Vision of the Future: Streamlining and Sophistication of Procurement Site Operation Using AI Agents

In response to these “reduced estimate precision” and “procurement site operation individualization” issues, utilization of AI Agents is expected to achieve both streamlining and sophistication of operations.
AI Agents will convert the past orders which form the basis of estimates into data, replace the specialized knowledge of purchasing managers with AI processing, and automate the decision-making process for orders. On the other side, purchasing managers will no longer need to carry out operations such as past results comparisons and surveying. This enables them to spare time for other important work that is difficult to replace with AI such as predicting demand, strengthening and diversifying sourcing site relations, and formulating procurement strategies.

Achieving this will require the promotion of a two-step plan: Step 1 “Building Business Infrastructure” and Step 2 “ Streamlining and Sophistication of Procurement Site Operation.”

There are many locations where past order records and estimates are not yet digitized. Internal document data conversion is an essential prerequisite for AI utilization. In the course of building the business infrastructure, we help digitize the estimates saved in various formats and carry out data deduplication between items.

When data from estimates and invoices is manually input, the mistakes, omissions, and inconsistency in formatting can complicate processing, causing issues. Also, in addition to the format differing between suppliers, saving formats vary between PDF, paper, and other formats, which makes data organization labor-intensive.
Utilizing AI-OCR instead enables handwritten and printed text to be automatically converted into digital data. Important information such as the supplier company name, product name, quantity, unit price, total amount, and order data is extracted and recorded accurately. While avoiding input mistakes and data inconsistencies, this method can also organize data in different formats consistently through machine learning pattern recognition. Combining AI-OCR with an large language model  (LLM), also dramatically improves reading accuracy for these financial records. In this way, work is both more accurate and more efficient, saving time and contributing to reduced costs through optimization of human resource utilization.

In addition to the format differing for each supplier’s estimates and invoices, the notation and names used for materials is also frequently not consistent. Even within the same company, different names or codes are used across departments in some cases, making data integration and analysis more difficult.
For this problem, generative AI can be used for deduplication, integrating the scattered notation. AI uses language learning and natural language processing to determine similarities in differing names while organizing data based on consistent notation rules. For example, it can automatically detect different notation such as “Company A: Hexagonal Bolt M10” and “Company B: Hexagonal M10 Bolt” and register the items in a consistent format, increasing the convenience of searching and analysis. This reduces the burden of manual data cleaning, enabling smooth procurement site operations and price analysis based on accurate data.

Purchasing departments select the supplier for an order based on the requested content and specifications from the ordering department, then acquire an estimate. The content of this estimate is evaluated based on factors such as past results, market prices, and the knowledge of the purchasing manager, and this manager makes the final decision on the supplier and submits the order. Within this work flow, we will determine the tasks which can be replaced with AI, organize the required data and output for each task, and define the AI requirements. Here, we have divided the tasks into six categories to show some examples of the processing automation AI Agents can achieve (see Figure 3).

Referencing the materials database constructed in Step 1, the AI Agent determines potential candidate suppliers to request estimates from.

■Current Status
In addition to it requiring specialized knowledge, another problem with investigating transaction history and procurement records is that the necessary address information can be hard to figure out. Especially for younger employees, it can be difficult to quickly grasp the prices and transaction requirements of similar orders, contributing to slower decision-making and increased procurement costs.

■After AI Agent Utilization
Through searching and extraction of digitized past transaction data, necessary information can be accessed instantly. In addition, search engine tuning can automatically correct similar and incorrect notation, enabling consistent searching even of differing formats. For similar items, you can also search related data, effectively increasing awareness of information that tends to be overlooked.

After selected the optimum suppliers for the requested content, the next step is requesting estimates from each company. In the future, this will involve conversation with the suppliers’ AI Agents, automating the process from requesting to receive an estimate.

■Current Status
The complexity of diverse estimate requests and material orders from multiple suppliers can cause work mistakes and communication errors.

■After AI Agent Utilization
By managing the communication between AI Agents, these kinds of business risks can be reduced.

Efficiently analyze the information listed on received estimates and organize the information needed for the continuing tasks.

■Current Status
Normally, the estimate format differs for each supplier, and organizing the information is time-consuming.

■After AI Agent Utilization
Once the AI Agent has organized the necessary information, use AI-OCR. Automating the extracting and analysis of information from estimates eliminated labor hours for manual organization, streamlining operations.

