Generative AI shifts M&A IT strategy: LLM affordances let acquirers keep and revive target systems instead of automatically ripping them out, by increasing system transparency, lowering perceived personnel dependence, and cutting conversion costs.

Main Finding

Generative AI (specifically LLM-enabled tools) can materially change post-merger IS-integration strategy by reducing perceived knowledge and conversion barriers to retaining proprietary target systems. In a longitudinal, revelatory multiple-case study of the same e-commerce target acquired twice, the first acquirer pursued a conventional rip‑and‑replace approach; the second, after adopting GenAI affordances, followed a “retain‑and‑revive” strategy that preserved the target system and recovered embedded knowledge via LLM‑assisted practices. LLM affordances (e.g., chat-based learning, code synthesis) increased knowledge transferability, aggregation, and efficiency, reconfiguring feasible action space and enabling new value-capture paths.

Key Points

• Problem: Post‑merger IS integration often destroys or fragments tacit, IT‑embedded knowledge (high cost of replacement, personnel dependence, poor documentation).
• Conventional logic: Acquirers typically absorb/replace target IS (rip‑and‑replace) to align systems and reduce long-term complexity, despite short-term knowledge loss and heavy conversion costs.
• GenAI affordances identified:
  • Learning system knowledge through conversational queries (chat) even when key personnel leave.
  • Automated or assisted code understanding and generation (accelerating migrations, adaptations, or repairs).
  • Aggregation of dispersed knowledge into coherent artifacts (documentation, scripts, mappings).
  • Efficiency gains in developer productivity and onboarding.
• Empirical contrast:
  • BetaCo (pre‑GenAI): executed rip‑and‑replace, resulting in loss of tacit/system-embedded knowledge and high integration cost/risk.
  • GammaCo (post‑GenAI adoption): retained proprietary ERP and revived it using LLM‑supported practices, achieving quicker integration and preserving target-specific capabilities.
• Theory: Using a technology‑affordance lens, GenAI alters the action potentials available to acquirers—constraints that once justified replacement (intransparency, personnel specificity, conversion cost) are partially alleviated, shifting integration logic and strategy choice.
• Limitations/contingencies: Affordances are context‑dependent; they require access to relevant data/code, engineering to prompt/validate LLM outputs, and are subject to model errors, privacy/IP and regulatory constraints.

Data & Methods

• Research design: Interpretivist, embedded multiple-case, revelatory longitudinal study.
• Setting: One digital/e-commerce firm (TargetCo) that experienced two consecutive acquisitions in 2023 by different buyers (BetaCo and GammaCo).
• Primary data: 39 semi-structured interviews (acquirers, target personnel, consultants, service providers).
• Secondary data: Archival sources including ~20 years of online media and documents; triangulation across sources.
• Timeline: Nearly three years of in-situ and retrospective data collection; analysis informed by knowledge-based view and technology affordance theory (Yin, Gioia, Majchrzak & Markus).
• Analytical focus: Cross-case comparison of integration choices and mechanisms through which LLM/GenAI affordances were actualized.

Implications for AI Economics

• Transaction costs and integration economics:
  • GenAI can reduce apparent conversion and search costs associated with proprietary/legacy systems by improving knowledge extraction and reconstruction, lowering the expected cost of retaining targets’ IT assets.
  • This changes the cost–benefit calculus between replacement and retainment strategies, making retention economically feasible where it previously was not.
• Asset specificity and bargaining:
  • Lower integration costs diminish the premium acquirers need to pay (or raise target valuations), altering bargaining dynamics—targets with idiosyncratic IT assets become less “sticky,” potentially increasing acquisitions of niche/legacy firms.
  • Conversely, expected efficiency gains may be capitalized into higher acquisition prices ex ante.
• Human capital and labor markets:
  • LLM affordances substitute for some forms of specialized knowledge work (documentation, code comprehension), reducing the marginal value of certain firm‑specific human capital and shifting bargaining power away from specialized personnel.
  • Potential for labor displacement or reallocation toward oversight, prompt engineering, model validation, and governance roles.
• M&A market structure and strategy:
  • With lowered barriers to integrating heterogeneous IT stacks, acquirers may pursue more aggressive inorganic growth strategies across diverse tech platforms, increasing consolidation in sectors with many specialized legacy systems.
  • Alternatively, easier integration might reduce the time-to-value, affecting post-merger financing and investor expectations.
• Risk, externalities, and public policy:
  • Reliance on LLMs introduces risks: incorrect or hallucinated outputs, IP/data‑leakage, compliance concerns (data residency, GDPR), and potential long‑tail maintenance costs (reliance on proprietary LLM providers).
  • Regulators and auditors may need frameworks to evaluate the reliability and governance of LLM‑assisted integration.
• Research opportunities for AI economics:
  • Quantify integration cost differentials (replace vs retain) with and without GenAI; estimate how much of acquisition premium is explained by anticipated AI-enabled savings.
  • Model equilibrium effects: do GenAI-driven integration efficiencies increase acquisition volume, raise target prices, or redistribute surplus between acquirers, targets, and labor?
  • Study labor market dynamics: skill premiums for prompt engineering/AI oversight vs depreciation of deep systems expertise.
  • Investigate market failures (IP leakage, information asymmetries) introduced by LLM use in integration and design optimal governance/regulation.

Suggestions for practitioners and economists: - Acquirers should explicitly model GenAI-enabled integration options when valuing targets with proprietary systems; run pilot LLM probes for knowledge extraction before committing to rip‑and‑replace. - Account for governance, privacy, and validation costs of LLM use; savings are not free and depend on data access, model quality, and engineering effort. - For economic modeling, treat GenAI as a technology that lowers certain transaction costs and human capital specificity but raises new governance and operational costs—incorporate these offsetting effects when predicting market outcomes.