Brazilian manufacturers reap innovation gains only after full digital upgrades: firms deploying a full suite — Big Data, Cloud, AI, IoT, additive manufacturing and robotics — are significantly more likely to introduce product and process innovations, while piecemeal technology adoption shows no reliable benefit.

Main Finding

The innovation payoff from digitalization in Brazilian industrial firms is non-linear: adopting a full, complementary portfolio of advanced digital technologies (Big Data, Cloud, AI, IoT, Additive Manufacturing, Robotics) is positively associated with product and/or process innovation — and with the joint occurrence of both — whereas the mere adoption of one or a few isolated technologies does not consistently increase innovation probability. Instrumental-variable estimates indicate endogeneity is concentrated in partial adoption, while comprehensive portfolios behave like a more stable strategic commitment.

Key Points

• Three propositions tested:
  • P1: Broad, complementary portfolios yield stronger innovation effects than isolated technology adoption.
  • P2: Digitalization is more strongly associated with combined product-and-process innovation than with narrowly separated outcomes.
  • P3: Partial adoption is more likely endogenous to prior innovation behavior; full portfolios reflect deliberate transformation strategies.
• Technologies analyzed: Big Data, Cloud Computing, Artificial Intelligence (AI), Internet of Things (IoT), Additive Manufacturing, Robotics.
• Main empirical pattern: a threshold/complementarity effect — innovation gains appear when firms adopt the full set of six technologies, not when they adopt only one or a few.
• Endogeneity asymmetry: partial adopters show evidence of endogeneity (selection on unobservables), while full-portfolio adoption is more robust to endogeneity concerns in iv estimates.
• Policy implication emphasized by authors: support systemic, integrated digital upgrading rather than piecemeal technology subsidies.

Data & Methods

• Sample: 1,435 Brazilian industrial firms with 100+ employees.
• Data sources: two 2022 PINTEC Semestral survey rounds (innovation and advanced digital technologies) merged with selected variables from the 2022 PIA-Enterprise (IBGE).
• Key independent indicators:
  • Binary indicator for adoption of at least one advanced digital technology.
  • Binary indicator for simultaneous adoption of all six listed advanced technologies (the "full digital portfolio").
• Dependent variables: product innovation, process innovation, and joint product-and-process innovation (binary outcomes from innovation survey).
• Estimation strategy:
  • Descriptive analysis to profile digitalization and innovation.
  • Probit models to estimate the association between digital adoption and innovation outcomes.
  • Instrumental-variable probit (ivprobit) models to address potential endogeneity (paper reports ivprobit results and finds asymmetry in endogeneity across partial vs full adoption). (Specific instruments are used in the paper; they are not detailed in the supplied abstract.)
• Controls and robustness: models include standard firm-level controls and sectoral fixed effects (as typical in such analyses); findings are reported as robust across baseline and IV specifications for full-portfolio adoption.

Implications for AI Economics

• Complementarities matter for AI’s innovation impact:
  • AI is unlikely to deliver consistent innovation returns when deployed in isolation. Its effective contribution depends on complementary technologies (Big Data, Cloud, IoT) and physical automation (Robotics, Additive Manufacturing).
• Measurement and empirical design:
  • Treat digitalization as multidimensional; analyses that collapse AI adoption into a binary indicator risk missing threshold effects and complementarities.
  • Look for non-linearities and threshold regimes (portfolio adoption) rather than only marginal effects of single technologies.
  • Account for heterogeneous endogeneity: selection into partial AI adoption may reflect pre-existing innovativeness, while comprehensive AI-centered transformations may be less endogenous.
• Policy and industrial strategy:
  • Policies aimed at fostering AI-driven innovation should prioritize systemic packages: data infrastructure, cloud access, integration with production technologies, workforce skills, and R&D capabilities, rather than small isolated AI grants.
  • Public support could focus on enabling complementarities (e.g., co-investments in data platforms, interoperability standards, training, and adoption pathways) to help firms cross the complementarity threshold.
• Research agenda:
  • Investigate mechanisms through which AI interacts with other digital and production technologies (data pipelines, organizational change, absorptive capacity).
  • Use panel data and causal designs to track dynamics of portfolio adoption and downstream productivity effects.
  • Explore heterogeneity by firm size, sector, and absorptive capacity: SMEs and less-capable firms may face higher barriers to reaching the complementarity threshold.

Limitations to note: cross-sectional 2022 firm-level data focused on large firms (100+ employees) in Brazil; results may not generalize to SMEs or other countries without considering institutional and capability differences.