In addition to internal knowledge, AI Agents can also use diverse data such as market trends to autonomously carry out predictive modeling, determine parameters to use, and apply current market conditions to past material price data to determine appropriate prices.

■Current Status
Due to the soaring prices of materials, investigating past estimates is increasingly becoming a poor point of reference, making fair price determination difficult. Actually, company assessments show a wide divergence between these prices and those provided by the supplier side, and managers struggle with the adjustments. We receive a large number of consultations regarding the influence on price negotiations and procurement processes as a whole. In addition, specialized knowledge related to each type of material is essential for price prediction, and because the number of people who can handle this work is limited, efficient procurement judgment is difficult, and the inability to procure materials with the appropriate timing causes problems.

■After AI Agent Utilization
Through the use of market condition data in addition to past estimates, validity of pricing can be evaluated based on both past transactions and market trends (such as raw material price fluctuations, exchange rates, and supply conditions), enabling real-time calculating of appropriate pricing. Also, AI Agents can autonomously select the market condition data and price prediction model parameters to use for each material, streamlining the optimum price prediction process.
In addition, an abnormality detection model can be used to catch potential supplier delivery delays and supply risks in advance. By analyzing past delivery records and supply conditions and predicting supplier performance, these risks can be prevented.

This step consists of collecting past transaction and order status information for the supplier.

■Current Status
Supplier information collection is often dependent on a specific supervisor, resulting in delays when this manager is absent or transferred, and the lack of smooth information sharing causes problems. In addition to this, orders may be submitted without awareness of the supplier’s order status, which causes many problems such as leading to delays as a result.

■After AI Agent Utilization
By utilizing an AI Agent to collect supplier information, work automation becomes standardized, which not only prevents individualization but also enables real-time analysis of supplier status, reducing work burdens and detecting risks.

Based on information collected and prepared by various AI Agents, this process outputs an easy-to-understand report for managers containing information on supplier candidates.

■Current Status
For the final supplier decision, when the volume or quality of information is insufficient, the supplier selection process does not go well, and this frequently causes overly expensive orders. Especially in cases where the suppliers’ proposed content is unclear or multiple options cannot be sufficiently compared, it becomes difficult to understand optimal contract conditions. In these conditions, not only price but also important aspects such as quality and delivery time also tend to be overlooked, resulting in risks of increased costs and delayed deliveries.

■After AI Agent Utilization
AI Agents that create text gather and analyze not only pricing but also other supplier information such as quality, delivery, and past results from various perspectives. They organize this data in an easy-to-compare format, promote rapid and smooth decision-making and enabling the optimum supplier to be chosen every time. As a result, they prevent overpriced orders, reduce delivery delay risks, and achieve both streamlining and cost reduction of the overall procurement process.

Accordingly, promotion of radical reforms is needed for deep-rooted problems in procurement site operations. Figure 4 shows the approach needed for problem-solving as six processes.

• Define price fluctuation factors, expose work individualization, and clarify bottlenecks for the overall procurement process.
• Especially, define materials easily affected by changing market conditions and cost increase causes through analysis of past procurement data and transaction records.

• Create an environment conducive to integrating data from inside and outside of the company.
• Construct a high-precision database by automatically extracting data from estimates, invoices, and other documents using AI-OCR technology and leveraging generative AI for deduplication and standardization.
• Read in external data such as market conditions, exchange rates, and shipping costs in real-time, and organize it in a format that enables it to be used for procurement material judgments.

• Codify knowledge in a format that AI Agents can use easily by documenting experienced personnel’s best practices related to judgment standards, price determination processes, price negotiations, and supplier evaluations.

• Verify the reduction in workload from the application of AI Agents.
• Conduct a test operation of price prediction using past data and evaluate the usefulness of AI-driven search and comparison of past results.
• Based on the PoC results, proceed with work integration and operation process establishment, gradually introducing AI into actual work to ensure it takes root.

• With the roles performed by AI Agents and human staff clearly defined, organize the work flow.
• Formulate AI utilization guidelines to ensure operations can continue smoothly even if the operator changes, and clarify operation rules.

• Evaluate the AI implementation results quantitatively and qualitatively, promoting optimized operations.
• Through continuous learning and tuning of AI Agents, improve precision and broaden the scope of use, achieving increased sophistication for the company-wide procurement strategy.