Research opportunity · Organizational absorption: Why micro gains do not become macro outcomes
Is there a generative-AI productivity J-curve inside firms?
Question
In a multi-year panel of firms, does deeper integration of generative AI, together with measured complementary investment (process redesign, training, data and capitalized software), show an early dip and a later rise in measured productivity, and do firms that adopt without that investment show neither?
Not yet measuredA PhD projectLow to medium costPartly addressed
Why it matters
The absorption-lag position (P1) rests on the J-curve, well grounded for computers but not yet measured for generative AI. Among S&P 500 firms, profit margins dip at pilot stages and rise with deep integration, yet revenue per employee shows no difference; industry-level productivity correlates with AI adoption only since 2020, and the authors stress that this is not causal. A firm panel that measures the complementary investment directly would tell a lag from a low ceiling.
The debate lists this among what we still do not know: Are the positive industry-level correlations causal, and do they survive longer series?
Evidence that motivates it
- In S&P 500 annual reports, 11% of firms had AI deeply integrated into their business processes in 2025 and a further 10% used it in production, against 5% in 2022; among non-technology firms profit margins dipped at the pilot stages and rose with deep integration, while revenue per employee showed no difference at the advanced stages.
In S&P 500 annual reports, profit margins of non-technology firms dipped at pilot stages and rose with deep integration, while revenue per employee showed no difference at advanced stages. - Brynjolfsson, Rock and Syverson model how general-purpose technologies such as AI need large, poorly measured intangible investments, so measured productivity growth is first understated and later overstated (a 'productivity J-curve'); adjusting for intangibles tied to computers put US total factor productivity 15.9% above official measures by the end of 2017.
The productivity J-curve: general-purpose technologies need large, poorly measured intangible investment, so measured productivity is first understated and later overstated. - Reviewing evidence on computers, Brynjolfsson and Hitt argue that organizational investments, such as new work practices, strongly shape the value of information-technology investments, and that the benefits are often intangible and hard to measure.
For computers, organizational investment such as new work practices strongly shaped the value of IT investment. - Across European and US industries, a 10-percentage-point rise in AI adoption is associated with 2 to 5 points of extra cumulative productivity growth in Europe to 2024 and about 3 points relative to trend in the US since 2022, with no such association in the European placebo period (2015–2019); the authors stress that these correlations are not causal.
Across European and US industries, AI adoption is associated with extra productivity growth since 2020 but not before; the authors stress that the correlation is not causal. - In a survey of nearly 750 corporate executives, the authors find positive labour-productivity gains from AI, reflecting revenue-based total factor productivity and expected to strengthen in 2026, but report a productivity paradox: executives perceive larger gains than are measured, which the authors attribute to a delay in revenue realizations.
Executives perceive larger productivity gains from AI than their firms' data show and attribute the difference to a delay in revenue.
A possible design
Build a firm panel from 2022 onwards linking (a) depth of generative-AI integration, coded from annual reports and earnings calls or taken from business-survey microdata, (b) complementary intangible investment, measured from capitalized software, training expenditure, data-infrastructure spending and postings for process-redesign roles, and (c) labour productivity and revenue-based total factor productivity. Estimate event studies around the move to deep integration, with heterogeneity by complementary investment, and use earlier non-generative AI adoption as a placebo. Update the panel annually, since a lag needs years to show.
Data needed
Firm accounts (Compustat, Orbis or national registers); annual-report and earnings-call text; job-posting data; where possible, restricted business-survey microdata through a secure data centre.
Feasibility
A PhD project. Low to medium; most data are licensed or restricted rather than costly to collect, and the main cost is the years needed for the lag to appear. Capitalized software and training spend are reported inconsistently across countries and firm sizes, so the intangible measures need validation against a subsample with survey data. A first paper can test the dip on 2022–2026 data and leave the recovery to later updates.
Has this been done?
Partly done already: related work covers some of this question (first-pass screen of 15 October 2026; not yet checked in depth).
The J-curve design has been applied to AI in general with firm panels, and once to generative AI with financial outcomes, but no multi-year panel was found that measures generative-AI integration specifically, links it to measured complementary intangible investment, and tracks productivity (not profitability) over time to test for a dip and a later gain. Existing work uses all-AI measures, uses profitability or return on equity without intangible measures, or infers organization capital from job text rather than investment data.
