CHATBOT PITFALLS AND HOW TO AVOID THEM
RISE OF CHATBOTS
The rise of chatbots signals a transformation for mobility: the decline of mobile apps as we know them. Chatbots can be however compared to “apps” delivered through messaging platforms. This trend has been possible because messaging platforms offer a very natural interface for human communication — text or voice, and the users prefer them over other channels- testimony to this is there are more users on chat solutions than of social media sites. So, bots are here to stay, in fact the future is bots as it replaces most of the chores carried out by today’s mobile apps.
Brands are already leveraging bots on messaging platforms, not only to improve end consumer experience, but also enhance enterprise productivity — bots provide information on back office functions like HR queries, IT ticket status or function as an intelligent co-workers guiding and providing recommendations to sales or field force personnel.
CHALLENGES TO BOT BUSINESS
So, bots are a rage and are growing at a fast speed but there are multiple challenges that organization face when deploying chatbots. Even with AI maturing rapidly, most chatbots fail to meet their objectives. Some of the causes for failure are because of lack of strategy while few of these are because of technology limitations
Strategic reasons for failure
1) Irrelevant use cases: To try and be the first in their category to successfully deploy a bot, most brands are deploying bots for irrelevant use cases
2) Trying too much, too soon: Brands are trying to address too many problems with bots. Bots that do one thing well are better accepted then the ones that do many things poorly
3) Disconnected from existing ecosystem: A common tendency when building a chatbot is trying to recreate functionality from scratch. Bots are part of a larger ecosystem & creating a chatbot in a silo may not be very useful — leverage other systems.
4) Lack of proper human escalation protocol: Very few chatbots have an escalation workflow in place to let a human take over the conversation when the bot is unable to help
Technical/AI limitations
While bots translate spoken words into text, it’s very difficult to parse their meaning consistently — to pinpoint the right verbs and nouns and take action. Yes, progress is being made on AI at a very rapid pace but let some of the current limitations include –
1. Negative Words, Repeated Letters and Sarcasm — Bots are based on keywords for intent matching and use of negative words such as “No”, “Not” are still difficult for bots to comprehend, especially when used as sarcasm. This makes it difficult for bots to comprehend statements like “I’m not sure .. I’ll take a burger” OR “All my life I thought Air was free.. until I bought a bag of chips”.
2. Lack of context– Midway through a conversation if a user wants to change one of the inputs provided in the initial part of the conversation, most implementations today require restarting the conversation. Something akin to a back button or the ability to jump to a specific position in the tree has been the ask by most users
3. Implicit statements — Many bot implementations are not able to interpret implicit statements which are obvious to humans. “I’ll take a Paneer Pizza with onions and capsicum and a Mushroom pizza with plenty of tomatoes and chilly”. Some platforms are not able to distinguish the two pizzas and the ingredients”. However, with the advancement of AI and NLP, most platforms should be able to address this.
4. Hidden Context — Some bots continue to have an IVR style implementation for decision trees and hence user inputs such as “Choice 1” which are indistinguishable across different branches on the decision tree may cause AI tools to disambiguate ineffectively i.e. these tools lose context
5) Varied Utterances — NLP and language libraries have come a long way but bots continue to struggle with all combination of utterances for a specific intent. “How is the weather in Bangalore?” and “Is it raining in Bangalore?” these 2 questions have the same intent but some of AI tools are still not able to perform the intent mapping right
SO HOW CAN ORGANIZATION OVERCOME THESE CHALLENGES?
CHAT Framework- Organizations need to treat bot implementation as a journey and not a one-time effort
A technical approach to solve this problem may involve use of a wide variety of preprocessing techniques including stemming, lemmatization, tagging, creation of context storage modules, etc that when combined with standard AI/NLP tools help track context, remove redundant information, etc. But a technical approach alone may not be able to solve these issues and enterprises need to take a more pragmatic approach and treat AI as a transformation journey and not a onetime effort

C- Contemplate & Commence: Create an organization wide strategy but start with a focused pilot
– Though the recommended approach is to start with a focused pilot but it should still be a part of an overall strategy or a big agenda that organization wants to drive. We have seen even different units of the same organization starting their own AI journey independently e.g.- HR department funding a chatbot project and IT support starting another project with a different vendor, with no overall strategy in place. Following a disjointed approach by different business unit may not give the expected result
– Pick a small set of use cases to keep the pilot focused and manageable. Leverage data to identify right set of use cases. Pick use cases that are repetitive in nature and do not require too much functional analysis. Identify the success factors upfront and analyze shortcomings before moving to production
– Do not use chatbots in cases where the customer is already likely to be sensitive or irritated. Also, avoid irrelevant use case — emphasis should be on creating a meaningful use case rather than being the first to deploy a bot. First impressions last, so surprise and delight customers by starting with a core set of relatively simple use cases and deliver those well before expanding
H- Hybrid approach: Combine with human agents
– Take a “hybrid” approach, with human agents closely supervising and training the chatbots and able to intervene in conversations when required. Build the botd with a seamless handover mechanism because there will be scenarios the bot is unable to answer the query. As the bots evolve the human agent component could be reduced
A- Acceptance Tests: Conduct user acceptance tests
– Thoroughly test the chatbot with users before launching to a wider audience. Ensure the conversation flow is smooth and that the chatbot is desirable
T- Training: Plan for end user training
– Treat bot deployment as a bigger exercise than just technology implementation. There is a big components of change management that needs to be taken care of after the bot implementation. Workforce needs to be educated that bot will not replace them but will help them be more productive as intelligent co-workers. Incorporate bot usage training especially for B2B/B2E use cases. This will provide end user with insights on the BOT capabilities
BEYOND THE STRATEGY & TECHNOLOGY- THE ART OF DESIGNING CONVERSATIONS
Another very important point to consider while Chabot implementation is the conversation design methodology. The chatbot conversations need to be designed not only keeping in mind the human aspect of conversation but also the brand characteristics.
A good conversation should keep in mind the target audience but also reflect the image & personality of the bot itself. It should be clearly articulated with clear entry & exit points and decision trees & branching logics. Conversation flow should be intuitive without having to provide explicit guidance on how to operate the bot, the conversation should be human like without pretending to be human. Handover to live agent should be well integrated and seamless
CLOSING THOUGHTS
Chatbots are effective tools for companies to improve internal efficiencies as well as enhance customer experience. But without a clear understanding of current limitations enterprises risk an experience that is frustrating and useless. Hence in addition to tracking progress on the AI/chatbot technology landscape, enterprises also need to treat AI implementation as an organization wide journey which impact multiple stakeholders. The systematic approach with the knowledge on current limitations will help create bots that are rich in user experience to propel their brand in this messaging era
