CRM Strategy Department Manager Naho Iiyama
Yamato Contact Service is a Yamato Group company that handles customer service for Yamato Transport. In addition to handling calls for collection and redelivery of parcels, the company also handles contact services for various companies, including group companies and general businesses, and receives 2,000 million calls a year.
We are very particular about improving quality in order to provide a service that is considerate to each individual customer. We have many certified personnel, including 137 who are qualified as supervisors under the Japan Contact Center Education Examination, as well as personnel with operation management qualifications, and we also have one who is a COPC Certified Registered Coordinator, which is the international standard for contact center management.
Improved operator quality and work efficiency with the introduction of AmiVoice Communication Suite2
We were considering introducing a voice recognition solution to achieve our goals of "high quality, high efficiency, and high added value." After comparing various products, we decided to introduce AmiVoice Communication Suite3, as it seemed to be able to achieve these three objectives. Since we were considering using it for operator support, we found that the recognition results were displayed in speech bubbles, which made them intuitively easy to understand, and the screen was easy to read, so it was easy to use. Another point was that the recognition rate before tuning was quite good.
AmiVoice Communication Suite2 transcribes the speech of each operator in real time, and is used for the purpose of improving quality and work efficiency. For example, we have operators who have just started work look at the transcripts of their actual conversations and use them to improve their speaking style and other aspects of their work.
AmiVoice Communication Suite Cloud was introduced to acquire VOC.
CRM Strategy Department Manager VOC Analysis Specialist
Mr. Shunsuke Suzuki
We had a challenge to improve customer experience and rebuild customer touchpoints by utilizing VOC (Voice of Customer). At that time, we considered using AmiVoice Communication Suite2, which we had already introduced, but since it was installed at a fixed desk, the tasks for which VOC could be obtained were limited. Our company is contracted to carry out various tasks for multiple companies, and we were internally considering whether we could obtain and analyze VOC for each task. Around that time, we heard that the cloud version, AmiVoice Communication Suite Cloud, was about to start service, and we started considering its introduction.
In order to understand the current situation, I wanted to first convert the conversations into text and try various things such as analyzing them. Unlike before, there is no need to install various equipment such as a server or PBX, and you can easily convert the recorded files into text. Furthermore, since the charge is based on the call time converted into text, you can easily try various things, such as converting the conversations into text.
For the tasks that were originally introduced with the aim of improving quality, the recognition rate was high from the start, as operators were repeating the text properly and the quality of the response was high. In fact, when we performed text mining on text with a high recognition rate, the topic content rate and VOC acquisition rate were higher than expected, and as we continued to operate, we were able to get an idea of how much VOC we could acquire and what kind of analysis we could perform if we had this much call time.
Post-operation issues
In fact, at the end of 2016, we received a request to collect technical support VOCs to create an FAQ, but when we tried it out, the recognition rate for operators was below 7%, and we were unable to obtain the valuable VOCs we had hoped for. After investigating various options to see if this was a problem with the recognition rate, we discovered that there was an issue with how we approached customers.
When making technical support calls, the caller often uses katakana technical terms, numbers, and demonstrative pronouns such as "that's it," which was one of the reasons why VOC were not collected.
When creating an FAQ, it is necessary to clearly understand what the customer is confused about and where they are calling. Rather than using demonstrative pronouns, operators need to actively listen to what the customer has to say.
We analyzed the misrecognition and identified what was going wrong by reviewing the operation of both people and the system, including improving our approach to customers through operator training, tuning the voice recognition engine, and registering technical terms in a dictionary.
Initiatives to improve recognition rates
CS Promotion Department CS Promotion Chief
Mr. Momoma Kobayashi
We transcribed the correct sentences from the actual conversations, compared them with the recognition results, and investigated the cause of the misrecognition. This is a very painstaking task, but it is a very important part. We checked for misrecognition for each utterance while actually listening to the audio source.
