My Experiments with Chatbot strategy

The above are some sample tweets by the Chatbot from Microsoft – Tay. It took Tay , less than 1 day to go from humans are super cool to being a mouthpiece which will make Nazis and MCP proud !! It was taught by humans to be abusive and racist.

As custodians of the brand name more than revenues, churn and customer experience, our strategy on Chatbots walks a tightrope between faster go to market, cost reductions on one hand and not getting boomeranged on another.

Now there are lot of such strategies and sermons floating around in market- am not sure how many of them were devised from the grind and hard labor, and how many of them look beyond UK or USA market to make one. From my experience across Telecom, Government, Healthcare and Banking industries across India and South Asian markets – I believe a Chatbot strategy should be very fluid one and should have these constituents.

  1. Start Small, Start Safe and Start Soon: Even though Chatbots are automating a very scripted portion of tasks its always better to start small and conquer piece by piece than go Big -Bang approach. Start with those transactions which are not the top 5 transactions at your customer care and more importantly which are not touching your core elements (and even if they do, they are not updating CRM or Billing). Example information about new products / services. While the Chatbot which is connected to CRM can identify the customer, all it does, initially is to address questions on topics like new services launched. If the customer is eager , hand over to real agent (point which I will discuss bit later). Also remember don’t wait for a muhurt for launching the Chatbots – get into the market ASAP.
  2. Assisted Learning : The core reason why Tay from MS failed was that it was cognitive and learning on its own from social media. But sure this boomeranged because there was no teacher or a guardian to guide it and tell it what was wrong. This hand holding by a human to bots is called assisted learning. Program the chatbots to be cognizant about abuses (tag the abusive words in a section of the learning database ) and create scripts such that it reacts in way ,it should, than learn , from abusive and mischievous rants.
  3. Handover / Hand off to Machine and vice-versa : The Version 1 of Chatbot strategy will not be the last , ever ! Its very difficult to look forward and create all possibilities and actions for customer conversations. PLUS there will always be some customers who are fatigued with conversation and need more human touch(ear, must we say?). Here comes the switching (or handover ) to real agent. This transition should be smooth and fluid. Vice-Versa , after the core tasks which need human interaction are done, the customer should be handing back to chatbots, so that your customer care staff can handle more important, more consuming and more prioritized tasks.
  4. Going vernacular: I just cant emphasize this more. Going vernacular is the key in markets like South Asia where English literacy remains a challenge. Further bots must be programmed to understand jargons and catch phrases. Hinglish for India and Spanglish for Puerto Rico. Also the variants of English with regions – example for Singapore market – the chatbot must be able to comprehend Singlish. Its understood that the chatbots will communicate back in native scripts just to make the user feel more empowered and powerful.
  5. Omni Channel : You are leveraging the Chatbot to hilt only when they are across the channels (except of course the physical retail stores , where you walk to a human). So have a FB , Twitter, company app, website , Customer care number Bots deployed.
  6. Map the Customer and Record the Transactions: An important point to remember when you are devising the Chatbot strategy is the first lesson from customer experience – know your customer and the previous transactions(read experience) which she has gone through. It will be a folly if your customer lands on FB , meets the BOT , asks for new healthcare plans and then calls the customer care ( a real agent or a bot) and they are clueless on what was the last inquiry all about. Go steps further , if you have the phone number of customer, and she logs on to FB using that phone, save her time , and make a guess on who she is . Of course, get that confirmed from her… your customer will like a smarter bot.
  7. A/B Testing, Revise Strategy : The best mantra to make your Chatbot successful is Revise, revise.. and revise. Check conversations flow (number of steps), Bot stability (speed, accuracy), Scale, NLP score, examine the response quality to human agents, others. There are lot of software and tools available in market for same , but its always better to have these tests done under strict supervision of your cross functional team and monitor it through brand, marketing and executive lenses.

I am sure these principles of Chatbot strategy will be valid not just with South Asian region but also across those markets where

  • the language of communication is not English (example Latin America) or where there are multiple languages being spoken (example Canada)
  • Market of Late majority and laggards : The market or company believes in technology settling down and stabilizing than being the first to experiment
  • Cost of going Chatbot (creating, implementing and executing chatbots) versus continuing with proven and low cost labor (continuing with as-is) is tight (example at Telcos in Bangladesh). If the cost of Chatbot to customer care is heavily tilted towards the former , it seems to be one way road only.

Makes sense, doesn’t it ?

Any thoughts/comments on this are most welcome !!

More on this here 

Share This Post

1 Comment

  1. Hi Suman, the screenshots of tweets fail to load, possibly a broken link. Please share the screenshots, will help set the context.

Leave a Reply