What THA will never automate.

Why draw a line

Speed is useful. But some moments carry feeling, and a machine cannot carry feeling. If you hand those moments over, people notice.

This is my line. Yours may differ. The point is to draw one on purpose.

Six things I keep human

These stay with a person:

• An apology when something went wrong.
• A hard no to a customer or a partner.
• A conversation about money.
• A sensitive or private question.
• The final yes before something goes out.
• A thank you that is meant.

Draw your own line

Pick a type of message, read the question, and decide who should send it.

Two ways to use tools well

A worked example: one upset customer (illustrative)

A buyer writes that the parcel arrived damaged. A tool can spot the word damaged and send an apology. It cannot feel that this buyer waited three weeks for a gift.

So the tool does one job: it flags the chat as upset and puts it at the top of a person's list. The person reads the whole chat, writes the reply, and decides on the fix.

This is the pattern behind every line on our never list. Software sorts and reminds. A human decides and speaks when feelings or money are involved.

Questions to ask before you automate anything

1. Would I be embarrassed if the customer knew a machine wrote this?
2. Does this message involve a complaint, a refund, a death, a worry or a big decision?
3. Can a wrong answer cost a customer or a relationship?
4. Is there a person who can step in within the hour?

If you answered yes to any of the opening three, keep a human in the loop. If you answered no to the last, fix that before you add any tool.

A sample policy page for your own team

1. Purpose: why we keep some replies human.
2. The list: moments a person always handles.
3. The tools: what they may draft or sort.
4. The check: who reviews drafts before they go out.
5. The handover: how a chat moves to a person.
6. The review: when we update this page.

Write it in plain words on one page. Share it with every new team member. Review it after any customer complaint. Policies work strongest when people helped write them, so ask your team what they would add.

Five situations where a person should reply

A tool can notify you that these messages arrived, but a person should write the reply.

How to write a good apology

1. Say you are sorry, plainly.
2. Name what went wrong, in simple words.
3. Say what you are doing to fix it.
4. Tell them when they will hear from you next.
5. Thank them for telling you.

Avoid words like “unfortunately” and “as per policy” in the opening line. Lead with the person.

A friendly honesty rule for tools

If a tool helps write or send a message, review it before it goes out. If a customer asks whether they are speaking to a person, answer honestly. A brand that is clear about how it works earns more trust than one that hides it.

A tool can speak in a warm voice. It should still know when to hand over.

Reviewing your list each month

Every month, read ten chats where a person stepped in. Ask whether the handover came at the right time. Ask whether the person had what they needed. Ask what the tool could have prepared better. Update the list. A living list is more useful than a perfect one written once.

A 7 day action plan

A seven day way to draw your own line.

1. Day 1: list your message types.
2. Day 2: mark emotional ones.
3. Day 3: mark risky ones.
4. Day 4: write the person just list.
5. Day 5: write saved replies for the rest.
6. Day 6: share the list with your team.
7. Day 7: review what happened.

A checklist before you start

• Is the list written down?
• Does the team know it?
• Is there a path to a person?
• Do we review monthly?
• Do customers see warmth?

Mistakes brands make here

1. Automating apologies.
2. Hiding the human option.
3. No clear owner.
4. Never reviewing the list.
5. Treating all messages alike.

Why we keep a list

Technology can do many things. Not all of them should be done by a machine. We keep a short list of moments we will not automate, because they need a person.

This is our own policy, written to keep trust.

What we will not automate

• Apologies for a mistake. A person says sorry.
• Replies to angry or upset customers.
• Difficult news, such as a refund refusal or a delay with real impact.
• Anything involving health, safety or money advice.
• The final decision to send a sensitive message.

Tools can help us prepare, but a person decides and writes the words that matter.

What we are happy to automate

• Reminders to follow up.
• Sorting messages into groups.
• Drafting a opening version that a person edits.
• Routine answers like order status, with a clear way to reach a person.

Honest about being an assistant

When a tool is used in a chat, we believe people should be able to tell and reach a human. A friendly assistant that is clear about what it is can still be warm.

How to build your own list

1. List ten kinds of messages your customers send.
2. Mark which ones involve emotion, money or risk.
3. Mark those as person just.
4. Draft saved replies for the rest and have a person approve them.
5. Review the list every month.

Why this protects your brand

Customers remember how you made them feel when something went wrong. A thoughtful human reply at that moment can matter more than a hundred fast replies about routine things.

What this guide does not promise

It does not promise sales, income or results. It describes how I work and what you can expect.

Written by Madhur Tyagi, founder of The Human Algorithm. 8+ years, 25+ brands, since 2017.

Scope note: Scripts and plans are examples, not a promise of results. Nothing on this page is sent unless you press the WhatsApp button. No outside facts or rates are quoted.