Anatomy of Approval: Bottlenecks and Automation

15/09/2026

The Trap of False 'Consensus'

If you ask a middle manager why a purchase request or leave application has been pending for three days, the answer is usually simple: "The boss is busy."

That is what everyone says. But the reality is that the problem isn't a lack of time; it's that internal approval processes have turned a simple decision into a struggle over power and accountability. When a document passes through five or six levels of sign-off, the average wait time doesn't increase linearly—it grows exponentially. Each time someone signs and sets it aside, the file reverts to a "pending" state, and the cycle repeats until someone follows up or simply... forgets.

I have seen many businesses in Vietnam, from large food processing plants to retail chains expanding into Thailand and the Philippines, stuck in this trap. They believe that more signatories mean lower risk. But the data tells a different story. The bottlenecks aren't in the final approval; they are in the intermediate stages where information is duplicated, re-verified, and unnecessarily delayed. To solve this, we don't need a more complex system; we need a different perspective on operational efficiency.

Anatomy of Internal Procurement: Who Is Holding the File?

Consider the most concrete example: the internal Purchase Requisition process. This is the most common workflow, occurring hundreds of times a week in manufacturing and distribution companies. Imagine a warehouse employee notices a raw material shortage, creates a requisition form, and sends it to the warehouse manager.

Stage one: The warehouse manager verifies the quantity. It takes about fifteen minutes. They approve it. But instead of moving immediately to the next step, the system only sends an email notification. The procurement manager, responsible for finding suppliers, usually doesn't read emails immediately. They wait until the afternoon to process a batch of emails at once. Thus, twenty-four hours pass because of a single email notification.

Stage two: The procurement manager approves the price and supplier. They transfer the file to the Finance Department. Finance checks the budget. This is the most congested stage. Why? Because Finance typically checks budgets only at the end or beginning of the week when they have time to reconcile the books. If the requisition arrives on Monday, it might sit there until Friday. Five days. For an item worth a few million VND.

Stage three: The CEO signs off. This is the stage everyone thinks is the slowest, but in reality, it is the fastest if the file is complete. The problem is that files are often missing information. The CEO has to ask: "Why this supplier instead of that one?" This question goes back to the procurement manager, then back up. This loop can repeat two or three times.

Where can AI fit in here? Not in strategic decision-making. AI excels at standardizing information and eliminating redundant verification steps. A data analytics model can automatically cross-check real-time budgets, confirm previously approved suppliers, and standardize file formatting before it reaches the CEO. When the file reaches the decision-maker, it is clean, complete, and free of formal errors. The wait time in the Finance stage can be reduced from five days to a few hours, provided the system is configured correctly.

Where AI Should Stop: Keeping Humans in the Loop

Many people tend to want to automate the entire internal process. That is a common mistake. Full automation means the computer automatically purchases goods when inventory runs out. This sounds ideal, but in operational reality in markets like Mexico or Southeast Asia, where supply chains are volatile, it is extremely risky.

Everyone says AI will replace humans. Reality shows that AI will augment human capabilities. There are places where humans must retain the decision-making role. This is when a new supplier appears with a lower price but no transaction history. The computer sees the low price and wants to buy. But humans know that supplier had quality issues in the last batch or a history of late deliveries. The decision here requires experience and judgment based on implicit information the system hasn't collected yet.

Similarly, in exception scenarios, such as an urgent purchase request for a special project, standard processes will reject or delay it. Humans need the ability to "escalate" or "handle exceptions" flexibly. AI should act as an alert: "This is an exception, here is the reason, here are the potential risks." But the final decision, especially those involving long-term partnerships or company reputation, must remain with humans.

At AIVISION, when we deploy data analytics and automation solutions for clients like Masan or Meat Deli, we always emphasize this: Design the system so AI handles boring, repetitive tasks, and humans handle creative, decision-making work. If you see your employees spending time copy-pasting data from one spreadsheet to another, that is when you need AI. But if they are spending time calling suppliers to negotiate, don't interfere.

What Remains Unsolved and What Is Needed

However, we must be honest about what remains unresolved. Optimizing processes with AI is not a magic wand. It only highlights root problems that have existed for a long time. If your input data is dirty, if item codes are inconsistent, if approval processes are built on personal trust rather than clear criteria, then AI is just a faster machine to make wrong decisions.

We need more data standardization at the system level. Not a massive project, but small rules: Each expense type must have a unique code. Each supplier must have a unique profile. This sounds basic, but in practice, many businesses are still living with fragmented data.

Additionally, we need a cultural shift. Employees should not view the reduction of approval steps as a threat to their positions. They need to understand that internal automation is to give them time to do higher-value work. If managers don't trust the system, they will still retain final approval authority and delay decisions in their own way, no matter how fast the system is.

True operational efficiency comes from the combination of computer speed and human sensitivity. We are in a transitional phase where the boundary between automation and human management is not yet clearly defined. But one thing is certain: Bottlenecks created by human administrative inertia will gradually disappear, making way for new challenges: How to make faster decisions in a volatile environment? How to ensure AI doesn't make cold, emotionless decisions? These are questions that we, as operators, need to answer now, not later.

If you are weighing up a similar project, our team can help you scope it before you spend anything. See what we build or book a conversation.