"Hello, how may I help you?" Using Welcome Agents in customer-centric retail environments

Organizing waiting times as efficiently as possible can be a make or break factor in terms of the customer experience in sales environments with extensive consulting. Requests vary from customer to customer: Does the customer require detailed advice on a specific product or does he just want to pick up his order in the shop? Both are typical retail scenarios, but with completely different time and effort requirements.

In order to avoid unnecessarily long waiting times during peak hours, assessing the purpose of the customer’s visit and prioritizing accordingly are essential steps. NTS Retail has responded to this challenge with the development of NTS welcome manager, an innovative software solution that uses efficient queue management to optimize the customer experience in the retail sector. The solution relies on a "welcome agent", a shop employee whose job it is to greet the incoming customers and to determine the reason for their visit. The Welcome Agent can then present the customer with a predicted waiting time, queue them in accordingly or arrange an appointment later on.

For a project as a part of a TM Forum initiative research was conducted as to whether the role of the welcome agent can also be taken over by a robot, which records the customer's request and name by means of speech recognition. In order to implement the solution, Qihan Sanbot Elf was used as the robotic agent and IBM Watson as artificial intelligence for speech recognition and conversation control.



Creating a natural conversation

IBM Watson as a cloud solution played a key role in the implementation process. Three Watson APIs were used: speech-to-texttext-to-speech, and Watson Assistant. Speech-to-text analyzes the recording of the spoken input and converts it into text. It uses an algorithm to convert people's speech into text, which is getting better and better through the use of machine learning. The text-to-speech API gives the robot the ability to verbally respond to what has been said. The corresponding answer to the interpreted input is transmitted as text and transmitted via the API to create an acoustic vocal output.

The Watson Assistant service handles the interpretation of the spoken word and the semantic analysis. The recorded and converted text is arranged in a tree structure representing the conversation and also semantically analyzed in order to evaluate the spoken word in terms of content as well as intention. By means of machine learning, this algorithm is getting better at recognizing what the human counterpart wants to communicate.

In order to "facilitate" the analysis for the artificial intelligence and to make the process as efficient as possible, it was considered in advance how a typical welcome talk in a telecom shop can be designed in such a way that the necessary information is gained by answering a series of yes / no questions. A display on the upper body of the robot can also be used to show possible answers.

The cloud-native structure of IBM Watson makes it possible to access the processing power required to complete the complex analysis process completely independently of the client device (in this case the robot). The modeled process thus docks to the interfaces of the cloud services and receives feedback almost directly. The device itself provides the visual and audible output.


"Roberta" accepts customer requests

The robot, named "Roberta" at NTS Retail, is a cloud-enabled service robot manufactured by the Chinese company Qihan. This robot has cameras and sensors to analyze the environment, as well as microphones and speakers to verbally interact with people. It is controlled by an Android tablet. Thus, "Roberta" is able to approach and greet people.

In her role as Welcome Agent, "Roberta" welcomes newly arrived visitors and asks them about their concerns. The customer is then asked for his name and added to the waiting list with the phone number and photo. In the meantime, customers can leave the store in between and receive a notification as soon as a consultant becomes available. Thanks to the photo taken, the consultants recognize the customer more easily and can address them directly.



Digital reception and queue management in the telecommunications sector

"Roberta" has been integrated into the NTS welcome manager workflows and optimized specifically for typical telecom retail workflows. This includes, for example, the exchange of SIM cards, the delivery of equipment for repair or product-specific advice. A particularly suitable field of application, as the telco space attracts a very diverse clientele with a wide variety of support cases and sales scenarios, resulting in high stress levels for sales staff during peak hours. Although the requirements for an industry-specific queue management solution have been defined for the telecoms sector, most of the added value of the approach can also be transferred to other industries. Digital solutions for queue management seem to be a promising answer to the increasing demands in customer service environments with high information requirements in the consulting process.

The key is not only to save time through efficient business operations, but above all in the automated exchange of information between employees. Since the consulting topics are also recorded, the employees are able to prepare in detail for each consultation. The customer does not have to explain his concerns several times, but is guided seamlessly into a conversation with a well-informed consultant. The experience from product implementations by NTS Retail strongly suggests a positive effect on up- and cross-selling, as customers are provided with offers, which actually meet their needs and requirements.


Factoring in emotions in the process

Possible extensions to the integration of a robot into the processes is still being researched at NTS Retail. In a next step, the idea is to analyze the communication with regard to the conveyed emotions and to determine the intention of the customer. Is the customer disgruntled? Is the matter particularly urgent? In interpersonal communication, many nuances exist apart from the semantic level and they play a crucial role. The exciting task going forward is to "teach" the robot to be able to respond to these nuances in a seemingly natural way.


We want to thank IBM for the great cooperation and their support in implementing the Watson APIs.
Initially, the article was posted in the IBM Think Blog DACH.