Vytech streamlines automotive spares inventory management with AI

Large-scale manufacturers often stock thousands of non-production items, such as consumables and spare parts. Mislabelled, inconsistent and duplicated product data can cause major headaches, reducing efficiency, productivity and lead times. Different naming conventions for the same item can lead to confusion and surplus orders; while outdated data leads to inaccurate stock levels, or product specs that waste time and resources. 

In response to automotive industry challenge identified through the Made Smarter Innovation | Digital Supply Chain Hub delivered by Digital Catapult, who aim to improve UK supply chain resilience, Vytech set out to create an artificial intelligence (AI) supported framework that would transform legacy records into a consistent, standardised format for seamless part identification. The aim was to streamline processes, reduce risk, and enhance productivity throughout the automotive supply chain.

Data matching for improved efficiency 

Poor data hygiene impacts productivity and inflates costs. Manually mapping stock items to universal references is a cumbersome process, especially as up to 40% of supplier entries can lack a valid Global Trade Item number (GTIN). Any attempt at creating a new industry-standard product classification would only create additional competition and abstraction, so Vytech opted to use Gs1’s GTIN codes as they are used most often given that GS1 is the only authorised provider of Global Trade Item Numbers (GTINs) for trading with leading global retailers. 

Vytech’s AI-driven solution cuts down the manual effort required to standardise SKU data, leading to quicker item lookups, reduced duplication, and more efficient supply chain operations. It automatically refines descriptions, pulling in verifiable details and flagging potential duplicates in real time. The data-matching system uses generative AI (GenAI) with large language models (LLMs) to tackle inconsistencies head-on.

Combining AI with a vector database 

After exploring different LLMs to select the best performance trade-offs, the Vytech team selected GPT 4o-mini, as it delivers consistent performance in refining product descriptions.  

Vytech’s solution stores product information in a vector database, allowing rapid search and retrieval of contextually similar items, enhancing the model’s effectiveness in matching product data. The model ecosystem combines prompt engineering (with post- and pre-processing filters) and real-time vector lookups to produce more accurate and reliable product descriptions.

Using funding and data-rich testing to refine and scale  

Through the Digital Supply Chain Hub, Vytech was able to access funding necessary to invest in custom model development, refining and fine-tuning instructions, and running countless tests to tailor them specifically for product data matching. The funding alone was not enough to take the solution to the next level, and so Digital Catapult played a key role in addressing some of Vytech’s challenges.

“The network of supply chain professionals and data experts we’ve engaged with has massively shaped our platform’s features. We’ve primarily expanded our approach to include the vector store for improved search and context retrieval and spent more time refining the LLM prompt structure to meet industry data security and accuracy standards.”

Daniel Cliff, Head of Data/Digital Transformation, Vytech Solutions

Firstly, with the help of Digital Catapult’s solution architecture expertise and deep industry knowledge, the Vytech team was able to scale their prototyping, piloting their solution more quickly across several different supply chain contexts.

Secondly, Vytech could deepen vector store capabilities if they could add as much high-quality product data as possible to broaden the solution’s coverage. Deployment of the solution within the Digital Supply Chain Hub Industrial Spares Testbed, developed by Digital Catapult and NBT Group, provided access to crucial industry data and was instrumental in developing and validating the solution. The testbed provided a safe, simulated supply chain environment that closely mirrored real-world conditions, eliminating risk while producing realistic data and system interactions.

Unipres, a key supplier to the automotive industry, hosted the test scenario, allowing Vytech to use product data sourced directly from companies within a supply chain, enabling the model to work with relevant, industry-grade information from day one.

The testbed environment allowed the Vytech team to simulate real supply chain dynamics, validate their matching solution, and receive grounded feedback from industrial stakeholders and academic developers.

Deepening integrations 

Vytech plans to integrate more closely with GS1’s systems to automatically validate GTINs in real time. Since their solution demo, they have developed a proof of concept using the GS1 API to validate the GTIN code output from the model. 

Mapping items to Amazon’s ASIN also streamlines eCommerce listings and helps unify product references across different sales channels. Although Vytech is initially focusing on B2B products, linking to Amazon has value, due to its vast range of products.

About Vytech Solutions 

Vytech Solutions builds smart inventory systems which digitise control, reduce stock issues, and drive supply chain efficiency. They have been delivering tailored tech for real-world impact since 2014, and their new AI-driven data match solution ensures cleaner, standardised product data from the start.