top of page

13th of June 2024

How Ami contains 50% of the incoming conversations for delivery-related queries

The issues typically faced by customers of retail companies are difficult for Ami to solve; Ami cannot make a delivery arrive faster, or refund customers who are unhappy with their purchase. This means that to serve retail customers, Ami will always have to escalate some number of queries. However, Ami can still assist customers in a wide variety of ways, increasing containment and lightening the workload of live agents.

Ami has been deployed on a large retail client for a number of years. Here, Ami deals with complex queries about late deliveries, returns and refunds and product questions. Ami contains between 45% and 50% of chats, even though retail-related queries are often impossible for an AI to solve. Ami uses numerous strategies to achieve this containment rate.


Ami has different ways of deflecting when a customer asks to speak to an agent. When a customer asks to speak with an agent in their first question, Ami will prompt the customer to state the topic of their query, so that Ami can try to help.

Since Ami has to escalate a customer if it doesn’t understand their query, Ami tries to understand the customer’s query even when the customer doesn’t provide a lot of information.

By prompting the customer to elaborate, Ami has an opportunity to give the customer advice or guide them to the correct section of the retailer’s website, without the need for a live agent.

Customer satisfaction

Key to improving containment is keeping the customer happy throughout their interaction with Ami. Often, issues that retail customers have can be quite frustrating, such as late deliveries, incorrect refunds and product quality issues. Ami has several strategies for preventing customers from getting frustrated.

Ami often has to ask the customer questions to be able to give them the correct advice. A customer trying to figure out how to return their faulty product might get frustrated at being asked numerous questions before getting advice. Ami keeps the customer from getting frustrated by letting the customer know that it will need to ask some questions to help.

Here, Ami tells the customer that it will help them track their delivery before asking the relevant questions. Ami can also keep context throughout the conversation. This way, Ami avoids asking questions the customer has already answered, which can make the customer feel like they are not being listened to.

Keeping Ami’s knowledge up-to-date

Ami will only ever provide content from the client’s website, or content given to Ami by the client. For this reason, it’s important that Ami’s knowledge is up to date. If Ami doesn’t have the information available to answer a customer’s question, Ami will have to escalate. Monitoring and updating Ami’s knowledge allows Ami to contain more chats.

When escalation is necessary

Because of the nature of customer queries in retail, some customers have to be escalated to a live agent. If a customer has been waiting for a refund for a considerable amount of time, or if they have a serious food quality complaint, it is important for Ami to recognise these queries and escalate appropriately.

Ami is able to distinguish between more and less serious issues. For example, Ami can recognise when a customer has to wait for their refund a bit longer, and when the customer has already been waiting for a while.

When Ami has to escalate a customer, Ami can gather useful information to escalate the customer to the correct agent to help with their issue.

Rather than escalating each customer to a random agent with no additional context, Ami can gather useful information and escalate the customer to the correct agent. For example, Ami can ask for the customer’s name, contact information and order number. This not only saves the agent time by already having useful information about the customer available, it also improves the customer’s experience.



Juliette van Steensel



bottom of page