Source-linked AI summary
Artificial Intelligence and Strategic Decision-Making: Evidence from Entrepreneurs and Investors
Felipe A. Csaszar, Harsh Ketkar, Hyunjin Kim
TL;DR
Strategic decision-making is inherently challenging, and uncertainty remains about how AI-augmented SDM may look. The paper examines AI’s role in strategy generation and evaluation, finding that current LLMs can achieve human-comparable performance in realistic strategy tasks while potentially altering search, representation, and aggregation.
Problem
Strategic decision-making is inherently challenging, and significant uncertainty remains regarding how AI-augmented SDM may look.
Method
The paper studies AI and strategy generation using an empirical model that includes an AI indicator, fixed effects, and an idiosyncratic error term.
Results
Current LLMs can achieve human-comparable performance in realistic strategy tasks involving generation and evaluation.
Takeaways & Limitations
AI integration has the potential to fundamentally alter SDM processes—search, representation, and aggregation—and boost the quality, efficiency, and heterogeneity of strategic analysis.
Takeaways & Limitations
Some capabilities remain beyond current AI systems, which excel at recognizing patterns in existing data.
Abstract
from arXiv · showhide
This paper explores how artificial intelligence (AI) may impact the strategic decision-making (SDM) process in firms. We illustrate how AI could augment existing SDM tools and provide empirical evidence from a leading accelerator program and a startup competition that current Large Language Models (LLMs) can generate and evaluate strategies at a level comparable to entrepreneurs and investors. We then examine implications for key cognitive processes underlying SDM -- search, representation, and aggregation. Our analysis suggests AI has the potential to enhance the speed, quality, and scale of strategic analysis, while also enabling new approaches like virtual strategy simulations. However, the ultimate impact on firm performance will depend on competitive dynamics as AI capabilities progress. We propose a framework connecting AI use in SDM to firm outcomes and discuss how AI may reshape sources of competitive advantage. We conclude by considering how AI could both support and challenge core tenets of the theory-based view of strategy. Overall, our work maps out an emerging research frontier at the intersection of AI and strategy.
1 Introduction
The paper asks how AI may participate in strategic decision-making, which has traditionally required nuanced judgment over open-ended, uncertain textual inputs. It combines conceptual analysis, empirical comparisons, and a framework for understanding AI's implications for strategic processes and competitive advantage.
- 1.1 Could AI make strategic decisions?: Strategic decision-making is difficult because it involves high-stakes, complex, uncertain choices and open-ended qualitative inputs and outputs.
- 1.1 Could AI make strategic decisions?: LLMs make AI-augmented strategy more plausible because they process strategic text, exhibit reasoning capabilities, and contain information relevant to strategic decisions.
- 1.1 Could AI make strategic decisions?: The paper examines how AI-augmented SDM may look, how effectively LLMs make strategic decisions, and what using AI implies for SDM.
- 1.2 Our approach and contribution: The authors illustrate AI-augmented versions of existing SDM tools and provide empirical evidence from realistic entrepreneurship settings comparing LLMs with entrepreneurs and investors.
- 1.2 Our approach and contribution: The paper argues that AI may alter search, representation, and aggregation in SDM, while changing how firms understand strategy, competition, and competitive advantage.
2 How may AI-augmented SDM look?
The paper reimagines familiar strategic decision-making tools as AI-augmented processes for generating, evaluating, challenging, and aggregating strategic ideas. Examples suggest that LLMs can produce diverse analyses quickly and support virtual crowds for strategy work.
- 2 How may AI-augmented SDM look?: AI could augment Scenario Planning, Porter’s Five Forces, Devil’s Advocate, and Wisdom of the Crowd for strategy generation and evaluation.
- Scenario Planning: GPT-4 generated multiple reasonable business-school scenarios and a six-point plan, including AI-company partnerships and faculty upskilling.
- Porter’s Five Forces: An LLM produced a Porter’s Five Forces analysis comparable in depth and breadth to an MBA student’s analysis, including force-intensity assessments.
- Devil’s Advocate: An AI devil’s advocate can generate contrarian arguments more frequently and without human social inhibitions when challenging strategic assumptions.
- Wisdom of the Crowd: Virtual individuals and AI-based aggregation could provide a quick, inexpensive way to simulate crowds and explore more strategies than relying solely on human input.
evaluation
The paper evaluates LLM strategy generation and evaluation using business plans from a European accelerator and a startup competition. LLM-generated plans performed comparably to or better than entrepreneur plans in evaluations, while LLM scores were positively correlated with experienced investors’ scores.
- evaluation: LLM-generated strategies attracted at least as much investor interest as entrepreneur-generated plans, while LLM evaluations correlated positively with experienced investors’ evaluations.
