Introducing new products into manufacturing scenarios is one of the most challenging and riskiest tasks in the industrial world. It requires a thorough balancing act between engineering, procurement, planning and quality control, all while rushing against strict deadlines. In the world of luxury eyewear, where the launch of new collections is tightly linked to presentation events and tight commercial deadlines, things can get complicated. This thesis dives into a structural issue I encountered during my internship at Thelios, the eyewear brand under the LVMH Group. The main problem is a consistent mismatch between the proposals for New Product Introductions (NPI) and the actual availability of components when it is time to kick off First-Lot production. This disconnect, often caused by delays in receiving materials from suppliers, is usually only recognized after production reactively witnesses a problem, leaving little time for any effective fixes. In this thesis, there will be a dive into the conceptual design of an AI-based system that seeks to proactively guarantee the availability of components during new product introduction. The key target is to resolve any possible discrepancies that may explicit itself. This system is crafted to automatically check the consistency between the launch proposal, the planning and MRP data handled through SAP, CyberPlan and the real physical and qualitative status of each component in the bill of materials. The proposed architecture is centred around a dual-layer logic. The first layer encodes deterministic business rules derived from established domain knowledge, designed to flag unambiguous risk conditions with certainty. The second introduces a machine learning classification model trained on historical NPI launch data, capable of detecting subtler patterns of risk that rigid rule sets alone would fail to discover. For each component, the system produces a quantified availability risk score and these scores are aggregated into a Launch Readiness Index at the product level and translated into standardised operational recommendations addressed to the responsible functions. The potential for future development and implementation of such a system, leveraging the expertise of managers, professors and researchers, is highlighted as a critical step in transforming this conceptual design into a functional tool. The efficacy of the system is intended to be demonstrated through a retrospective back-testing approach, applying the model to reconstructed historical data states and comparing its ex-ante risk signals against the delays and corrective actions applied. The thesis contributes to three interconnected domains: NPI governance, by introducing a structured and automated launch gate validation mechanism; supply chain coordination, by reconciling heterogeneous data from multiple enterprise systems into a unified readiness assessment and AI-based decision support in operations management, by demonstrating the applicability and practical value of hybrid rule-based and machine learning architectures in a real industrial context.

Introducing new products into manufacturing scenarios is one of the most challenging and riskiest tasks in the industrial world. It requires a thorough balancing act between engineering, procurement, planning and quality control, all while rushing against strict deadlines. In the world of luxury eyewear, where the launch of new collections is tightly linked to presentation events and tight commercial deadlines, things can get complicated. This thesis dives into a structural issue I encountered during my internship at Thelios, the eyewear brand under the LVMH Group. The main problem is a consistent mismatch between the proposals for New Product Introductions (NPI) and the actual availability of components when it is time to kick off First-Lot production. This disconnect, often caused by delays in receiving materials from suppliers, is usually only recognized after production reactively witnesses a problem, leaving little time for any effective fixes. In this thesis, there will be a dive into the conceptual design of an AI-based system that seeks to proactively guarantee the availability of components during new product introduction. The key target is to resolve any possible discrepancies that may explicit itself. This system is crafted to automatically check the consistency between the launch proposal, the planning and MRP data handled through SAP, CyberPlan and the real physical and qualitative status of each component in the bill of materials. The proposed architecture is centred around a dual-layer logic. The first layer encodes deterministic business rules derived from established domain knowledge, designed to flag unambiguous risk conditions with certainty. The second introduces a machine learning classification model trained on historical NPI launch data, capable of detecting subtler patterns of risk that rigid rule sets alone would fail to discover. For each component, the system produces a quantified availability risk score and these scores are aggregated into a Launch Readiness Index at the product level and translated into standardised operational recommendations addressed to the responsible functions. The potential for future development and implementation of such a system, leveraging the expertise of managers, professors and researchers, is highlighted as a critical step in transforming this conceptual design into a functional tool. The efficacy of the system is intended to be demonstrated through a retrospective back-testing approach, applying the model to reconstructed historical data states and comparing its ex-ante risk signals against the delays and corrective actions applied. The thesis contributes to three interconnected domains: NPI governance, by introducing a structured and automated launch gate validation mechanism; supply chain coordination, by reconciling heterogeneous data from multiple enterprise systems into a unified readiness assessment and AI-based decision support in operations management, by demonstrating the applicability and practical value of hybrid rule-based and machine learning architectures in a real industrial context.

An AI-based system to integrate New Product Introduction and Planning. The Thelios case.

