Summary of Key Points
This article demonstrates the capabilities of Baidu DuMate AI assistant by simulating the daily work of a procurement specialist, Ah Hui, in the manufacturing industry—organizing material lists, comparing prices, and evaluating alternative materials. The test reveals that while AI can efficiently handle routine tasks such as Excel data organization and formatting, it has several significant limitations:
1. Rigorous inaccuracies in real-time price checks: The AI provides completely incorrect data on prices and inventory levels.
2. Poor information isolation: It fails to distinguish between relevant and irrelevant documents, leading to the misuse of unnecessary information.
3. Unstable results: The same task may yield different outcomes when using different algorithms or settings.
The overall conclusion is that AI can serve as a valuable assistant but cannot make independent decisions. Any financial decisions involving spending must be carefully reviewed by a human.
Detailed Analysis
1. AI as an Excel Expert: Efficient and Accurate at Routine Tasks
Ah Hui’s most challenging task was organizing the BOM (Bill of Materials). Issues included inconsistent notation for the same component (e.g., “10KR” vs. “10kΩ”), the need to mark missing delivery dates, and ensuring proper formatting. During the test, DuMate completed this task in just 5 minutes. Notable improvements included:
- It correctly identified two different capacitors as the same part (“C1 (10uF/±20%/0603”) and (“C2 (10μF 20% 1605)” and noted that “1605” was an abnormal variant.
- It automatically converted the unit from “10KR” to “10kΩ” and marked the delivery date as “To be confirmed” without any arbitrary changes.
- For potentially duplicate components (“J1” and “J2”), it appropriately labeled them as “Suspected duplicates” rather than deleting them, which aligns with procurement logic.
Ah Hui stated, “This is much faster than doing it manually and saves a lot of time.” Although there were minor errors (e.g., changing the tolerance from “1%” to “±1%”), the AI performed well for handling repetitive and tedious spreadsheet tasks.
2. The Dangers of False Information: Misleading Price Checks
The most problematic aspect was the real-time price comparison test. When DuMate was asked to check the price of a capacitor on Lichuang Mall, it generated a report with brand, model, and price details that seemed credible. However, upon verification, Ah Hui found significant discrepancies:
- The actual price of the component was 0.92 yuan, but the AI reported 0.038 yuan (a 20-fold difference).
- It claimed there were 3850 units in stock, but in reality, there were none.
- The model “Sanye Yuan C0603X5R106M160NT” did not even exist on Lichuang Mall.
This type of misleading information is more dangerous than simple errors because it could lead to serious consequences if used in negotiations with suppliers or in placing orders at incorrect prices, resulting in financial losses.
3. AI’s Misuse of Relevant Information
During the test, the team included a test instructions document with an exchange rate of “1:7” in the folder. DuMate used this rate instead of the correct one (“1:6.8”), causing a 3% deviation in the total cost estimate (which could amount to tens of thousands of yuan for a large purchase order). Ah Hui expressed concern that if the AI also processed outdated quotes or colleagues’ notes, the results would be even more inaccurate.
This highlights the issue with AI’s inability to filter out irrelevant information and its tendency to rely on incorrect data.
4. Unpredictable AI Outcomes
Even for the same task, different settings led to different results from the AI:
- In the “work area” mode, DuMate identified a certain capacitor as having an abnormal packaging.
- In the “no work area” mode, it considered the same component to be a metric version of another part (“1605” vs. “0603”).
This indicates that AI’s performance is not consistent and cannot be reliably relied on for critical decisions.
5. The Right Role for AI
Ah Hui concluded that AI is useful for:
- Routine tasks: Organizing Excel files, formatting data, and summarizing information, which can save time.
- But not for: Making pricing decisions, conducting real-time price checks, or selecting alternative materials that require financial approval. In these cases, human review is essential (e.g., verifying exchange rates, checking official prices, and correcting dates in reports generated by AI).
In summary, while AI can significantly enhance efficiency, it should be used as a tool, not a substitute for human judgment and responsibility.
Final Takeaway
AI is like a talented intern who has just graduated: it excels at simple tasks like Excel processing but often fails when dealing with real-time market data or professional judgments. Technology advances rapidly, but in procurement, every detail—every penny spent and every component purchased—requires the careful verification of a professional like Ah Hui. AI can create detailed reports, but only humans can truly understand the value of each decision.