Decision Intelligence and Operations Optimization in the Automotive Industry: Concepts, Applications, Challenges, and Practical Experience

Guest speaker: Lin Lin, Data Analysis Expert; Editor: Liu Qian, Zhejiang University of Technology; Platform: DataFunTalk.

Overview: With the rise of new energy, autonomous driving, and AI, the automotive industry’s intelligence level is rapidly increasing. Decision intelligence, a flagship AI technology, drives digital transformation in the sector. This talk shares ten years of data‑analysis experience, focusing on four topics: automotive industry chain overview, operations‑optimization empowerment, project implementation challenges, and practical exploration with experience summary.

1. Automotive Industry Chain Overview

The automotive chain consists of four core parts: (1) end‑to‑end R&D and technology covering procurement, manufacturing, sales, and after‑sales; (2) component procurement from numerous suppliers; (3) vehicle manufacturers producing core components such as engines and transmissions; (4) sales and service, including traditional dealer distribution, direct sales, after‑sales, insurance, and used‑car services. AI can be applied throughout, e.g., knowledge graphs for parts, predictive maintenance, defect detection, invoice recognition, and visual inspection.

2. Operations‑Optimization Empowering the Automotive Industry

Operations optimization seeks the best decision under constraints, involving two key steps: modeling (defining decision variables, objectives, constraints) and solving (using exact or heuristic algorithms). Traditional applications include routing, facility location, and supply‑chain design.

In automotive supply chains, optimization supports demand capture (e.g., new‑energy credit policies), R&D scheduling, inventory and order management, vehicle allocation, logistics planning, and dispatching. Applications can be classified into three decision‑making levels:

Strategic level: long‑term capacity planning, process planning, inventory and warehouse planning; low time sensitivity, high optimality requirement.

Planning level: weekly/monthly production, distribution, logistics, and material plans; balanced optimality, timeliness, and stability.

Execution level: shop‑floor scheduling, picking routes, material supply; high timeliness and stability, lower optimality demand.

Examples:

Machining process optimization: Mathematical modeling reduced manual layout effort by 80% and provided a cost‑optimal solution for engine manufacturing.

Production planning optimization: Integer‑programming models balance color, configuration, daily and monthly outputs across multiple factories.

Inventory optimization: Safety‑stock and reorder‑point models combine demand forecasts, service‑level targets, and stock‑out costs; machine‑learning enhancements further improve accuracy.

3. Project Implementation Difficulties

Complex business logic: Numerous domain‑specific terms make communication between business owners and modelers challenging.

Modeling difficulty: Qualitative goals (e.g., “shortest processing time”) are hard to quantify; many problems are nonlinear.

Solver infeasibility: Data quality or model formulation issues can lead to no solution, requiring extensive debugging.

User acceptance: Optimized solutions may conflict with established habits, reducing adoption.

Iterative cost escalation: Incomplete requirements cause repeated iterations and higher project costs.

4. Practical Exploration and Experience Summary

The project lifecycle is divided into four stages:

POC (Proof of Concept): Verify concept feasibility; for optimization, validate that target cost reductions are achievable, but recognize that later constraint additions may invalidate early results.

Define: Capture detailed business goals, profit maximization, efficiency, capacity, and sequence constraints; thorough understanding of each constraint is essential.

Build: Develop the mathematical model and data pipeline; anticipate data gaps and constraint conflicts that may require model revisions.

Test: Conduct functional, performance, and algorithm‑specific tests; ensure constraints remain satisfied after any late‑stage changes.

5. Q&A Highlights

• Production‑plan insertion is usually handled manually if it occurs after the weekly plan is locked; recurring constraints can be modeled, but ad‑hoc insertions often require manual adjustment.

• Optimization adoption varies by OEM maturity; older equipment and delayed data collection hinder real‑time scheduling.

• When model results clash with user habits, both parties must compromise; simplifying complex business scenarios is necessary for a tractable model.

• Inventory management differs for mature, new, and low‑volume parts; safety‑stock calculations rely on demand forecasts and variance, while low‑volume items often follow a “order‑as‑used” approach.

Thank you for attending.