The Biggest Barrier to AI's Firm-Level Impact Isn't Access — It's Knowing Where and How to Deploy It
May 8, 2026
In "Mapping AI into Production: A Field Experiment on Firm Performance" (March 2026), Hyunjin Kim and Dahyeon Kim of INSEAD and Rembrand Koning of Harvard Business School conducted a field experiment across 515 high-growth startups to answer a critical question: why do AI's well-documented task-level productivity gains fail to translate into firm-level performance improvements? The answer lies in what they call the "mapping problem" — the challenge of discovering where and how AI creates value within a firm's production process.
The experiment was run within a three-month global startup accelerator. All firms received identical resources — API credits, technical training, mentorship, and pitch opportunities. The treatment group additionally received case studies showing how AI-native companies had reorganized their entire production processes around AI, not just adopted it for isolated tasks. The control group received standard entrepreneurship case discussions instead.
The results were striking. Treated firms discovered 44% more AI use cases (2.7 additional applications), concentrated in product development and strategy rather than obvious tasks like drafting emails. This broader AI adoption produced measurable performance gains: 12% more completed tasks, 18% higher likelihood of acquiring paying customers, and 1.9x higher revenue. Revenue and investment gains were largest at the 90th percentile, suggesting AI expands the upper range of what firms can achieve rather than modestly lifting the average.
Treated firms achieved this faster growth without proportionally scaling inputs — demand for external capital fell by 39.5% relative to controls, and labor demand stayed flat. The gains were broad-based regardless of founder technical background or baseline firm performance.
Point of View
This research validates that enterprise AI impact operates on two distinct levels.
The first — and the one this paper illuminates — is the strategic mapping of high-impact use cases that cut across the way an organization operates: rethinking product development pipelines, customer acquisition workflows, and decision-making processes end to end. This is a leadership and vision challenge, not a technology one.
The second is enabling teams and individuals within functional areas to leverage AI where their needs are most dynamic. Functions across an organization — sales, operations, finance, product — constantly need to connect and draw from different systems, processes, and data, both structured and unstructured, from inside and outside the organization. These needs shift frequently, are difficult to anticipate centrally, and resist traditional top-down automation. This is where AI-driven automation will increasingly occur at the edge: individuals and small teams composing AI solutions on the fly to bridge data sources, accelerate decisions, and adapt to changing requirements in real time.
Companies that pursue only top-down mapping get strategies without execution. Companies that pursue only bottom-up enablement get incremental efficiency. The firms that win will do both simultaneously.
Source: Kim, H., Kim, D., & Koning, R. (2026). "Mapping AI into Production: A Field Experiment on Firm Performance." INSEAD Working Paper 2026/20/STR. Available at https://ssrn.com/abstract=6513481