Banks that chase volume risk unprofitable growth; integrating historical customer-profitability with predictive analytics and real-time profitability simulation can reorient acquisition toward higher-value customers, though the proposal is conceptual and only illustrated across two national banks.

Main Finding

The authors develop an integrated conceptual Customer Profitability Analysis (CPA) model that combines historical (retrospective) and predictive profitability perspectives and embeds them into the bank customer acquisition process. Using Soft Systems Methodology (SSM) and a comparative case study of two national commercial banks, they show that current CPA practice is partial, siloed, and backward-looking; integrating predictive CPA, real-time profitability simulation, package-deal evaluation, and cross-functional data flows can materially improve value‑based acquisition decisions and portfolio quality.

Key Points

• Problem: Banks commonly use volume-based KPIs and unit-level profit metrics, which can drive growth that is economically suboptimal; 20–40% of customers may generate negative returns when full cost-to-serve and risk are attributed.
• Gap: Most CPA implementations are historical/descriptive; predictive CPA is limited and not embedded into acquisition workflows (lead scoring, solution design, negotiation).
• Study contribution:
  • Conceptual dual-mode CPA framework that links historical CPA with predictive CPA across acquisition stages.
  • Process redesign using SSM to integrate CPA into three acquisition stages: (1) prospect identification & qualification, (2) needs analysis & solution/package design, (3) proposal presentation & negotiation.
  • Practical tools: profitability-based prospect screening, profitability-optimised package-deal evaluation, and real-time deal-simulation support for Relationship Managers.
• Implementation constraints identified: lack of standardized profitability frameworks, fragmented data and system interoperability, siloed decision-making and KPIs, limited historical data and predictive capability, and regulatory constraints.
• Outcome: An integrated model is proposed (conceptual, not quantitatively validated) to support value-oriented acquisition and more sustainable portfolio growth.

Data & Methods

• Methodological approach: Qualitative, Soft Systems Methodology (SSM).
  • SSM stages applied: problem situation analysis (rich pictures), root definitions via CATWOE, conceptual model development, comparison to real-world conditions and gap analysis.
• Empirical design: Comparative case study of two national commercial banks differing in KPI approach (one using unit profitability, the other adopting customer profitability).
• Data sources: Document analysis and stakeholder insights across functions (management, Relationship Managers, sales, operations, accounting, IT, credit/risk).
• Analytical focus: Three acquisition stages where CPA integration matters most (prospect qualification; needs analysis & package design; negotiation & pricing).
• Limitations noted by authors: qualitative/conceptual output; CPA adoption remains partial; limited historical data and predictive models in the studied banks — future work needed for quantitative validation and operational deployment.

Implications for AI Economics

• Evaluation metric design: Predictive models for customer selection and lead scoring should be evaluated by economic metrics (expected profitability, CLV, profit-at-risk), not solely by classification accuracy or conversion probability.
• Model objectives and optimization: AI/ML systems in banking acquisition should optimize for multi-dimensional objectives (net profitability after cost-to-serve, funding cost, credit risk) and support package-level rather than product-level decisions.
• Real-time simulation and decision support: Deploying AI requires low-latency integration into RM workflows to simulate profitability trade-offs during negotiations; this creates demand for real‑time model inference, fast scenario simulation, and interpretable outputs for front-line staff.
• Data & infrastructure needs: Effective predictive CPA requires integrated data (transactions, product bundles, service cost allocations, credit/risk indicators). Fragmentation and system interoperability are key bottlenecks for AI deployment.
• Organizational and incentive alignment: Adoption depends as much on governance and KPIs as on model quality. AI systems must be embedded into cross‑functional processes and incentive structures to avoid perverse volume-driven outcomes.
• Research directions:
  • Develop profit-optimizing lead-scoring algorithms and compare them to conversion-focused models in field experiments.
  • Methods for attribution of cost-to-serve and dynamic funding costs at customer-bundle level.
  • Causal approaches to estimate counterfactual profitability from acquisition offers and pricing decisions.
  • Explainable ML for profitability predictions to ensure RM trust and regulatory transparency.
• Regulatory and market impacts: Profit-driven acquisition may change competition dynamics (targeting higher‑value customers), raise fairness and access concerns, and interact with prudential regulations—AI economists should study welfare and distributional consequences.
• Deployment risks: Over-reliance on imperfect profitability models risks excluding potentially valuable relationships (especially if data are sparse); model uncertainty, concept drift (customer behaviour changes), and feedback effects on pricing and market structure must be managed.

Summary: The paper argues for embedding predictive CPA into acquisition pipelines and highlights the technical, organizational, and data prerequisites. For AI economics, that implies shifting model development and evaluation toward profit-centric objectives, investing in integrated data and real-time deployment, aligning incentives, and studying broader market and regulatory effects.