How SpeechKit helped improve offline customer service for British American Tobacco

British American Tobacco (BAT), like many other companies, is always working to improve customer service in its stores. Previously, BAT managers had manually analyzed just 8–10% of consultant interactions with customers. They wanted to increase this figure to 100%, and understand that they needed a solution that offered cloud technologies within a wearable device, like the audiobadges developed by VocaTech based on Yandex SpeechKit’s technology. These audio badges automatically record, recognize, and analyze what their salespeople say.

However, BAT automatically collected and analyzed 100% of all of their 240 consultants' conversations using the audiobadges. The summary analysis became available the following day, providing a digital basis for effective management decisions.

Observing that consumers are becoming increasingly knowledgeable, businesses like BAT recnognize the need to communicate with their audience better and anticipate what level of service their customers expect. In light of this, BAT sought to improve customer service quality by making sure that store consultants interacted with visitors politely, providing comprehensive and accurate information.

In the past, BAT had recorded and analyzed consultants' conversations in shopping centers via the tablets that consultants used to present products. Recordings were manually transcribed and analyzed. Managers checked whether consultants followed set scripts and used keywords, and collected information for several other metrics. This manual analysis was time-consuming — only 8–10% of conversations could be processed, which was insufficient for managers to make effective decisions.

In 2020, BAT decided to automate the process: at the stores were consulants sell BAT’s tobacco heating system in Russia, they began recording and transcribing consultants' conversations, analyzing the data, and making adjustments to consultants' work processes in real-time.

The task before BAT was to automate the analysis of each recording and getting summary statistics on sellers by region and period.

After looking at the available offers on the market, BAT chose to go with Yandex SpeechKit. BAT turned to Yandex Cloud, who acquainted them with the Audiobadge solution developed, using SpeechKit technology, by Yandex Cloud’s partner Vocatech. During the launch of the pilot project, it showed the highest indicators in terms of sound quality (including the amount of background noise) and the speed of speech recognition.

VocaTech created customized BAT audiobadges with a built-in recording devices for each consultant, with a note indicating that they are being recorded. Audiobadges can be used during employee shifts using a simple authorization mechanism. When the consultant speaks, the device saves an audio recording of their speech to its internal memory.

At the end of their shift, when the consultant leaves the audiobadge to charge, the device synchronizes with Yandex Cloud servers and sends data to the cloud.

First, the VocaTech filtering module processes the audio recording by removing unwanted sound (in this case, the shopping center’s background noise, as well as the customer’s speech).

The processed recording is transcribed using the SpeechKit service for further analysis, using VocaTech’s technology, which finds keywords and calculates the necessary metrics (the number of interactions, cold approaches, product mentions, compliance with scripts, politeness, payment method, product shortages).

The data is collected in an analytical dashboard for BAT’s specialists to work with.

Combining Yandex Cloud’s SpeechKit with VocaTech’s audiobadges, BAT got a fully automated speech recognition and analysis system which helped them process 100% of consultant interactions with customers. Every day, BAT managers receive the metrics and results for the previous day’s interactions, which helps managers to quickly make changes to consultants' work processes and pay special attenion to employees whose results deviate significantly from average values.

In addition, each salesperson is assessed based on the results of the speech analysis system, which factor into their financial compensation. This motivates BAT employees to do a better job by following the recommended conversation scripts, being more polite, and customer-oriented.

In the future, BAT, VocaTech, and Yandex Cloud specialists will continue to improve the Speechkit user recognition model taking the the company’s specific terminology and its customers into account. There are plans to expand the project to include consultants and other product categories, totaling over 150 people.