Decision Intelligence and Operations Optimization in the Automotive Industry: Practices, Challenges, and Lessons Learned

Introduction With the rise of new energy, autonomous driving, and AI, the automotive sector’s intelligence level is rapidly increasing. Decision intelligence, a flagship AI technology, drives digital transformation in the industry, yet faces practical pain points that require systematic solutions.

Automotive Industry Chain Overview The chain consists of four core parts: (1) R&D and technology spanning procurement, manufacturing, sales and after‑sales; (2) component procurement from numerous suppliers; (3) vehicle manufacturers producing core parts such as engines and transmissions; (4) sales and service, including traditional dealer networks, direct sales, after‑sales, insurance and used‑car services. AI techniques—knowledge graphs, predictive maintenance, defect detection, computer vision, NLP, etc.—can be applied throughout these stages.

Operations Optimization Empowering the Automotive Industry Operations optimization seeks the best decision under constraints and involves two key steps: modeling and solving. Modeling translates real problems into mathematical forms (variables, objectives, constraints); solving applies exact or heuristic algorithms. Typical applications include supply‑chain, routing, site selection, and network design.

In the automotive supply chain, optimization is used for: • Strategic decisions such as capacity planning, long‑term inventory and warehousing; • Planning decisions like production, distribution, logistics, and material plans; • Execution decisions such as shop‑floor scheduling, picking routes, and material supply. Strategic level emphasizes optimality, planning level balances optimality with timeliness, and execution level demands high speed and stability.

Three Illustrative Optimization Cases

Machining Process Planning Traditional manual planning for engine machining requires dozens of senior engineers and yields only feasible solutions. A mathematical model provides an optimal plan, reducing manual effort by about 80%.

Production Planning Optimization Balancing color, configuration, daily and monthly production targets across multiple factories is modeled as an integer‑programming problem to achieve balanced and timely schedules.

Inventory Optimization By integrating demand forecasts, service‑level targets, and shortage costs, a safety‑stock model (enhanced with machine‑learning predictions) minimizes inventory while meeting service requirements.

Implementation Difficulties Major challenges include complex business logic, difficulty quantifying qualitative goals, non‑linear constraints, infeasible models, solver performance, low user acceptance of optimized solutions, and costly iterative refinements.

Practical Exploration and Experience Summary

POC (Proof of Concept) Phase Validates feasibility and accuracy of algorithms; however, changing requirements can quickly invalidate POC results.

Define Phase Clarifies processes, reports, roles, performance needs, and explicitly lists objectives and constraints for the optimization model.

Build Phase Develops the model and data pipeline; challenges include data quality, constraint conflicts, and the need for a complete feasible solution set.

Test Phase Executes functional, performance, and algorithm‑specific tests; additional time is required to accommodate late‑emerging constraints.

Q&A Highlights

Insertion planning in production is feasible when incorporated as a constraint; ad‑hoc insertions after schedule finalization often require manual adjustment.

When optimization results conflict with user experience, both modelers and users must compromise; successful projects rely on close collaboration and mutual understanding.

Inventory management varies by part type; mature parts use statistical safety‑stock models, while low‑volume parts often follow a simple “order‑as‑used” approach.

Thank you for attending the session.