption from several perspectives.

#AI #Engineering #Software #Metrics #Leadership #ML #Metrics