Carving Efficiency: Using AI to Transform Order Management for a Wood Manufacturer

Leveraging cutting-edge AI, QuantSpark streamlined a wood manufacturer's order process, automating 50% of tasks and significantly enhancing efficiency. This case study explores our rapid integration of Generative AI models to achieve impactful results.

Case Study Context

In today's wood manufacturing industry, manual processes and data entry bottlenecks are more than just operational headaches - they're profit killers. From production floors to quality control stations, these inefficiencies cascade into escalating costs and frustrated customers.

Enter AI: a game-changing force revolutionising the sector. By automating tedious tasks and bringing precision to order management, AI isn't just eliminating errors - it's liberating skilled workers to focus on strategic initiatives. Manufacturing leaders are witnessing transformed production schedules, razor-sharp quality control, and a level of operational efficiency previously thought impossible.

The impact? Beyond impressive cost savings, companies are discovering a new competitive edge in an increasingly digital marketplace. This isn't just automation - it's a complete reimagining of what wood manufacturing can achieve.

The Business Problem

Our client was seeking to automate complex processes across their business through the incorporation of AI. Potential use cases were identified through a series of ideation workshops, where order management was determined to be high-priority.

The order management team faced daily challenges of deciphering customer orders arriving via email, and entering each order into their system. Incoming orders lack template conformity; arriving in multiple languages and layouts, with large variations in product descriptions that may deviate significantly from the official item. This diversity requires a degree of translation and a dedicated team to manually interpret and enter orders into the system.

This labour-intensive process takes up valuable resources and is prone to human error. Many companies are still heavily reliant on manual processes, yet business growth can be stunted due to inefficiencies:

• Time-Consuming: A team of full-time employees is required to carry out the order entry process, taking time away from higher-value tasks.
• Error Prone: Mistakes such as typos, missed data points, and formatting inconsistencies can lead to order fulfilment issues and a decline in customer satisfaction.
• Unscalable: Managing an increasing volume of orders requires a larger team and significant investment to process, which is not a sustainable solution and can result in lower efficiency.

The Solution: Leveraging AI

Recent advancements in Generative AI, particularly Large Language Models (LLMs) ability to understand and interpret human language, bring about opportunities to optimise these processes.

QuantSpark conducted a 4-week feasibility study to investigate how leveraging AI and LLMs could streamline the order entry process. Following a comprehensive assessment of various techniques for extracting order details, we developed a bespoke solution employing a multi-faceted approach to accurately determine customer orders.

Our proof-of-concept solution followed a 3-step process:

1. Order Details: A general-purpose multimodal Generative AI model was employed to extract order details and detect individual line items from a PDF customer order prior to further processing
2. Product Matching: Classification algorithms trained on numerical representations of the order descriptions were used to identify key product fields, allowing the prediction of a unique SKU value for each line item
3. Historic Data: To improve predictions and increase the accuracy of item-dependent information, historic purchase data is incorporated

During the study, OpenAI released the latest state-of-the-art model, GPT-4o. Realising the potential benefits that could be gained, QuantSpark pivoted and successfully integrated it within 24 hours of release.

This integration achieved significant gains in efficiency during a time-critical project, leading to superior results in a shorter time. This rapid adoption not only facilitated improved outcomes, exceeding the project’s success criteria, but it also delivered significantly reduced processing time and a 50% reduction in inference costs. Through this switch, we identified numerous model improvements:

• Multilingual Capability: Improved capacity to correctly identify and process text in multiple languages, occasionally within a single document.
• Expanded Document Structure Comprehension: The newer model exhibited superior proficiency in extracting order details from a wider range of purchase order formats, layouts, and styles.
• Higher Efficiency: Faster processing speeds and cost reductions when compared with older models.

Impact

Our study showcased the transformative power of Large Language Models (LLMs) for business processes. The benefits are multifaceted:

• Effortless Efficiency: Automating manual data processing can free up employee time for strategic tasks. LLMs can extract key information in seconds with a high degree of accuracy, minimising errors and allowing businesses to process higher volumes in less time.
• Customer Satisfaction: Faster processing, fewer errors, and rapid response times lead to satisfied customers.
• 24/7 Operations: Processing can be done outside business hours, with results ready for the next day.
• Processing Power: LLMs can process large volumes of purchase order data to learn preferences and identify inconsistencies.

QuantSpark's success stemmed from an agile approach that combined cutting-edge Generative AI models with traditional machine learning algorithms. This diverse technical expertise ensured we leveraged the right tools for the job. By prioritising swift testing and adoption of the latest advancements, we maximised client efficiency gains while building a solution with continuous improvement and expandability in mind.

Our approach

• Identifying opportunities: We worked closely with the client to understand their needs and brainstorm potential applications of AI across their business.
• Prioritising value: Use cases were evaluated based on feasibility and potential cost savings/revenue generation.
• Feasibility study: We conducted a focused study on the most promising use case - automating order management - demonstrating significant potential for efficiency gains.