unit economics modeling

Traditional revenue forecasting approaches struggle to uncover the underlying customer behavioral drivers—such as customer acquisition, repeat purchase rates, and average transaction value—that influence revenue dynamics. To address this limitation, this work proposes the Customer-Based Multi-Task Transformer (CBMT), which uniquely integrates multi-task learning with a Transformer architecture to jointly model customer behavioral metrics and total revenue through shared representations. Furthermore, CBMT incorporates a downstream alignment mechanism to enhance both interpretability and predictive accuracy. Empirical evaluation on real-world customer transaction panel data demonstrates that CBMT outperforms existing methods across 23 out of 24 evaluation metrics, achieving a 30% reduction in total sales prediction error compared to the strongest baseline and significantly surpassing single-task models employed by 74.3% of firms.

This study addresses the trade-off between model performance and training cost in large language model (LLM) development to maximize profit. By integrating scaling laws with microeconomic theory, it establishes the first rational decision-making framework for LLM training, distinguishing between compute-constrained and data-constrained regimes and deriving optimal strategies for model scale and training budget allocation. The analysis reveals that under compute constraints, the optimal training cost grows nearly linearly with hardware efficiency (FLOPs/$), while total cost scales sub-quadratically; under data constraints, optimal expenditure scales quadratically with available data volume and inversely with hardware efficiency. The findings indicate that current industry practices only partially align with theoretical optima, offering quantitative guidance for more efficient LLM training.

Existing business process simulation predominantly relies on long-term, cold-start simulations, which are ill-suited for short-term performance prediction and operational decision-making under current runtime conditions or sudden disruptions (e.g., demand surges, resource shortages). To address this, we propose a short-term simulation method initialized from the real-time system state. Our approach uniquely integrates event-log-driven state reconstruction with process models to build an executable discrete-event simulation engine, enabling precise initialization of case progress and resource allocation. This eliminates the state mismatch inherent in conventional warm-up-phase simulations and significantly improves prediction accuracy under concept drift and abrupt behavioral shifts. Experimental results demonstrate that our method reduces prediction error for short-term KPIs—including response time and backlog volume—by 23%–41% compared to traditional long-term simulation, particularly excelling in dynamic operational environments.

This paper addresses the challenge of accurately identifying demand under substantial temporal fluctuations and absent cost-variation information, focusing on the French railway industry. We systematically evaluate the economic performance of revenue management (RM) strategies using a novel identification framework that integrates time-series relative price changes, consumer rational expectations, and firms’ weak optimality conditions in pricing. Our methodology combines structural econometric modeling, counterfactual demand estimation, endogenous price treatment, and censoring-handling techniques to overcome identification issues arising from sales cutoffs and the lack of exogenous price variation. Results show that current RM practices significantly outperform uniform pricing but still incur a 16.7% revenue loss relative to theoretically optimal dynamic pricing. This study provides the first empirical quantification of RM’s net economic value in a real-world industrial setting and reveals its critical role in aggregating and processing information under demand uncertainty.

This paper addresses the limitations of conventional classification models in decision optimization—specifically, their neglect of cost sensitivity and causal effects. We propose the first unified evaluation framework integrating cost-sensitive learning and causal inference. Methodologically, we formalize standard classification as a special case of single-action causal classification and, grounded in decision theory and axiomatic performance measurement, construct an extensible family of causal performance metrics. Theoretical contributions include: (i) the first systematic unification of cost-sensitive and causal learning paradigms; (ii) rigorous proof of their intrinsic consistency; and (iii) reconstruction and generalization of established industry metrics (e.g., Qini, ROI). Empirical evaluations demonstrate that our framework significantly improves profit-maximizing decisions in real-world business applications—including customer retention and response modeling—outperforming both standard classification and isolated causal or cost-sensitive approaches.

This study addresses the lack of a systematic framework for identifying critical input variables and conducting sensitivity analysis under uncertainty in complex simulations, particularly in military decision-making contexts. The authors propose a unified sensitivity analysis framework that integrates local and global methods—including variance-based, derivative-based, screening, and uncertainty quantification techniques—and strategically maps these approaches to specific decision objectives such as factor prioritization, fixing, variance reduction, and mapping. Innovatively, the framework introduces a “sensitivity audit” mechanism to enhance traceability of model assumptions and promote responsible model usage. By providing a structured guide for high-dimensional, complex simulation systems, this work significantly improves model interpretability, transparency, and the credibility of decisions derived from such models.

This study addresses the challenge of accurately attributing prediction discrepancies between CCAR and CECL forecasting processes to specific input changes, without relying on the order of input substitutions. Framing attribution as a cooperative game-theoretic problem, it presents the first systematic evaluation of multiple attribution methods—including Exact Shapley values, Hierarchical Shapley values, Integrated Gradients, Gradient SHAP, Permutation SHAP, and Kernel SHAP—in real-world production settings. By analyzing these methods across dimensions such as allocation properties, computational cost, implementation requirements, and inherent limitations, the work proposes a practical selection framework that balances interpretability, reproducibility, and engineering feasibility. This framework offers financial institutions actionable guidance for choosing attribution techniques that support transparency and governance in regulatory forecasting systems.

This study addresses the problem of allocating collective surplus in hierarchical organizations when agents have heterogeneous basic needs. To ensure that total revenue is distributed fairly only after all parties’ essential requirements are met, the paper proposes two novel classes of allocation rules: one based on either aggregating net surplus upward through the hierarchy or equally sharing it between each agent and their immediate superior, and another comprising a family of geometric and sequential rules adjusted for individual needs. Employing axiomatic analysis and mechanism design theory, the work fully characterizes allocation mechanisms that simultaneously satisfy fairness, efficiency, and hierarchical consistency, thereby achieving a theoretically coherent integration of individual needs with organizational structure.

Traditional quantitative investment systems typically optimize a single metric—such as the information ratio—and thus struggle to meet professional investors’ multifaceted objectives, including pure alpha generation, style control, drawdown resilience, and turnover and capacity constraints. This work proposes an Objective-Oriented Quantitative Investment (OOQI) framework that formally encodes investment intent as strategy specifications and compiles them into composable, constraint-satisfying strategy assemblies. Key innovations include establishing a dual lattice structure between specifications and assemblies, designing a satisfaction-driven synthesis mechanism, and introducing rolling recertification via e-process-based validation. Empirical results demonstrate that the specification-driven approach satisfies 100% of target constraints across 32 strategies, at the cost of only a 5.5% reduction in information ratio, whereas conventional outcome-oriented methods—despite higher in-sample information ratios—fulfill merely 25% of the specified requirements.

In non-contractual settings, customer churn is unobservable, rendering accurate counts of active customers challenging. This study identifies a category error in the conventional P(alive) metric, which conflates finite-horizon, verifiable repurchase probabilities with infinite-horizon extrapolations of customer survival. To address this, we propose counting customers based on auditable, finite-horizon repurchase probabilities and develop an interval estimation framework using the beta-geometric family of models, replacing prevailing point estimation approaches. Empirical analysis leveraging a seven-year panel dataset of 31,683 customers reveals that alternative model specifications can yield customer counts differing by up to 7.6-fold, while default parameter choices introduce biases as high as 42%. The proposed method substantially enhances both predictive accuracy and verifiability.