- AI and strategy generation: In the accelerator experiment, LLM-generated plans received 0.14 standard deviations higher ratings on average than entrepreneur-generated plans (p < 0.001).
- AI and strategy generation: Evaluators were 5 percentage points more likely to recommend LLM-generated plans for acceptance and 3 percentage points more likely to invest in them.
- AI and strategy generation: LLM versions outperformed rejected entrepreneur plans by 7 percentage points in acceptance recommendations, 8 points in introduction interest, and 6 points in investment likelihood.
- Implications: The findings suggest that AI may help entrepreneurs generate broader, higher-quality strategies and help investors screen ideas and provide feedback at scale.
- AI and strategy evaluation: LLM and investor evaluations had an average correlation of 0.52, with standardized scores showing similar distributions.
4 Implications of AI-augmented SDM
AI may reshape strategic decision-making by relaxing bounded rationality and expanding the speed, scale, and flexibility of search, representation, and aggregation. Current evidence suggests LLMs can generate and evaluate strategies comparably to human participants, while stronger AI could enable virtual experimentation and more complex strategic representations.
- Current LLMs can generate and evaluate strategies at a level comparable to humans participating in leading startup accelerators.
- The paper treats these implications as conjectures because the analysis relies on preliminary AI-augmented decision-making tools and results.
- AI can process more information more rapidly and at lower cost than human strategists, potentially producing wide-ranging effects across strategic decision-making.
- Search: AI could substantially accelerate search, enabling firms to analyze more alternatives, launch more products, and increase the pace of competitive evolution.
- Search: Weak AI may increase search speed without substantially improving fitness or heterogeneity, whereas strong AI could improve both and uncover opportunities overlooked by human managers.
- Representation and aggregation: AI may enable virtual crowds, in-silico experimentation, and offline strategic search, while increasing representational complexity and making reconceptualization easier.
5 Conclusion and future directions
The paper concludes that current LLMs can perform human-comparable strategy generation and evaluation, while AI may reshape search, representation, aggregation, and the sources of competitive advantage. Its effects on firm performance depend on AI progress, adoption, complementary assets, and competitive dynamics.
- Current LLMs achieve human-comparable performance in realistic strategy-generation and strategy-evaluation tasks, the two main outputs of strategic decision-making.The authors present these findings as answers to their central research questions.
- AI may relax bounded rationality and alter search, representation, and aggregation, potentially improving the quality, efficiency, heterogeneity, and availability of strategic decision-making.The paper frames these changes as affecting both SDM processes and the generation and evaluation of strategies.
- 5.1 Effects on performance: Under weak AI, adoption may reduce costs and increase productivity but also intensify competition, encourage convergence on common applications, and eventually commoditize AI use.After prolonged stability, AI may become a standard tool with no significant differentiation.
- 5.1 Effects on performance: Under progressing AI, differences in firms’ ability to adopt and integrate capabilities may create temporary advantages, while complementary assets help determine rents and profitability.The paper identifies both AI-specific and traditional complementary assets as relevant to competitive outcomes.
- 5.3 Conclusion: The paper maps an emerging research frontier while recognizing that AI progress is difficult to predict and that extensive further exploration remains necessary.Even if AI plateaus, the authors argue that current systems already perform useful analyses and that SDM may still need to adapt.
- 5.2 Implications for strategy research: AI may enable strategy frameworks to become executable algorithms, increasing the precision, speed, scale, and breadth of strategic analysis and theory development.The paper also discusses virtual strategy simulations and AI-assisted pattern recognition as routes to new strategic insights.
A.1 Scenario Planning
The paper uses AI to generate multiple business-school futures while considering competitor responses. The scenarios differ in technological orientation, global structure, and entrepreneurship focus, each with distinct benefits and execution challenges.
- Scenario Planning: AI-generated scenarios compare a tech-forward blended-learning school, a distributed global network, and an entrepreneurship incubator while incorporating competitors’ potential responses.The scenarios emphasize digital infrastructure, global campuses, or student startups, respectively.
- Tech-Forward Titan: The tech-forward scenario could scale to a global student audience, but depends on seamless online-offline execution and continual technology updates.Competitors may respond by upgrading digital infrastructure and partnering with technology companies.
- The Global Networker: The global-network scenario emphasizes experiential learning and cultural exchange across multiple campuses, while competitors may expand international partnerships and exchanges.Its central challenge is maintaining consistent educational quality across locations.
- Incubator of Innovation: The innovation-incubator scenario finances student startups and could create a self-sustaining alumni ecosystem, but requires substantial mentorship, resources, and business-community ties.Competing schools may strengthen entrepreneurship programs and relationships with venture investors.
- Cross-scenario implications: Across scenarios, reshaping business education depends on effective execution, continual adaptation, and staying ahead of competitor responses.The scenarios present different benefits and challenges rather than a single superior model.
