By InsyteGroup

Customer acquisition has become an expensive business for many ecommerce brands.

Getting someone to make that first purchase can involve paid search, social media, retail media, promotions and increasingly sophisticated personalisation.

Which makes the second purchase rather important.

A lot of retention investment goes into marketing. Loyalty programmes, email campaigns, personalised offers and reactivation journeys all have an important role to play. But there is another retention lever that retailers already have.

Customer service.

Not the marketing message sent to a customer. The interaction that happens when something has gone wrong.

A delivery is late. An order is damaged. A product isn't what the customer expected. A refund hasn't arrived.

The customer has a problem, and they have contacted the retailer to get it resolved. That interaction may be one of the most important moments in deciding whether they come back.

Your customer data tells you what happened. Their conversations tell you why.

Most retailers have made significant progress joining up their structured customer data: Orders, returns, delivery status and loyalty are connected, or at least close enough to give the business a useful picture.

But there is another half of the customer record that is much harder to use.

What the customer actually said: call transcripts, customer service tickets, chat conversations, emails, reviews and notes from previous interactions. This is often where the reason a customer is unhappy actually sits.

A customer might be categorised as high value and showing signs of lapsing. But that doesn't tell you they were promised a callback three months ago and never received one.

The structured data tells you what happened. The conversation can tell you why.

And that distinction matters.

Identity first, or nothing else works

Before a retailer can bring all of this information together, it needs to answer a deceptively simple question:

Who is this customer?

Guest checkout, multiple email addresses, marketplace purchases, loyalty accounts and orders delivered to different addresses can all make that harder than it sounds.

In one implementation, we achieved more than 99% matching across eight source systems using a combination of direct ID matching, fuzzy matching and AI-assisted matching.

The retail challenge is the same. The identity resolution cannot stop at orders and transactions. It needs to reach the customer conversations as well. Otherwise you can have a beautifully integrated sales history sitting next to a completely separate history of what the customer has actually been telling you.

What does a complete customer view actually look like?

The obvious answer is to connect everything and let an AI assistant query the underlying systems whenever somebody asks a question. But there is a problem with that approach.

If an agent has to search five systems every time a customer calls, the experience quickly becomes complicated.

A better approach is to continuously assemble the breadth of the customer history into a usable customer view: support history, orders, returns, billing, lifecycle stage, product information and previous conversations. Then, when something needs to be current, query the relevant system directly.

This is particularly important in retail. A customer calling about a delivery that failed an hour ago doesn't want yesterday's data. They need to know what is happening now. The goal isn't necessarily to put every piece of data into one place. It's to make the right context available at the right time.

Not every question needs the same AI. This is where I think some customer service AI strategies go wrong. We don't need one general-purpose bot trying to answer everything. Different questions need different approaches.

“Where is my order?”

That should be a direct lookup against the order management and carrier information.

“Am I still inside the returns window?”

That's a combination of policy and the order date. It should be deterministic.

“Why does this customer keep complaining?” That's different. Now we need to synthesise information across previous tickets, returns and conversations. And:

“Should we offer a goodwill gesture?”

The goal isn't to make AI answer every question. It's to make sure the right question gets the right type of answer, and that when a human is needed, they have all the information they need to make a good decision.

The bot and the human need the same customer view

This is also where AI can change the role of the customer service agent. The objective shouldn't simply be to reduce the number of contacts reaching a human. It should be to make both the automated and human interaction better. If the bot can resolve a straightforward question reliably, great. But when the conversation needs to be handed to a person, the customer shouldn't have to start again. The agent should inherit the context. They should know what the customer has bought, what has happened previously, what they've already been told and why they're contacting the retailer today.

That changes the conversation. The customer doesn't have to explain their history again. The agent doesn't have to search across multiple systems while the customer waits. And the retailer has a better opportunity to turn a difficult interaction into a positive one. There is still an important boundary here.

AI should not be used simply because it can be.

Where the answer needs judgement, empathy or a decision involving significant customer value, a human may still be the right person to handle it. But that person should have better information.

Five questions for retailers

1. Can your service team see the customer's purchase and interaction history in one place?

2. Can you reliably identify the same customer across your different systems?

3. Can you use customer conversations alongside structured data?

4. Can an agent understand what has happened previously without searching multiple systems?

5. When AI hands a customer to a person, does that person inherit the context?

If the answer to several of these is no, there may be a significant opportunity sitting inside your existing customer service operation. Because the next retention opportunity may not be another email campaign or loyalty promotion. It may be the conversation you're already having with the customer.

Customer service isn't just where you solve the problem.

It could be where you save the customer.

Published 12/10/2026