📈 economics

AI‑Powered Capabilities and Green New Product Development: The Dual and Asymmetric Mediating Roles of Green Exploratory and Exploitative

Drawing on dynamic capabilities theory and survey data from 333 Chinese manufacturing firms, this study reveals that AI-powered capabilities significantly enhance green new product development through dual mediating roles of green exploratory and exploitative processes, with the exploratory pathway demonstrating a significantly stronger mediating effect than the exploitative one.

Original authors: Jun Zhou, JianHui Chen, ChengZi Liu, RongFu Zhan, JunFeng Cao, Tao Liu, DeYu Zhong

Published 2026-07-08

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Jun Zhou, JianHui Chen, ChengZi Liu, RongFu Zhan, JunFeng Cao, Tao Liu, DeYu Zhong

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

The Big Picture: AI as the "Super-Scanner" for Green Inventions

Imagine a manufacturing company is like a giant chef trying to create a new, healthy, eco-friendly menu (Green New Product Development). In the past, this chef had to guess what ingredients to use or how to cook them.

Now, this chef has a Super-Scanner (AI). This AI can look at millions of data points—like weather patterns, customer complaints about pollution, or new government rules—and instantly spot trends.

The main question this paper asks is: Does having this Super-Scanner automatically make the chef create better green dishes?

The answer is yes, but not directly. The AI doesn't cook the meal itself. Instead, it gives the chef two different ways to work:

The "Explorer" Chef: Someone who uses the AI to find brand new ingredients and invent wild, never-before-seen recipes.
The "Optimizer" Chef: Someone who uses the AI to make the existing recipes cook faster, use less water, and taste more consistent.

This study found that while both chefs help the restaurant succeed, the "Explorer" Chef gets much better results when using the AI.

The Two Ways AI Helps (The Mediators)

The researchers looked at how the AI actually helps the company. They found two distinct "paths" or "routines":

1. Green Exploration (The "Treasure Hunter")

What it is: This is about searching for things you've never seen before. It's risky. It's about asking, "What if we made a car out of mushroom leather?"
How AI helps: The AI acts like a metal detector in a vast desert. It finds tiny, hidden signals (like a new trend in eco-friendly packaging) that a human would miss. It lowers the fear of trying something new because the AI can simulate the result before you spend real money.
The Paper's Claim: The AI is really good at helping companies find these new, radical ideas.

2. Green Exploitation (The "Efficiency Expert")

What it is: This is about taking what you already have and making it slightly better. It's about asking, "How can we make our current solar panel 5% cheaper to produce?"
How AI helps: The AI acts like a precision mechanic. It tweaks the assembly line, fixes small errors, and makes sure the recycling process runs smoothly.
The Paper's Claim: The AI does help here too, but it's more like a standard tool that just makes things run smoother.

The Big Surprise: The "Explorer" Wins

The most important finding in this paper is what the researchers call "Asymmetric Mediation."

Think of it like this:

If you give a flashlight (AI) to a Treasure Hunter (Explorer), they can find a diamond they would have never seen in the dark. The flashlight changes everything for them.
If you give that same flashlight to a Mechanic (Optimizer) who is already working in a well-lit workshop, it helps them see better, but they were already doing a good job. The flashlight is helpful, but it doesn't change their life as much.

The Result: The study found that the "Treasure Hunter" path (Exploration) is significantly stronger than the "Mechanic" path (Exploitation).

The AI's ability to spot new opportunities (Sensing) is its superpower.
The AI's ability to just make things faster (Optimization) is good, but it's not the main reason companies get huge green breakthroughs.

In numbers: The "Explorer" path explained about 23% of the success, while the "Optimizer" path explained about 16%. The "Explorer" path was the clear winner.

Why Does This Happen? (The Three Reasons)

The paper gives three reasons why the AI helps the "Explorer" so much more than the "Optimizer":

Finding the Invisible: AI is amazing at spotting weird patterns in huge piles of data. This is exactly what an "Explorer" needs to find new ideas. An "Optimizer" just needs to follow a rule, which is easier for humans to do anyway.
Lowering the Risk of Failure: Trying new green ideas is scary and expensive. AI can run thousands of virtual simulations to say, "This idea might work," which makes companies brave enough to try it. This "bravery boost" is huge for exploration but less critical for tweaking existing products.
Being First: In the green market, being the first to invent a new eco-product is worth a lot of money. AI helps you get there first. Just making an old product slightly cheaper (Optimization) doesn't give you that same "first-mover" advantage.

What Should Companies Do? (The Takeaway)

Based only on what this paper says, here is the advice for managers:

Don't just buy AI to save money on the assembly line. While that's nice, it's not the most powerful use of the technology for green innovation.
Buy AI to find new ideas. Invest in tools that help you scan the market for new trends, predict what customers will want next, and simulate radical new product designs.
Balance, but lean toward the new. You still need to improve your current products (Exploitation), but if you have limited money, spend more on the "Treasure Hunting" (Exploration) side because the AI makes that path much more effective.

Summary

This paper is like a map showing that while AI is a useful tool for both fixing old problems and inventing new ones, it is a magic wand for invention. If you want to create amazing new green products, you should use AI to help you explore the unknown, not just to polish the known.