For example, the reason why "WEB" is written as "Ebu" is due to a problem with pronunciation, "Reissue" is written as "Rebox" is due to a problem with the speed of speech, and "Yamato" is written as "Yamato" is due to a problem with dictionary registration. By analyzing where in the conversation the misrecognition occurred and why it occurred, it is possible to determine whether the problem is on the operator's side or the system's side. From the correct sentences, four types of data were created: dictionary word learning data, transcription learning data, unnecessary word data for text mining, and training data for operators.
The dictionary word learning data and transcription learning data were directly reflected on the AmiVoice side for tuning, and the unnecessary word data was registered in the text mining tool. The training data was then used to train the operators.
Operator training with an emphasis on customer service quality
When we talk with the VOC analysis team, the main topic is the recognition rate, but the recognition rate has no direct relevance to the operators on the front lines. Therefore, when we tell the people on the front lines, "Do this to improve the recognition rate," they respond, "Why do we have to adjust to the machine?"
The most important thing for an operator is the customer in front of them. The training focused on how to communicate clearly with customers and what benefits it brings. For example, repeating back what the customer said can help prevent mistakes on important matters and demonstrate the intention to "listen to the customer." By clarifying the topic, you can confirm what the customer does not understand and grasp the main points before speaking. This is the perspective of customer service quality, which puts customer service first.
As we continued our efforts, we realized that collecting VOCs and improving our FAQs and systems would ultimately lead to improved customer service quality, so we made improvements on-site with an emphasis on the customer's perspective.
Subsequent effects of introduction
As we continued these efforts, we were able to achieve a recognition rate of 90%, a VOC acquisition rate of over 130%, and an average 10% improvement in response quality scores. The improved response quality made it easier for customers to understand, and reduced the time it took to resolve inquiries. Although operators are actively working to acquire VOC, the average talk time per call has remained the same as before.
We show the operators the text of the previous day's call, and it seems like it's very refreshing to see what was said in the speech bubbles. By reviewing the conversation by themselves, unnecessary conversations were reduced, and as a result, productivity did not decrease. Since the text and sound are linked, it also led to improvements in voice and speaking style, such as when the sound breaks up when speaking loudly, making it difficult for customers to hear.
Furthermore, through this technical support VOC collection project, we realized that response quality and recognition rate are synchronized. Without quality, valuable VOC cannot be obtained. Because it is text data and not voice data, it can be used in a variety of ways, such as through text mining or as AI data. In order to obtain VOC with maximized quality and quantity, the quality of the contact center and operators is extremely important.
Future utilization
I would like to expand what I have done in technical support to other areas of my business. By converting voice to text, I was able to see the issues in how to approach customers to obtain VOC. I would like to have the operators look at what they have said, and use the realizations and new discoveries to lead to further improvements.
Message to customers considering purchasing
With the advent of cloud services, I think the hurdle to introducing voice recognition has been greatly lowered. It's very easy to convert phone calls into text, and there are many things you can learn by looking at the text. Improving quality may seem abstract, but I feel that voice recognition makes it possible to visualize quality.
I think there are many companies that want to mine text from conversations at contact centers, but if they don't pursue a high recognition rate, the mined data will be biased. If you want to use VOC analysis for something, it's best not to give up on the recognition rate. Improving the quality of service will improve the recognition rate.
I think that voice recognition technology has reached a fairly high level, so the rest is down to our efforts and how we collect data to utilize it. If it doesn't work, I think the root cause is the calls that generate the data. The quality of the data is very important.
It might be a good idea to try out a cloud service first, and then deploy it on-premise once you have a good understanding of how it actually works. In that sense, I think AmiVoice Communication Suite Cloud has great value.
Customer profile
Yamato Contact Service Co., Ltd.
Responsible for customer support operations for Yamato Group companies and general companies. Handles 9 million calls per year at 15 centers in 2,000 locations nationwide. Since its establishment in 2003, it has been the contact center for many companies in Japan, regardless of industry or size, including distribution, manufacturers, local governments, producers, mail order and e-commerce.
