What's the Difference Between "Visibility" and "Accuracy" in the Supply Chain? —— Four Levels of Visibility and Three Criteria for Deciding Which Level to Implement First
An automotive parts company launched a supply chain visibility platform in 2025: over 300 suppliers were online, GPS tracking was available for all shipments, and arrival nodes were automatically transmitted. The project cost over three million yuan. A year later, the downtime due to material shortages only decreased by 8%, and the expedited shipping costs remained almost unchanged. Purchasing agents were still making phone calls daily to chase materials. The supply chain director's confusion is typical: We can see everything, so why are we still firefighting every day?
The problem isn't with the platform, but with the goals. This project bought "visibility," but was evaluated based on the benefits of "accuracy."
Conclusion: Visibility addresses "what is happening now and when will there be a problem," while accuracy addresses "based on current trends, what will happen in the future and how much should be prepared in advance." The former buys reaction time, while the latter buys preparation time. There is a layer of decision rules and a layer of data foundation between the two. Expecting the latter from the former is the most common reason for the failure of such projects.
1. Visibility Has Four Levels, and Most Companies Stop at the First Two
The phrase "launched a visibility platform" itself carries no information. The real question is: At which level did it stop?
| Level | Answers What Question | Typical Implementation | Improves What | Typical Reasons for Failure |
|---|---|---|---|---|
| 1. Post-Event Visibility | Why was the last delivery delayed? | ERP reports, arrival records, traceability queries | Only improves post-mortem quality | Used as real-time management, always one step behind |
| 2. Real-Time Visibility | Where is the shipment now, and what stage is the order at? | Supplier portal, in-transit GPS, scanning nodes, EDI | Communication costs, frequency of material chasing | Data is available, but no one is assigned to monitor it |
| 3. Early Warning | Which node is about to have a problem? | Threshold rules, event-driven, exception kanban | Missed warning rate, earlier detection | Thresholds set for "already happened" rather than "about to happen" |
| 4. Decision-Making | What should be done, who should do it, and how long to complete it? | Suggested actions, assignment of responsible persons, disposal time limits, and escalation | Decision and execution time | Only provides information without actions, warnings become new to-do items |
Criterion: The time from "information received" to "action completed" must be shorter than the lead time required to address the issue. If a warning is issued when there is no operational space left—such as discovering missing customs documents when the goods arrive at the port—then this visibility is just for record-keeping, not for management.
This criterion can filter out most "large screen projects." Large screens solve the information presentation at Level 2, while Levels 3 and 4 require rules and responsibilities, regardless of the screen size.
2. Three Breakpoints Where "Visibility" Fails to Be Useful
Breakpoint 1: Data without Rules. When the platform goes live, it lists hundreds of indicators but fails to answer three questions: Who will monitor, what value triggers action, and how long must the loop be closed after action? A simple test: Pick any warning and ask, "Who will take action within 30 minutes of this warning being issued?" If there's no answer, the function is just for show.
Breakpoint 2: Mismatch Between Warning Time Window and Decision Time Window. Events that can be known three days in advance often require decisions to be made thirty days in advance. An appliance company implemented supplier capacity visibility, but the capacity data was monthly. By the time they saw that the "key process capacity was fully occupied," their orders were already scheduled six weeks out. The information was correct, but it was too late.
Breakpoint 3: The Wrong Person at the Wrong Level. The dashboard is installed on the director's phone, but the planner who needs to call suppliers to adjust shipments can't see it in the workshop; or, conversely, exceptions are pushed directly to the general manager, leaving three layers of management without responsibility. The visualization should target the "person who can take action," not the "highest-ranking person."
Passing Line: For A-class materials (the 20% of categories that account for 70% of procurement costs), the time from when an abnormal event occurs to when the responsible person confirms it should not exceed 4 hours; the closure rate of warning actions (with action records and conclusions) should be no less than 80%. These two metrics can immediately distinguish between "really being used" and "just a show at launch."
3. The Prerequisites for "Accuracy" Are Not Models, but Three Data Foundations
Many companies interpret calculability as buying a set of predictive algorithms. In reality, the algorithms are the last step, and three foundational elements are missing:
First, lead times should be based on distributions, not averages. Materials with an average lead time of 30 days may have a P90 lead time of 48 days. Setting safety stock based on the average means accepting a material shortage once every ten times.
