The Practicality of Implementing #ChatBots

Here’s an article published from our Founder & CEO, Ujwal Makhija‘s blog:

It’s been three weeks since my last post about #ReadyFor2020 and how bots are there to change the future. It’s also been close to a similar amount of time since the launch of Phonon’s Chatbot Trixie! Have been asked many questions through the sales process. I thought it best to jot down some of my thoughts.

Chatbot Implementation

 

 

Response Time: We’ve had a close to 90% positive response rate (usually within seconds) to tweets for flight status received on those #hashtags. Usually quicker than a 20 minute average response time and approximately 50% response time from manual #hashtag tracking. (Reminder to self: get a quantitative analysis done). And that’s great. We’ve had some excellent likes and appreciations too.

 

 

 

Delays at Airport

Tweeting About Delays at Delhi Airport

 

 

 

Predictive Responses: While our information is sourced from public sources, our Bot tweeted the updates about Delhi airport delays in the last two weeks accurately. Here’s a tweet from the drone incidence delays.

 

Transactional

Keep quiet when you don’t understand it

 

 

The Bot is Supposed to Transact and Not Converse: This is a death of many technology implementations. We go for the overkill. How often does a customer make general conversation with a call center employee? But why do line managers often want Bots to be super-humans. Many discussions I have had have had the product owners ask the bot the toughest, most convoluted question they can. Nope, that’s not how it works! That’s probably something your top performers do. Your bot is supposed to reduce the work of bottom-performers so that you can afford better top-performers.

The best implementation strategy in bots is to focus on transactions and compromise on some more false negatives to have zero false positives. So, if in doubt, ask the bot to STFU and get a human to intervene. :-)

Here’s an example of how our bot STFU when it was asked a question beyond it’s capability. We tracked #jetInstant and there was a counter question to which our bot kept quiet.

 

Accurate Information

Accuracy is Paramount

 

 

 

 

Bots reduce human errors and ambiguity. Here’s an example of how our bot made a reply unambiguous and more importantly accurate. A flight 9W-856, was LKO/DEL/PNQ. The manual reply made a wee bit of a mess! It gave the departure time of the flight at LKO as the status at DEL. Luckily DEL departure was after the LKO departure. And compare that with the response of the bot!

 

In a nutshell – #ChatBots are the future. But be clever. Don’t use them for #Chatting. Use them for #Transacting.