What 17 years of CRM taught me about what not to automate

I have spent 17 years running CRM and marketing for beauty, fashion and luxury brands. In that time I have automated thousands of customer messages. Welcome flows, abandoned baskets, replenishment reminders, birthday emails.

Most of it worked. Some of it cost us customers we never got back.

Now AI can write and send almost anything, so every business owner asks me the same question: what should we automate? I think that is the wrong first question. Start with what you should never automate. I have three answers.

1. The apology

When an order goes wrong, the customer is not reading your reply for information. They are reading it to find out whether anyone cares.

A templated apology answers that question badly. “We are sorry for any inconvenience caused” tells the customer that nobody read their message.

At one brand, we tested this. Complaints that got a personal reply from a named team member led to a repeat purchase 62% of the time. Complaints that got the automated template came back 31% of the time. The message was the same length and offered the same refund. Only the personal reply read as if someone had written it for that customer.

Use AI here to do the preparation. It can pull the order history, summarise the complaint and draft a reply in seconds. A person should still read it, change it and put their name on it.

2. Your best customers

In most businesses I have worked with, the top 5% of customers bring in around 35% of revenue. Those people know they are valuable. They expect to be treated as if you know it too.

The mistake is to put them in the same flows as everyone else. Your best client of six years gets the same “We miss you! Here is 10% off” email as someone who bought once in a sale. It tells them they are a row in a database.

In luxury, the best relationships I saw were run by people. A store manager who remembered a client’s daughter was getting married did more for retention than any campaign I ever built.

AI is very good at telling you who those customers are and what they have done lately. Let it build the list and the briefing. Then a person should pick up the phone or write the note.

3. The moment someone is about to leave

Every CRM system has a win-back flow. Customer goes quiet for 90 days, system sends a discount. It is the most common automation in retail and one of the weakest. In my experience it brings back around 4% of the people it reaches.

It fails because it treats every quiet customer the same. One left because of a bad delivery. One moved country. One is still loyal and just buys once a year. A discount fixes none of those.

For your highest-value accounts, silence should start a conversation with a real person. When we replaced the automated win-back with a personal check-in for top-tier clients, 22% of them came back.

AI should be the alarm here. It can spot the drop in activity weeks before a person would. What happens next is a human decision.

What I would automate tomorrow

None of this makes me anti-automation. I have built my career on it.

Order confirmations, delivery updates, replenishment reminders, review requests, appointment reminders. Customers want these fast and accurate. Nobody needs a handwritten note to learn their parcel has shipped.

Behind the scenes, automate even more. Segmentation, reporting, data clean-up, first drafts of campaigns. In a typical CRM team, this kind of work eats 40% of the week. That is time your people could spend on the three things above.

The rule I use

Ask one question about any message: is the customer looking for information, or for proof that someone cares?

If it is information, automate it and make it instant. If it is care, let AI prepare everything and let a person deliver it.

Most businesses get this backwards. They spend human hours on reports and data entry, then send their most important messages on autopilot.

If you want to know which way round your business has it, get your business X-rayed.