AI and NLP in Antitrust: Simulating Merger Price Increases

Antitrust economists and merger regulators increasingly rely on natural language processing (NLP) and artificial intelligence (AI) to evaluate the competitive impact of corporate consolidations. By automating the analysis of massive datasets, these technologies help define relevant markets, calculate consumer substitution patterns through diversion ratios, and accurately simulate potential post-merger price increases. This article explores the direct mechanics of how NLP and AI modernize antitrust enforcement by predicting unilateral pricing power before a merger is approved.

Market-Definition Natural Language Processing

Defining the relevant market is the foundational step in any merger review. Traditionally, economists relied on surveys, executive testimonies, and limited retail scanner data to determine which products compete directly with one another. Today, market-definition NLP processes millions of unstructured text documents—including internal corporate emails, strategic pitch decks, product descriptions, and consumer reviews—to map the competitive landscape objectively.

Regulators use NLP algorithms, such as word embeddings and transformer models, to convert text into numerical vectors. Products and companies that frequently appear in similar semantic contexts are grouped together in a high-dimensional space.

  • Product Substitution Mapping: NLP scans e-commerce descriptions and customer reviews to identify functional similarities between products. If consumers consistently compare Product A and Product B in review texts, the NLP model flags them as close substitutes.
  • Internal Document Analysis: During the discovery phase of a merger investigation, companies submit millions of internal documents. NLP tools rapidly extract sentiment and co-occurrence patterns, revealing who the merging executives actually view as their primary competitors, regardless of their legal arguments.
  • Dynamic Market Boundaries: Unlike static industry classification codes (like NAICS), NLP captures real-time shifts in competition, allowing economists to draw precise boundaries around the relevant product and geographic markets.

Diversion-Ratio Artificial Intelligence

Once the market is defined, economists must measure how closely the merging firms compete. The standard metric for this is the diversion ratio: the fraction of consumers who would switch from Product A to Product B if the price of Product A increased. High diversion ratios between merging firms indicate a strong incentive to raise prices post-merger, as the merged entity will recapture a large portion of the lost sales through its new subsidiary.

Artificial intelligence, particularly machine learning (ML) models, calculates diversion ratios with unprecedented granularity. Instead of relying on broad aggregate assumptions, AI analyzes granular consumer-level data.

  • Clickstream and Search Data: Machine learning algorithms analyze user behavior on digital platforms. By tracking what a consumer views, adds to a cart, and ultimately purchases within a single session, AI predicts the exact second-choice product for millions of individual buyers.
  • Geospatial Movement: For brick-and-mortar mergers (such as supermarkets or hospitals), AI processes anonymized mobile location data. It learns how far consumers are willing to travel and predicts which specific competing location they will visit if their primary choice raises prices or closes.
  • High-Dimensional Demand Estimation: Traditional econometric models struggle when markets have hundreds of differentiated products. Machine learning techniques, such as LASSO regression and random forests, select the most relevant variables from massive retail datasets to estimate cross-price elasticities of demand without overfitting the data.

Simulating Post-Merger Price Increases

The ultimate goal of combining market-definition NLP and diversion-ratio AI is to feed accurate inputs into a merger simulation model. Regulators use these simulations to quantify the "unilateral effects" of a merger—the ability of the newly combined firm to profitably raise prices on its own, without needing to coordinate with remaining rivals.

The simulation process follows a clear, data-driven pipeline:

  1. Quantifying Upward Pricing Pressure (UPP): Economists input the AI-generated diversion ratios and the firms' pre-merger profit margins into UPP formulas. If the diversion ratio between the merging firms' products is high, and the profit margin on the diverted sales is substantial, the model flags a strong upward pricing pressure.
  2. Calibrating Structural Demand Models: For a full simulation, economists use the NLP-defined market boundaries and AI-derived elasticities to calibrate structural models of supply and demand (such as Bertrand-Nash pricing models). AI helps estimate the exact shape of the demand curve, determining how sensitive consumers are to price hikes.
  3. Predicting the New Equilibrium: The simulation changes the ownership structure in the model's code, uniting the two competing firms under a single profit-maximizing algorithm. The model then calculates the new equilibrium prices. It accounts for how the merged firm will raise prices and how non-merging rivals in the NLP-defined market will respond (usually by raising their own prices slightly in response).
  4. Evaluating Efficiencies: Finally, the simulation calculates the exact level of marginal cost reductions (efficiencies) the merging firms would need to achieve to offset the predicted price increases. If the AI models show a 7% predicted price increase, the merging parties must prove the merger will generate enough operational savings to bring that number back down to zero.

By integrating NLP and AI into merger simulations, antitrust regulators minimize guesswork. They move from theoretical assumptions about how markets behave to empirical, high-resolution predictions of exact dollar-and-cent impacts on consumers.