Decision matrix for choosing between automatic responses, assisted AI, and human agent

How to decide between automatic responses, AI-assisted, and human agents

Choosing how to handle WhatsApp messages is not a binary decision between 'bot or human.' It is a decision by layers: each message can go through an automatic response, an AI suggestion that an agent approves, or directly to a human agent. What most guides lack is a concrete way to decide when to use each layer, with your own numbers and not opinions.

This page is a reference table: the goal is for you to have it on hand when you need to size up your business's customer service. We do not recommend a brand or a plan; we give you the criteria and the formula so you can apply your own numbers.

The table: which channel to use based on complexity, volume, and budget

The table below crosses three variables: message complexity (low, medium, high), daily volume (how many messages come in per day), and available budget. Each row tells you which combination of layers makes sense. It is not an exact recipe: it is a starting point for you to adjust later with your own data.

How to read the table: the three variables

The table assumes you can classify a message into three levels of complexity. This classification is the most important part and the one almost no one does. Without a clear definition of what is 'complex' for your business, the matrix is useless.

• Low complexity: the answer is unique, short, and does not depend on context. Examples: hours, address, fixed prices, order status with tracking number.
• Medium complexity: the answer requires understanding the message, but there is a standard procedure. Examples: shift changes, simple complaints, inquiries about products with variants.
• High complexity: the answer involves a risky decision, professional judgment, or a case not covered in the manual. Examples: complaints with compensation, legal matters, technical issues requiring diagnosis.

Daily volume is measured in incoming messages, not conversations. A customer who writes five times counts as five messages. The difference matters because automatic responses and AI are charged per conversation, not per message, but the agent's work is measured in messages.

The method for sizing how many agents you need

When volume exceeds 50 daily messages, the question stops being 'what tool do I use' and becomes 'how many people do I need'. The formula is simple and works for any business, with any tool:

1. 1Measure the daily volume of incoming messages (V). If you don't have it, estimate it: count the messages for a week and divide by 7.
2. 2Define the average response time per message (T), in minutes. A simple message may take 1 minute; a complex one, 10 or more. If you haven't measured it, use 3 minutes as a starting point.
3. 3Calculate the work minutes per day: M = V × T. For example, 100 messages × 3 minutes = 300 minutes, or 5 hours.
4. 4Divide by the number of productive hours per agent per day (H). An 8-hour workday has about 6 productive hours, because no one types for 8 straight hours. The number of agents is M / (H × 60).

With the example above: 300 minutes / (6 hours × 60) = 0.83 agents. That is, one person is enough with margin. If volume rises to 200 daily messages with the same average time, you need 1.67 agents: two people, or one person and a lot of automation.

What assisted AI does and what it doesn't do

Assisted AI, like the one used by Wando, works with a human-in-the-loop model: the AI suggests a response and an agent approves it with a click before it is sent. This changes the sizing calculation: the agent does not write from scratch, but is still responsible for what is sent.

What assisted AI does do: reduces writing time, standardizes responses, learns from the responses the agent approves, and becomes more accurate with use. What it does not do: decide on its own, assume legal responsibility, or replace human judgment in complex cases. That is why in the matrix, high-complexity messages always end up with a human agent.

Common mistakes when using this matrix

• Confusing 'messages' with 'conversations' when measuring volume. A customer who writes five times counts as five messages; if you measure conversations, the sizing falls short.
• Using average response time without separating by complexity. An average of 3 minutes can hide messages of 1 minute and others of 15; the average only works if the mix is stable.
• Assuming that assisted AI eliminates the need for agents. AI speeds things up, but someone has to approve; if no one is there, the message goes unanswered.
• Thinking that automatic responses cover everything. Automatic responses work with keywords and exact phrases; if the customer writes differently, it does not match and the message stays in the inbox.
• Not checking Meta's per-conversation prices. The cost of each conversation varies by country and changes over time; if you do not have it updated, the budget does not add up.

How to apply the matrix to your business, step by step

1. 1Classify your messages: take the last 100 incoming messages and classify them as low, medium, or high complexity. Count how many there are of each type.
2. 2Measure daily volume: count incoming messages for one week and divide by 7. If you already have a tool with statistics, use that number.
3. 3Estimate the average time per message: if you have not measured it, use 1 minute for low, 3 for medium, and 10 for high. Calculate the weighted average according to the mix you found.
4. 4Apply the sizing formula: M = V × T, and then divide by productive hours. That gives you the number of agents needed.
5. 5Cross-reference the result with the main table: based on your dominant complexity and your budget, define which layers you will use (automatic, assisted AI, human).
6. 6Check Meta's per-conversation prices in your country before building the budget. That number changes and cannot be invented.

When this matrix does not apply

The matrix assumes that the business receives messages from customers expecting a response. It does not apply if the volume is so low that one agent can handle everything without help (fewer than 20 messages per day), nor if the business uses WhatsApp only to send campaigns without expecting responses. In those cases, the decision is simpler: no automatic layer is needed.

It also does not apply if the business has no record of its messages. If you do not know how many messages come in per day or what type they are, the matrix is theory. The first step is always to measure, even if it is by hand for a week.

The final decision is always human

The matrix organizes the options, but it does not decide for you. The correct combination depends on your industry, your team, and your risk tolerance. A healthcare business will put a human in cases that an e-commerce solves with an automatic response. The table gives you the starting point; the fine-tuning is yours.

If you want to try the assisted AI layer, Wando offers a 14-day trial without a card. But the matrix works the same with any tool: the important thing is that you measure your numbers and apply the formula before buying anything.