A.2 Porter’s Five Forces
The Porter’s Five Forces analysis describes business-school education as increasingly competitive and technologically disrupted. It highlights pressure from entrants, students, faculty, substitutes, and established institutions.
- Threat of New Entrants: Online education platforms and innovative teaching methods are lowering entry barriers despite traditional requirements for infrastructure, faculty, accreditation, and reputation.New institutions can offer courses without the heavy infrastructure of conventional schools.
- Bargaining Power of Buyers: Students have substantial bargaining power because more business schools and online courses increase choice and heighten attention to employability, curriculum relevance, and return on investment.Schools compete for students through scholarships, innovative programs, and other incentives.
- Bargaining Power of Suppliers: Top faculty possess significant bargaining power because specialized expertise is scarce and prestigious professors can demand competitive salaries, research grants, and benefits.Outside offers from other prestigious institutions strengthen this bargaining position.
- Threat of Substitute Products or Services: Online courses, vocational training, certifications, corporate training, and self-study platforms substitute for traditional business-school education through greater flexibility, affordability, or accessibility.Platforms such as Coursera, edX, and Udemy expand these alternatives.
- Rivalry among Existing Competitors: Competition among business schools is fierce, with institutions contesting students, faculty, corporate partners, rankings, research, employability, and alumni earnings.Rivalry also drives innovation in curriculum, pedagogy, and extracurricular offerings.
A.3 Devil’s Advocate
The devil’s-advocate analysis identifies risks in a technology- and entrepreneurship-focused business school. The main concerns involve narrow curricular focus, volatile technologies, startup failure, resource demands, student fit, and institutional sustainability.
- Curricular focus: Overemphasizing technology and entrepreneurship could narrow the curriculum and neglect foundational disciplines such as finance, strategy, operations, and organizational behavior.The analysis treats these traditional disciplines as still important to business education.
- Technological volatility: Rapid technological change could make a future-focused curriculum reactive to current trends and difficult to maintain over the long term.The concern is that what is groundbreaking today may become obsolete within a few years.
- Financial and resource risks: Financing student startups may produce weak returns because most startups fail, creating concerns about the school’s financial sustainability.Large campuses and digital platforms also require expensive ongoing maintenance and updates.
- Student fit: The proposed model may not meet students seeking traditional MBA experiences, networking, or career transitions, potentially limiting student-body diversity.A blended model may also exclude students who prefer fully traditional or fully online approaches.
- Implementation and governance: Entrepreneurship-heavy programs may struggle to provide adequate quality control, mentorship, and attention across many simultaneous startups.The analysis also identifies accreditation, reputational, ethical, theoretical-learning, and long-term funding challenges.
Appendix B: Generation study
The generation study constructs LLM-completed business plans from accelerator submissions and compares them with entrepreneur-authored plans through randomized evaluator assessments. Evaluators score plans across five dimensions and an aggregate index.
- Sample and generation: The study selects five accepted and five rejected accelerator business plans, then creates LLM-generated versions of each selected plan.The underlying accelerator set contains 17 accepted and 142 rejected plans after screening.
- Plan construction: LLM-generated plans complete the remaining business-plan sections, including Solution, Market, and Go-to-market strategy, from an original first section.Formatting and length-adjustment prompts standardize the generated versions.
- Evaluation design: Evaluators are randomly assigned original or LLM-generated versions, with display order randomized across ten plans in a Qualtrics survey.The sample includes 250 online evaluators screened for United States location and angel or venture-capital experience.
- Reported comparisons: Table B.2 compares entrepreneur- and LLM-generated plans by factor and in an aggregated index, while Table B.1 reports evaluator descriptive statistics.The supplied passages identify the comparison structure but do not report the table’s numerical outcomes.
- Evaluation rubric: Plans are evaluated on writing quality, innovation and value proposition, execution plan, investment potential, viability, and a standardized sum index.Each factor is scored from 1 to 10 and standardized for the comparison table.
Appendix C: Evaluation study
The evaluation study uses LLM prompts to assess startup business plans across 10 investor-relevant dimensions and compares those evaluations with human judgments. Human–LLM evaluation correlations remain robust across plans submitted before and after the LLM’s training window.
- The LLM is prompted to act as an experienced venture capitalist whose textbook integrates practical experience with academic research in strategy and entrepreneurship.
- The scoring framework operationalizes investor evaluation through criteria such as team competence, execution feasibility, financial prospects, market strategy, differentiation, customer fit, and long-term viability.
- The prompts require structured analyses of each plan’s pros and cons, followed by identification of its two strongest and two weakest dimensions.
- Human–LLM evaluation correlations remain robust across business plans submitted before and after the LLM’s training window.