Generative AI can sharply raise efficiency and accuracy in corporate finance and tax operations, cutting routine costs and freeing staff for higher‑value work; but realising these gains at scale hinges on addressing data privacy, model reliability, system integration and regulatory accountability.

Main Finding

Generative AI (GAI) materially improves efficiency, accuracy, and automation in corporate finance and taxation—demonstrated by Xiaomi’s Financial Shared Service Center and Deloitte—while introducing technical, governance, labor, and regulatory challenges that constrain diffusion and raise distributional and measurement issues.

Key Points

• Case evidence
  • Xiaomi: introduced GAI into its shared services since 2020; reported a ~40% reduction in financial report preparation time; RPA+AI reconciliation covers >1,000 suppliers with an 82% data matching rate and claimed 100% accuracy in the workflow cited.
  • Deloitte: developed DelphAI (five-layer modular platform) and deployed AI agents and edge solutions (OpenVINO, Core Ultra) plus Intel Granulate optimizations; Smart Chain Audit Platform reportedly reaches ~95% automatic voucher verification; invoice processing time reduced from days to minutes.
• Primary applications
  • Automated voucher generation from invoices/orders, first-draft financial reports, account reconciliation, invoice authenticity checks, tax return automation, targeted tax planning (e.g., R&D deductions), compliance screening, anomaly detection.
• Benefits
  • Time savings, higher throughput, improved consistency, faster anomaly detection, potential narrowing of tax gaps, cost reductions in cloud/computing (with optimizations), and reorientation of workforce toward higher-value finance roles (management accounting, business partnering).
• Challenges & risks
  • Model errors/hallucinations with plausible-but-wrong outputs in high-stakes contexts.
  • Data quality, integration and embedding into legacy systems.
  • Large capital and compute costs (barrier for SMEs).
  • Privacy and leakage risks in human–machine interactions.
  • Labor displacement for routine roles and need for reskilling.
  • Incomplete legal/regulatory frameworks and unclear liability and auditability standards.

Data & Methods

• Methods
  • Case-analysis of two organizational settings (Xiaomi Financial Shared Service Center; Deloitte China) and literature review of academic and industry sources.
  • Theoretical framing using Technology Acceptance Model (TAM) and Innovation Diffusion Theory to interpret adoption drivers (perceived usefulness, ease of use, relative advantage, compatibility, complexity, trialability).
• Data sources cited
  • Xiaomi 2023 ESG report and company materials (for metrics like 40% prep-time reduction, reconciliation coverage).
  • Deloitte public descriptions and platform material (DelphAI architecture, Smart Chain Audit Platform metrics).
  • Academic and industry literature on GAI in finance/tax (selected references spanning 2023–2025).
• Empirical scope & limitations
  • Largely descriptive/case-based evidence; limited cross-sectional or causal empirical analysis.
  • Reported performance metrics are organizational claims; paper notes a broader gap in systematic, generalizable empirical measurement across industries and firm sizes.

Implications for AI Economics

• Productivity and cost structure
  • Short-term productivity gains from automating routine accounting/tax tasks; potential reallocation of labor toward higher-skilled finance roles.
  • Capital intensity increases (compute/model costs, platform investments) may shift firm cost structure toward fixed costs and create scale economies favoring larger firms and incumbents.
• Labor market effects
  • Task-level substitution: routine bookkeeping likely declines, increasing demand for analytical, regulatory, and systems-integration skills.
  • Transitional policies needed: retraining, dual-track talent development, and social safety considerations.
• Market structure & competition
  • Adoption advantages for large firms and professional-service incumbents (e.g., Big Four) could increase concentration in some services unless SMEs gain affordable access to GAI tools.
  • Specialized AI platforms (DelphAI-like) create new analytic complementarities and switching/lock-in risks.
• Taxation & public finance
  • GAI-enabled auditing and screening can narrow tax gaps, change compliance costs, and alter the administrative capacity of tax authorities.
  • Automated tax planning may shift firm behavior; measurement needed for how incentives and loopholes interact with model-driven recommendations.
• Measurement and research priorities for AI economics
  • Quantify causal productivity gains attributable to GAI (difference-in-differences, experiments, firm-level panel studies).
  • Measure distributional impacts across firm sizes, sectors, and worker skill levels.
  • Estimate fiscal effects: changes in tax base, enforcement yields, administrative cost savings.
  • Model general-equilibrium effects: capital–labor substitution, scale economies, and dynamic investment in AI.
  • Evaluate externalities: privacy breaches, model errors leading to regulatory penalties, and systemic risks from common-model dependence.
• Policy and governance implications
  • Need for model governance, audit trails, explainability, and liability standards specific to finance/tax contexts.
  • Public investment to reduce SME barriers (shared infrastructure, subsidized APIs, standardized data connectors).
  • Regulatory design to integrate policy APIs and ensure compliant, auditable automated tax filings.
  • Support for pilot & phased rollouts, benchmarking, and validation frameworks to compare AI vs. manual outcomes.

Overall, the paper documents promising business-case outcomes from GAI in finance and taxation while highlighting open empirical questions and public-policy tasks for economists and regulators to measure, manage, and distribute its benefits safely.