Second, supplier commitments must be quantifiable. Besides on-time delivery rates, you should also consider "the number of commitment changes" and "the distribution of the deviation between committed and actual delivery dates." A supplier with an OTD of 95% but an average of three rescheduling changes is more difficult to plan for than a supplier with an OTD of 88% and stable commitments.
Third, the frequency of demand and supply linkage should match the material attributes. Monthly S&OP cannot manage weekly material shortages; A-class materials must be managed on a weekly rolling basis. C-class materials can be managed with monthly inventory buffers; forcing a weekly rolling basis only increases unnecessary workload.
Criterion: If the P90 lead time of key materials deviates from the P50 lead time by more than 40%, it indicates significant supply volatility. Safety stock must be set based on P90, not the average. This calculation often reveals that many companies have had incorrect safety stock parameters since the platform's launch.
Expected Range: Implementing "real-time visibility + early warning rules + responsible person closure" without changing the algorithm typically reduces downtime due to material shortages by 20% to 35% and expedited shipping costs by 15% to 25%; adding predictive replenishment can further improve inventory turnover by 0.3 to 0.8 times. It's important to note that any solution promising a "30% reduction in inventory and a 50% reduction in stockouts" is likely misrepresenting the metrics—either pushing inventory to suppliers or redefining stockouts as "ordered but not yet delivered."
4. Which Level to Implement First: Three Criteria to Determine Priority
Not all events are worth predicting, and not all events require real-time monitoring. Prioritize based on three metrics:
| Criterion | How to Measure | Usage |
|---|---|---|
| Event Value | Single event loss (downtime cost per hour × impact area) | Events with losses below the threshold should be added to the "inventory buffer" list and not included in the system |
| Early Warning Window | How many days remain from "when it can be known" to "when a decision must be made" | Events with a window shorter than one week should be managed with real-time visibility and automatic escalation; predicting them is meaningless |
| Action Certainty | Whether there are standard actions and whether someone can execute them | For events without standard actions, define the rules first before implementing the system |
Combined, this forms a priority table: High-loss, long-window, and clear-action events should prioritize calculability (predictive replenishment, capacity pre-occupation); high-loss, short-window events should prioritize real-time visibility and automatic escalation; low-loss events should be managed with inventory and service levels, and not included in the project scope.
Counterexample: A construction machinery company's first version of the visualization was a large screen covering "all suppliers, all materials, all indicators," with over 600 indicators. After three months, the average daily visits dropped to 4. The second version, which focused on 12 events—each corresponding to a responsible person, a threshold, and an action—was actually used. The value of visualization lies not in how much it covers, but in the few events where people actually follow through.
5. Five Steps for Implementation
- List Event Inventory. Identify the events that truly caused downtime, delayed deliveries, and expedited shipping costs over the past 12 months. Select 10 to 15 events, listing them by event, not by system module. More than 20 events indicate a lack of prioritization.
- Label Each Event with Three Metrics. Single event loss, early warning time window, and whether there are standard actions. Events without all three metrics should be excluded from the project until the metrics are complete.
- Sort by "Loss × Early Warning Window." The top 5 events should be included in the first phase, and the rest should be on the observation list.
- Install Rules Before Screens. Clearly define who will monitor, what the threshold is, and who will be escalated to if the threshold is exceeded. Do not launch warnings without these details.
- Discuss Predictions After Three Months of Stability. The value of predictions is based on the complete recording of historical events. Without event data, the model learns noise, not patterns.
The underlying judgment behind this sequence is: Visibility improves reaction speed, while calculability improves preparation depth. Reaction speed has an upper limit—materials are only in transit for a few days; preparation depth has no upper limit, but it requires clean data. Therefore, start with visibility for a few key events, ensuring they are monitored and acted upon, before moving to predictions—doing it the other way around turns the platform into an expensive monitor.
In a Nutshell
"Visibility" buys a few days of reaction time, while "accuracy" buys a few weeks of preparation time. The former relies on rules and responsible persons, while the latter depends on lead time distribution, commitment reliability, and linkage frequency. The sequence is to first ensure that a few key events are monitored and acted upon, and then discuss predictions—reversing this order turns the platform into an expensive display.
This article belongs to the "Digitalization and AI Zone · Operational Digitalization" direction, targeting operations, supply chain, and digitalization leaders, and does not involve quality management system clauses.
Knowledge code: 16.2.3 Author: QTank