5 Strategic Recommendations for Your AI Chatbot

5 Strategic Recommendations for business people who are considering deployment of a natural language chatbot.

Have you have heard about “the one thing” productivity hack? If not, it’s the ONE Thing you can do such that by doing it everything else will be easier or unnecessary.

Chatbots too, have ONE thing.

I’ll get to that in a minute but first, some backstory. We built a natural language chatbot using IBM Watson. IBM has a solid suite of tools, cognitive services, and cloud infrastructure required to build sophisticated AI-powered applications.

A chatbot is one such application. As we got familiar with the Watson environment, capabilities, and limitations we came to places where we debated going left, right or straight ahead.

If you’re thinking about deploying an NLU chatbot I have a set of guidelines learned from doing it.

5 Strategic Recommendations

1. Vision is the ONE thing. What business problem is your bot addressing? What’s it going to do and how will people use it?

Our chatbot lives to solve a specific problem faced by airlines and some other travel industry businesses. Before we started making the bot I envisioned a person driving through town, both hands on the wheel, asking Siri to book a flight.

2. Type of chatbot. Can you picture my motorist cruising along with both hands on the steering wheel? What your bot does, and how it does it, comes from your Vision of it in action.

When we got deep into building the bot our focus naturally shifted to head-down developer mode. At one point, dialog buttons would have helped us simplify some work and advance the conversation. Dialog buttons are Yes/No; Small/Med/Large, options in a conversation. When we hit that point in development our vision led the way. It’s a voicebot first, no buttons. (Later, if the vision changes by way of product evolution then that’s fine.)

3. KIS. Keep it simple. AI-powered chatbots need to be trained to understand the language of the domain in which they are used. The tighter the bot’s purpose, the less data it needs for training. Starting with a few tasks for it to handle made development and improvement easier. It also makes measuring progress that much easier.

4. Set KPIs. Define your progress or success markers. Common performance metrics include ROI and CSAT. Chatbot ROI can come from cost reduction or revenue increases. CSAT from performance after the bot is adopted. A chatbot can ask the user how satisfied he or she was with the engagement in the same way a call center human would.

5. Experience & Talent. For small and mid-size organizations who don’t plan to engage the consulting arm of Preferred Big Consulting Company, get people who have experience with the platform you plan to use.

Artificial intelligence tools are new and constantly evolving. IBM Watson has extensive documentation but there were a few times when the documentation was obsolete or when we needed a functionality today that was on tomorrow’s product road map. In one situation there was no Watson solution to a dialog problem we had. This meant building an external solution to work with Watson. Resourcefulness and talent will push through those situations.

Finally, if you need some general advice on deploying your own NLU chatbot or IBM Watson in particular I can share my experiences in the comments section.

Leave a Reply

Your email address will not be published. Required fields are marked *