US companies have poured tens of billions into generative AI initiatives, yet the vast majority remain stuck in pilot mode with little measurable impact. A new report from MIT’s NANDA initiative, cited by Computerworld, points to integration and learning shortfalls rather than model quality as the main reason projects fall short.
Findings from MIT’s NANDA initiative
According to Computerworld, US firms have invested between $35 billion and $40 billion in generative AI projects. Only about 5% of these efforts lead to rapid revenue growth, while most deliver little or no impact. The publication attributes the figures to a report from MIT’s NANDA initiative and notes that Fortune reported the same conclusion.
The report’s central takeaway is that the bottleneck is not the quality of the underlying models. Instead, projects commonly falter due to a lack of integration with existing systems, insufficient learning, and weak alignment with real corporate workflows. As a result, initiatives remain proofs of concept rather than scaling into production.
Where companies see returns
While many organizations have directed spending toward sales and marketing solutions, the biggest returns appear to be emerging in back-office automation and the streamlining of internal processes, according to the coverage. These areas seem to benefit more readily from structured integration and clear workflow alignment.
Approaches linked to better outcomes
The report also highlights differing strategies between companies that see impact and those that do not. Successful firms tend to purchase specialized solutions and form partnerships that help embed technology into operations. In contrast, in-house development efforts are reported to fail significantly more often.
Computerworld’s article underscores that moving from pilot to production depends on how well tools are integrated, how teams learn from early iterations, and how closely solutions map to day-to-day tasks. Without these elements, investment levels have not translated into broad revenue gains.
In sum, the findings frame the challenge less as a question of model performance and more as an operational issue: aligning generative AI with existing processes, choosing targeted solutions, and partnering to ensure adoption. As the report indicates, those factors separate the small set of projects driving rapid revenue growth from the many that remain stalled.