How strong this evidence is
- Who searched and checked: an independent AI reviewer with fresh context (not a person), 15 October 2026 (review of PR #77): verdicts, query counts and 13 abstract-read labels checked; label slips corrected. Every check so far was done by AI; no person or domain expert has checked it yet.
- What was read: 4 of the 6 nearest works below were judged from an abstract, summary or search listing only, and 1 has details that could not be confirmed; each is marked. Not every entry says how much of the work was read.
- Where the search could not reach: Screened by web search (7 queries) and arXiv, NBER and BIS abstract pages on the date shown; central-bank, OECD and national statistical-office microdata projects and the accounting literature on software capitalization were not searched in depth. An earnings-call firm panel presented at a 2026 Bank of Finland conference, which treats generative AI as intangible investment, was seen in listings and not read. Absence of a hit is not proof that the work does not exist.
- Before building a project on it, run your own search in a scholarly database such as Scopus, Web of Science, Google Scholar or SSRN.
Nearest work found
- Babina, He & Jiang, "Canaries in the Gold Mine: Early Productivity Gains from Artificial Intelligence Creating Organization Capital," NBER Working Paper 35684 (2026) Judged from an abstract, summary or search listing only
Abstract read. A firm-level AI investment measure based on AI-skilled employment, spanning machine learning to generative AI: productivity gains appear in recent years but not in the previous decade, consistent with a J-curve, and are traced to organization capital measured from job descriptions. Not specific to generative AI, and no direct measure of intangible spending. - Kikuchi, "The Innovation Tax: Generative AI Adoption, Productivity Paradox, and Systemic Risk in the U.S. Banking Sector," arXiv:2602.02607 (2026) Judged from an abstract, summary or search listing only
Abstract read. A panel of 809 US banks, 2018–2025, linking generative-AI adoption disclosed in filings to regulatory data around the release of ChatGPT: an early fall in return on equity. One sector and one author; profitability rather than productivity, and no complementary-investment measures. - Yu, Fleming, Hampton, Combemale & Thompson, "AI Adoption in S&P 500 Firms," arXiv:2607.08920 (2026)
Full text read for the LD-03 debate (LD03-SL-C01). Adoption depth coded from annual reports, 2016–2025; a profit-margin pattern across adoption stages but no difference in revenue per employee at advanced stages. A cross-section by adoption stage, without intangible investment. - Aldasoro, Gambacorta, Pál, Revoltella, Weiss and Wolski, "AI adoption, productivity and employment: evidence from European firms," BIS Working Paper 1325 (2026) Judged from an abstract, summary or search listing only
Abstract read. Survey and accounts data on more than 12,000 European firms with an instrument from US peers' adoption: AI raises labour productivity in the short run, and complementary investment in software, data and training matters. AI in general and mainly short run, with no dynamic J-curve test. - McElheran, Yang, Kroff & Brynjolfsson, "The Rise of Industrial AI in America: Microfoundations of the Productivity J-curve(s)," US Census Bureau CES Working Paper 25-27 (2025) Judged from an abstract, summary or search listing onlySome details not confirmed
Judged from search listings; not read directly (the Census site could not be opened). Reportedly manufacturing microdata for 2017 and 2021 showing short-run productivity and profit losses before later gains. Industrial AI, not generative AI. - Fang, Yuan, Zhang, Donati & Sarvary, "Generative AI and Sales Productivity: Field Experiments in Online Retail," arXiv:2510.12049 (v6, 2025)
Full text read for the LD-03 debate (LD03-SL-C15). Randomized experiments on seven workflows at one platform, with sales effects from none to 16.3%. Short-term and within one firm, so no panel or intangible measures.
Search queries used
generative AI firm-level panel intangible investment productivity J-curvegenerative AI adoption firms productivity training complementary investment panel evidence 2025Babina AI investing firms productivity 2018-2024 J-curve employees AI implementation working paper"generative AI" firm productivity panel data software capitalization intangible capital earnings calls arXivAldasoro "AI adoption, productivity and employment: evidence from European firms"generative AI productivity J-curve firms 10-K profit margin revenue per employee arXivgenerative AI firms Census Business Trends and Outlook Survey productivity panel worker training process changes
Provenance
RO-LD03-04 v1. Drafted by Claude (SocioLens Project Lead), AI-drafted from the LD-03 debate on 2026-10-15. Checked by an independent AI reviewer with fresh context (not a person), 15 October 2026 (review of PR #77): verdicts, query counts and 13 abstract-read labels checked; label slips corrected. Not yet reviewed by a domain expert. This is a research hypothesis, not a finding.