AI-ER separates how much AI threatens a software company's business model from its capacity to adapt, offering a compact metric set and an evidence/confidence layer to support due diligence; the framework is practical and well-grounded but remains to be validated empirically.

Main Finding

The paper introduces AI-ER, a two-dimensional assessment framework that separates (a) AI exposure — the external economic pressure AI creates on a software or software-based business model — from (b) AI resilience — the firm’s capacity to absorb that pressure and adapt. It operationalizes both dimensions with a compact set of metrics, a rule-based scoring/aggregation proposal that keeps evidence and confidence explicit, and a procedure for empirical validation. The framework is designed for strategic assessment, tech due diligence, and investment/M&A decisions; it is intentionally not a single aggregate “company grade.”

Key Points

• Conceptual separation:
  • AI exposure = pressure for business-model change arising from AI capabilities and diffusion.
  • AI resilience = firm-level ability to absorb, adapt to, and economically use AI.
  • Dimensions are assessed and reported separately (no single combined score).
• Derived economic impact mechanisms (how AI changes business economics):
  • Substitution, Compression, Bundling, Re-intermediation.
• Exposure metrics (numerical + categorical modulator):
  • Substitutability of the core paid customer benefit (how closely AI capabilities map to the core service).
  • Replicability of the offering (effort to recreate offering given company data, integrations, domain expertise, regulation).
  • Competitive dynamics (how competitors or platforms can change speed, price, bundling using AI).
  • Customer access / demand pressure (changes in customer expectations, willingness to pay, and intermediated selection).
  • Business-model modulator (categorical): direction/amplification vs dampening of impacts on current business model.
• Resilience metrics (five numerical):
  • Protective positions (robust, company-specific resources or barriers that remain valuable under AI change).
  • Adaptability (organizational capacity to sense and change product/process/business model).
  • Technical AI maturity (ability to develop, integrate, monitor, and operate model-based functions reliably).
  • Implementation capability (software delivery, skills, governance to convert technical possibilities into deployed solutions).
  • Economic viability (balance of development/operation/control costs vs expected value capture).
• Evidence & confidence:
  • Each metric must be linked to documented evidence.
  • Confidence scores capture quality and independence of evidence (not the rating value).
  • Assessments can be done with public information initially and refined with internal data.
• Scoring & aggregation:
  • Metrics are combined only within their respective dimension to produce an exposure and a resilience profile.
  • Non-compensatory rule-based variant is proposed to keep individual critical scores visible.
  • Encourages multiple independent assessment runs where possible to improve reliability.
• Technical foundation and limits:
  • Focuses on economically relevant AI services (analytical, predictive, optimization, generative).
  • Emphasizes deployment conditions: need for integration, monitoring, human oversight, regulatory compliance, and cost/viability constraints.
• Validation: paper proposes empirical validation via independent assessments, inter-rater/reliability analysis, and sensitivity testing; provides a numerical example (specification) for testing.

Data & Methods

• Methodological approach:
  • Integrates three literatures: AI technical capabilities & deployment limits; technological exposure & business-model change; organizational resilience & dynamic capabilities.
  • Procedure: assess AI capabilities & limits → review research foundations → derive recurring economic impact mechanisms → define assessment dimensions → define metrics → construct AI-ER profile.
• Evidence sources (used to motivate metrics and limits):
  • Task-exposure studies (e.g., Eloundou et al., OECD) for occupations/task risk.
  • Benchmarks and technical literature for generative/predictive/analytical capabilities and their failure modes.
  • Business model and platform economics research for bundling/re-intermediation effects.
  • Organizational resilience and digital transformation literature for adaptability and capability formation.
• Implementation details:
  • Metrics were selected to be distinct, observable, and limited in number to preserve tractability and avoid double-counting.
  • Proposed rating logic: rule-based aggregation inside dimensions; evidence-anchored ratings; explicit confidence scoring; independence of multiple runs required for treating them as separate evidence.
  • The framework is conceptual and prescriptive; it specifies how to collect, rate, and combine evidence but does not assume particular proprietary AI technologies.
• Validation plan:
  • Empirical validation recommended through (1) independent assessor comparisons, (2) inter-rater reliability metrics, and (3) sensitivity analyses on scoring rules and evidence quality.

Implications for AI Economics

• For tech due diligence, M&A, and investors:
  • AI-ER provides a structured, traceable way to evaluate AI-induced risk and adaptation capacity beyond code/architecture checks — capturing economic capture, platform risk, bundling, and customer-access threats.
  • Keeps technical substitutability distinct from economic vulnerability: a company may be technically replaceable yet economically resilient (and vice versa).
  • Explicit confidence/evidence treatment helps surface uncertainty and prevents overconfident single-number grades.
• For firm strategy and management:
  • Emphasizes the need for investments not only in models but in integration, monitoring, governance, and business-model redesign (implementation capability + economic viability).
  • Highlights the importance of protective positions (data assets, regulatory/contractual barriers, customer lock-in) and organizational adaptability as sources of resilience.
• For policy and market structure analysis:
  • Draws attention to platform-mediated re-intermediation and bundling as systemic mechanisms that can reshape market access and rents.
  • Suggests regulators/investors should look beyond model performance to deployment, intermediaries, and concentration risks.
• For future research:
  • The framework fills a conceptual gap by combining exposure and resilience but requires empirical validation and calibration.
  • Opens avenues to quantify how specific firm-level attributes (e.g., data exclusivity, platform dependence, delivery capability) mediate the distribution of AI gains across incumbents, entrants, and platforms.
• Limitations to bear in mind:
  • Currently conceptual and prescriptive; effectiveness depends on implementation fidelity, quality of evidence, and assessor judgment.
  • Designed for software and software-based business models; applicability to heavily physical/regulatory businesses will be limited or need adaptation.