A few considerations to keep in mind when you launch your conversational business chatbot
Congratulations! You decided to upgrade your skyrocketing customer service activity with a NLP (1) solution using a chatbot, a great scalable and fast-responding tool for your business.
Additionally, the ability to handle free typing from your customer inquiries is beneficial for
- the brand image, your company cope with the latest tools
- the customer segments which culturally value human services over bots or forms, and still enjoy the user experience in this situation
In this article features some considerations to design your bot accordingly to your business and avoid many obstacles. Your chatbot is product-oriented : think of the semantics around your product and the Call to Action sentences that are meaningful to your business compared to their generic meaning in common language.
(1) Natural Language Programming - ability from computer to understand human spoken and written language
1 ) Challenge the intents identification of your dialog flow
Your bot answers the customer with a set of automated answers (which syntax can randomly vary for a human look-alike experience), and which can contain dynamic data depending on the customer current purchase or profile data.
This is called the dialog flow, the bot is identifying the topic - or intent - from the inquiry and it answers accordingly.
As you may understand, the intents tree can expand to a thousand leaves even though some intents can be assisted by bot’s conversational “small talk”. This is the core of the technology, your bot assists or activates actions to fulfill most of the customer’s needs. You want to address the most precisely the issue and it is crucial to identify correctly the intent, not to deliver an unrelated answer or reply the dead-end message “I cannot understand, I’ll redirect you to … “ which will annoy the customer. The intents identification can be very tricky as shown in the following use cases.
Case #1: thebot understands which object executes the action (grammatical case)
Case #2: the bot understands unrelated action to business and assigns it to the correct object (vocabulary case)
The main obstacle belongs to the writing comprehension skills field and is still a problem to tackle for scientists. To avoid this situation, we advise measuring a confidence level score of the bot to identify the intent and set a minimum grade to allow the bot reply. Otherwise, it is reasonable to handover the chat to a human operation executive. After each chat, the operation executive ticks the customer intent to confront results with the bot. This allows to follow on a dashboard some valuable KPIs like the bot’s confidence level and bot’s mistakes metrics, and engage reinforcement trainings on the least performing intents identified.
From this dashboard, we can read that the bot struggles to identify shipping prices inquiries to Egypt. However, when the bot replies to this intent, it addresses correctly the topic in 85% chats.
2 ) Reinforce the vocabulary designed to your business
The vocabulary is closely linked to your business intents, although it is not the only factor as grammar can reverse the sentence meaning. When training your bot, input as many chats as you can which highlight your product - its name, its aspects and its attributes - as well as brand names from partners or competitors. Imagine that your products are gardening tools, your bot must identify the tool from its use or its aspect if customer cannot name it.
This consideration broadens to multiple languages chats, when customer writes foreign words or idioms. The natural language programming has mainly been language-centered, respectful to formal language, and very prolific for the main worldwide spoken languages. However the biggest worldwide marketplaces - Asia & Africa - experience a lot of mix spoken languages among their populations. Customer would swap from English to regional dominant language for better comfort or practicality, even if they selected a bot’s language when entering chat room. Here is an example.
This case highlights the product-oriented needs of your chatbot, a regional untranslated product wouldn’t be understood by the default English conversational bot.
The multiple languages barrier is even worse to overcome for languages which adopt a lot of foreign words and idioms. For instance, English-tagalog is the spoken and written language in the Philippines, although the chances to launch a chatbot are left to whoever dares. The main issues are the unpredictable swap of vocabularies, prefixes and suffixes added to English words, and the lack of written documents as data sources to train the technology.
Without promising start-ups developing the research in mixed languages, companies miss the opportunity to reach hundreds of millions of literate-enough people, through a scalable automated tool they could interact with.
For further info, you can meet us, call us, message us or take a look at our website www.drive-agency.com
Talk to you soon!
Created in 2017, DRIVE is a digital performance company providing clients with a data-driven marketing strategy, powered by skilled talents, reliable data and scalable technology system. Certified on major analytics and conversion solutions such as Google Analytics, AT Internet and Adobe Analytics, we combine our expertise to meet your business expectations. With experience in both brand and agency matters and working within international environments for more than ten years, we can accompany you in facing your challenges.