How Selsey automated 96% of customer conversations with Text AI Agent

Buying furniture online often comes down to one last question. Will this sofa fit my family and my living room? Will this fabric survive life with a dog? Will this table fit through my doorway? Selsey’s Customer Care team had the expertise to help customers make those decisions. But too much of their time went to questions that didn’t need it: Where is my order? When will it arrive? Is this model available?

Justyna Jamrozińska-Żelazna, Head of Customer Care at Selsey, sought out a way to automate these repetitive conversations and free up her team for the work where their expertise could have a greater impact. Today, Text AI Agent resolves 96% of Selsey’s customer conversations, giving the team more time to advise customers, follow up with potential buyers, and contribute directly to sales.

We’re moving from being a department that handles what happens after a sale to one that actively helps make the sale happen.

The customer

Selsey is a leading Polish furniture brand and ecommerce retailer selling across Poland and international markets. Alongside products from other manufacturers, the company develops its own furniture collections.

Its 22-person Customer Service team supports customers across Selsey’s e-commerce stores and marketplaces, covering the entire journey around a purchase, from product questions and order changes to deliveries, returns, complaints, and transport claims.

The challenges

Selling furniture online means asking customers to make expensive decisions about products they haven’t seen in person. That makes Selsey’s Customer Service team an important part of the buying process. But much of the team’s time was spent checking deliveries, confirming availability, providing dimensions, and answering questions about payments and returns.

As Selsey grew, these routine questions increasingly competed with conversations that required the team’s expertise.

I had people on the team who could handle a difficult complaint or advise on an order worth thousands, and they were spending most of the day telling customers where their package was.

Keeping up with the incoming queue also left little time to follow up with customers considering a purchase or those who had ordered fabric samples.

Availability was another challenge. Customers often shopped outside Selsey’s Customer Service hours, meaning someone on the verge of making a purchase might have to wait overnight or even over a weekend for an answer, putting the sale at risk. With a purchase worth several hundred dollars, nobody wants to guess. The customer either postpones the decision or goes somewhere else where they can get an answer right away. For Justyna, the goal was to give customers immediate answers to routine questions while freeing specialists to focus on conversations where their expertise could make a difference.

I wanted customers to get answers when they had questions, not when we happened to be at work.

The solution

Selsey introduced AI Agent as the first line of Customer Service, available 24/7 to answer product and order questions, recommend products, and provide information about availability, delivery, and returns. When a conversation requires judgment or a more individual approach, it is passed to a specialist.

Configuring AI Agent correctly meant writing instructions and keeping the knowledge up to date – and making adjustments when needed. It’s much closer to working with procedures than writing code, and procedures are my job.

Justyna led the initial setup, using Selsey’s existing knowledge and data from real customer conversations. Before launch, she tested AI Agent against difficult and ambiguous cases, including questions it couldn’t answer.

I needed to see that when it didn’t know something, it would say so and pass the conversation to a specialist instead of improvising.

She also set clear boundaries around what inquiries AI Agent should handle. It wouldn’t decide complaints, negotiate prices, or promise exceptions. These would be handed off to a specialist, or a customer could ask for one at any point.

If it’s a question with a clear answer based on the trained data or nature of inquiry, AI Agent can handle it. But complaints, negotiations, exceptions — anything where someone has to make a decision and take responsibility — stays with the team.

Once AI Agent went live, Justyna could easily observe how this setup worked with real customers. The first morning, she opened the conversations from the night before and found that customers had been asking about dimensions, availability, and delivery throughout the night, getting answers when Selsey’s team had previously been unavailable.

That’s when I realized these weren’t conversations that would have waited for us until morning. Before AI Agent, customers could see that no one was available, leave the site, and continue shopping somewhere else.

From that point on, improving AI Agent’s performance became part of the Customer Service team’s work. Specialists review its conversations daily flagging answers that could be improved and identifying gaps in knowledge. Those insights are then used to train AI Agent’s knowledge and instructions.

You don’t have to build everything perfectly from the beginning. Customers show you what’s missing. You see the conversation, add what AI Agent needs, and the next time it knows how to handle it.

The results

Today, AI Agent resolves 96% of conversations without human involvement. Product questions, delivery updates, availability checks, and other recurring requests are handled automatically, including in the evenings, overnight, and over the weekend.

Mondays were always the worst. We would come in after the weekend and already have more than 100 conversations waiting for us. Now we open the queue and there are usually just a few.

With most of the customer service chat volume handled by AI Agent, specialists can spend more time on complex cases and customers who need human advice and expertise.

My team stopped answering and started helping.

That extra capacity is also changing how Selsey uses its Customer Service team. Instead of focusing almost entirely on incoming requests, specialists can be more proactive with customers who are still deciding what to buy and help them move towards a purchase.

If someone orders fabric samples, they’re already seriously considering the purchase. Now we actually have the time to come back to that person and ask: did they arrive, what are you deciding between, can we help?

The same applies to high-value or more complex purchases. AI Agent can answer questions about dimensions, availability, delivery, and other product information, while specialists have more time for customers who need advice based on their individual situation.

At some point someone has to say: ‘In your place, I’d choose this one, and here’s why.’

That time matters beyond sales, too. Difficult complaints, unusual delivery situations, and exceptions can get more attention from specialists instead of competing with routine questions for the same capacity.

There are conversations where the customer doesn’t need another piece of information. They need someone who understands the situation, makes a decision, and takes responsibility for it.

For Selsey, 96% automation isn’t the end of the transformation. The team is continuing to expand what AI Agent can do while using the capacity it creates to take on more valuable work with customers.

We always had ideas for what we could do more proactively with customers. The problem was finding the time to actually do it. Now we finally have that space.