Technical Summary: AI‑Powered Capabilities and Green New Product Development

Problem Statement and Research Gaps
The convergence of artificial intelligence (AI) and environmental sustainability represents a critical frontier in innovation management, yet the mechanisms through which AI capabilities translate into Green New Product Development (Green NPD) remain theoretically underdeveloped. While prior research has largely focused on direct performance outcomes (e.g., cost savings, efficiency), it often treats AI as an isolated technological input, overlooking the internal organizational routines that convert AI-generated insights into tangible product results. This study identifies three specific gaps in the literature:

The Mediation Black Box: Existing studies rarely unpack the process architectures through which AI influences Green NPD, implicitly assuming benefits materialize without organizational intermediation.
Undifferentiated Treatment of Innovation Modes: Research often fails to distinguish between green exploratory (search, discovery, radical novelty) and green exploitative (refinement, efficiency, incremental improvement) innovation activities, despite their distinct operational requirements.
Unexamined Asymmetry: Even when multiple mediators are considered, studies typically assume symmetric effects. This paper questions whether AI's specific capabilities (pattern recognition, uncertainty reduction) might make one mediation pathway significantly stronger than the other.

Theoretical Framework and Hypotheses
Grounded in Dynamic Capabilities Theory (DCT), the study posits that AI-powered capabilities do not directly drive Green NPD but rather act as a catalyst for specific adaptive processes. The framework distinguishes between:

Green Exploratory: A sensing-driven process where AI enhances opportunity identification and the discovery of novel low-carbon trajectories.
Green Exploitative: A seizing/transforming-driven process where AI optimizes execution, refines existing green technologies, and improves process efficiency.

The study proposes a dual mediation model with an asymmetric hypothesis:

H1a & H1b: Both green exploratory and green exploitative mediate the positive relationship between AI-powered capabilities and Green NPD.
H1c (Asymmetry): The mediating effect of green exploratory is significantly stronger than that of green exploitative. This is theorized based on AI's comparative advantage in sensing augmentation, its disproportionate ability to reduce the high uncertainty associated with radical innovation, and the strategic value of first-mover advantages in volatile low-carbon markets.

Methodology

Data Collection: The study utilized cross-sectional survey data from 333 manufacturing firms in China across five high-priority sectors: electronics/semiconductors, automotive, machinery, pharmaceuticals/medical devices, and home appliances/consumer electronics.
Sampling Criteria: Firms were selected based on their implementation of Green NPD practices and the adoption of AI technologies for low-carbon operations or decision-making.
Measures: Constructs were measured using multi-item scales adapted from established literature (e.g., Gupta et al., 2023; Singh et al., 2024) on a 7-point Likert scale. Key variables included AI-powered capabilities, green exploratory, green exploitative, and Green NPD performance. Control variables included industry sector, firm age, size, AI investment, and ISO 14001 certification.
Analysis: The study employed hierarchical regression and bootstrapped mediation analysis (PROCESS Model 4 with 5,000 resamples) to test hypotheses. Common method bias was assessed via Harman's single-factor test and CFA comparisons, while non-response bias was checked by comparing early and late respondents.

Key Results

Dual Mediation Confirmed: Both green exploratory and green exploitative significantly mediate the relationship between AI-powered capabilities and Green NPD.
- AI capabilities positively predict green exploratory ( $\beta = 0.624, p < 0.001$ ) and green exploitative ( $\beta = 0.578, p < 0.001$ ).
- Both mediators positively predict Green NPD ( $\beta_{exploratory} = 0.375, p < 0.001$ ; $\beta_{exploitative} = 0.289, p < 0.001$ ).
Asymmetric Mediation: The indirect effect via green exploratory is significantly stronger than via green exploitative.
- Indirect effect (Exploratory): $\beta = 0.234$ , 95% CI [0.139, 0.354].
- Indirect effect (Exploitative): $\beta = 0.167$ , 95% CI [0.093, 0.258].
- The confidence intervals exhibit partial non-overlap, supporting the hypothesis of asymmetry.
Model Fit: The full model explains 54% of the variance in Green NPD ( $R^2 = 0.540$ ).

Key Contributions and Significance
The paper claims three primary contributions:

Unpacking the Black Box: By specifying two distinct process architectures (exploration and exploitation), the study clarifies how AI generates value in Green NPD, moving beyond direct-effect models to a process-oriented view grounded in DCT.
Challenging the "Balance Assumption": The findings challenge the traditional ambidexterity assumption that firms must balance exploration and exploitation equally. Instead, the study demonstrates the primacy of exploratory mediation in AI-enabled green innovation contexts, suggesting that AI's unique capabilities are more potent in sensing and discovery than in incremental refinement.
Methodological Template: The study provides a rigorous template for testing dual and asymmetric mediation using bootstrapped confidence interval comparisons, offering a replicable method for future research comparing competing mediation pathways.

Practical Implications

For Managers: Firms should prioritize AI investments that enhance sensing and exploratory capabilities (e.g., predictive analytics for emerging markets, generative design) over purely optimization-focused tools, though a balanced but exploration-skewed approach (e.g., 60–70% to exploration) is recommended.
For AI Vendors: Product development should differentiate between exploration-oriented tools (generative design, weak signal detection) and exploitation-oriented tools (process optimization, predictive maintenance).
For Policymakers: Subsidies and regulatory frameworks should be recalibrated to support higher-risk, AI-enabled exploratory green innovation, which yields greater performance gains per unit of investment than incremental exploitation.

Limitations and Future Directions
The authors acknowledge limitations including the cross-sectional design (limiting causal inference), reliance on single-respondent data, and the specific context of Chinese manufacturing. Future research is suggested to employ longitudinal designs, investigate behavioral micro-foundations, and examine the interplay of AI with other Industry 4.0 technologies (IoT, blockchain). Additionally, the study notes that the proposed moderation effects of technology uncertainty were not fully tested in this manuscript and remain an avenue for future analysis.