7 Questions Business Leaders Must Ask Before Starting on a Conversational AI System

An AI-based virtual assistant is a class of conversational AI system which helps with tasks, often in customer care domain. Virtual assistants have sophisticated language (text and voice/speech) capabilities, knowledge retrieval skill, process automation and integration support for enterprise systems such as customer data, voice gateways and for handoff to a live agent.

Building such conversational AI systems can be deceptively hard. The key elements include:

• Business goals (rules / policies, controls, integrations)
• Natural Language Capabilities (NLU, NLG) which make the assistant intelligent, helpful and powerful
• Conversation state and context
• Capabilities such as search, display, channels
• Nice to have: low-code UI which allows business users to build flows and tasks

There’s more.

The takeaway for executives and decision makers is that even though most of the technical focus is usually on AI-driven language capabilities there are many other pieces of the puzzle that must fit together to deliver a wholistic solution. The illustration below — inspired by the ML technical debt paper — attempts to drive home this point.

Conversational Assistants — © Babar M Bhatti

Here are the top 7 questions that executives (leaders, decision makers) should ask about a conversational AI initiative:

1. Do we have key goal, business justification and success metrics? what exactly are we solving for (eg large volume, low-complexity or something else.)
2. Is there clarity around scope and awareness of the complexity level — or, do we have clear requirements? (what are we getting into? how will it play out over time, what phases are there to divvy up the work: languages, custom speech models, searching our knowledge base.)
3. What is the expected cost? (enterprise virtual assistants can take 12–24 months to build and cost somewhere in the $250K to $2M range, according to Gartner.)
4. What data do we have? how good is the data? how much data engineering is needed? (how much do we know about our customer issues, what happens at our call center, what is the top…