HAMMOUDI, ZIAD
2025/2026

Abstract

Introducing new products into manufacturing scenarios is one of the most challenging and riskiest tasks in the industrial world. It requires a thorough balancing act between engineering, procurement, planning and quality control, all while rushing against strict deadlines. In the world of luxury eyewear, where the launch of new collections is tightly linked to presentation events and tight commercial deadlines, things can get complicated. This thesis dives into a structural issue I encountered during my internship at Thelios, the eyewear brand under the LVMH Group. The main problem is a consistent mismatch between the proposals for New Product Introductions (NPI) and the actual availability of components when it is time to kick off First-Lot production. This disconnect, often caused by delays in receiving materials from suppliers, is usually only recognized after production reactively witnesses a problem, leaving little time for any effective fixes. In this thesis, there will be a dive into the conceptual design of an AI-based system that seeks to proactively guarantee the availability of components during new product introduction. The key target is to resolve any possible discrepancies that may explicit itself. This system is crafted to automatically check the consistency between the launch proposal, the planning and MRP data handled through SAP, CyberPlan and the real physical and qualitative status of each component in the bill of materials. The proposed architecture is centred around a dual-layer logic. The first layer encodes deterministic business rules derived from established domain knowledge, designed to flag unambiguous risk conditions with certainty. The second introduces a machine learning classification model trained on historical NPI launch data, capable of detecting subtler patterns of risk that rigid rule sets alone would fail to discover. For each component, the system produces a quantified availability risk score and these scores are aggregated into a Launch Readiness Index at the product level and translated into standardised operational recommendations addressed to the responsible functions. The potential for future development and implementation of such a system, leveraging the expertise of managers, professors and researchers, is highlighted as a critical step in transforming this conceptual design into a functional tool. The efficacy of the system is intended to be demonstrated through a retrospective back-testing approach, applying the model to reconstructed historical data states and comparing its ex-ante risk signals against the delays and corrective actions applied. The thesis contributes to three interconnected domains: NPI governance, by introducing a structured and automated launch gate validation mechanism; supply chain coordination, by reconciling heterogeneous data from multiple enterprise systems into a unified readiness assessment and AI-based decision support in operations management, by demonstrating the applicability and practical value of hybrid rule-based and machine learning architectures in a real industrial context.
2025
An AI-based system to integrate New Product Introduction and Planning. The Thelios case.
Introducing new products into manufacturing scenarios is one of the most challenging and riskiest tasks in the industrial world. It requires a thorough balancing act between engineering, procurement, planning and quality control, all while rushing against strict deadlines. In the world of luxury eyewear, where the launch of new collections is tightly linked to presentation events and tight commercial deadlines, things can get complicated. This thesis dives into a structural issue I encountered during my internship at Thelios, the eyewear brand under the LVMH Group. The main problem is a consistent mismatch between the proposals for New Product Introductions (NPI) and the actual availability of components when it is time to kick off First-Lot production. This disconnect, often caused by delays in receiving materials from suppliers, is usually only recognized after production reactively witnesses a problem, leaving little time for any effective fixes. In this thesis, there will be a dive into the conceptual design of an AI-based system that seeks to proactively guarantee the availability of components during new product introduction. The key target is to resolve any possible discrepancies that may explicit itself. This system is crafted to automatically check the consistency between the launch proposal, the planning and MRP data handled through SAP, CyberPlan and the real physical and qualitative status of each component in the bill of materials. The proposed architecture is centred around a dual-layer logic. The first layer encodes deterministic business rules derived from established domain knowledge, designed to flag unambiguous risk conditions with certainty. The second introduces a machine learning classification model trained on historical NPI launch data, capable of detecting subtler patterns of risk that rigid rule sets alone would fail to discover. For each component, the system produces a quantified availability risk score and these scores are aggregated into a Launch Readiness Index at the product level and translated into standardised operational recommendations addressed to the responsible functions. The potential for future development and implementation of such a system, leveraging the expertise of managers, professors and researchers, is highlighted as a critical step in transforming this conceptual design into a functional tool. The efficacy of the system is intended to be demonstrated through a retrospective back-testing approach, applying the model to reconstructed historical data states and comparing its ex-ante risk signals against the delays and corrective actions applied. The thesis contributes to three interconnected domains: NPI governance, by introducing a structured and automated launch gate validation mechanism; supply chain coordination, by reconciling heterogeneous data from multiple enterprise systems into a unified readiness assessment and AI-based decision support in operations management, by demonstrating the applicability and practical value of hybrid rule-based and machine learning architectures in a real industrial context.
NPI
AVAILABILITY
RISK
AI
CROSS-FUNCTIONAL
File in questo prodotto:
File Dimensione Formato  
FINALE TESI.pdf

Accesso riservato

Dimensione 2.6 MB
Formato Adobe PDF
2.6 MB Adobe PDF

The text of this website © Università degli studi di Padova. Full Text are published under a non-exclusive license. Metadata are under a CC0 License

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/20.500.12608/116731