AI Chatbots: 5 Key Considerations for CXOs

Adarsh Srivastava
Published in
9 min readDec 3, 2019


As enterprises worldwide ride the digital wave of Conversational AI, there are key considerations to make before incorporating a sustainable roadmap towards Digital Transformation with AI chatbots for enterprises.

The following article accentuates key criterion for assessment of chatbot companies, and helps in setting a base-line for successful deployment of multilingual-omnichannel chatbots for enterprises, that are scalable, robust and secure.

Businesses off all sizes are adopting conversational interfaces to make inroads towards Digital Transformation
Photo by Samson on Unsplash

Artificial Intelligence is a muse of great promise for CXOs, as they look to adopt the technology to improve customer experience, reduce OpEx, augment 24*7 sales, marketing, support, and boost customer agent-productivity.

84% of businesses say AI will enable them to obtain or sustain a competitive advantage.

However, there’s more than meets the eye, as a contrarian viewpoint is offered by the fact that many CXOs are yet apprehensive about adopting chatbots. 9 on 10 enterprises think that AI chatbots need to get smarter before they can be adopted.

And a general lack of stand-alone solutions has made adoption further difficult, with 47% of CXOs claiming that it is difficult to integrate AI into existing systems.

As enterprises are on their way to adopt AI-driven chatbots or conversational agents for customer service automation, augmenting sales and marketing, there are some quandaries to consider.

AI Chatbots have been touted as key digital convergence for enterprises and customers
Conversational AI and Chatbot solutions, to make critical inroads towards Digital Transformation, are being adopted by enterprises of all sizes.

What are the questions enterprises need to ask to bring their digital transformation strategy to fruition with AI chatbots?

This article depicts the need for robust language technology, exploration of business-specific automation and viability of solutions with enterprise needs.

We explore the following points:

  1. Guarantee of ROI by boosting business-specific KPIs.
  2. Enterprise need for end-to-end life-cycle management of a chatbot.
  3. Robustness in Language Understanding Engines and Multilingual Solutions.
  4. Human-in-the-loop-AI with Contextual Routing.
  5. Compliance to scale and security.

1. Do AI Chatbots guarantee ROI and can your vendor claim it?

Conversational AI and AI chatbots for enterprises
A business transformation framework should assess plausible solutions to set a base-line and measure for success.

The possibilities of Conversational AI augmenting various functions in business and customer service are endless. The right approach is to identify key long-term and immediate goals to boost critical KPIs that help in realizing ROI. Successful implementation of bots is directly commensurate with how bots can boost these critical KPIs.

These goals could be anything from cost-cutting to boosting employee productivity, automating business processes, implementing new digital strategies or even finding new avenues of service and sales through Omnichannel experiences.

Laying out business priorities helps in identifying key use-cases for automation and calibrate a definitive and measurable strategy for implementation.

Also read:”The True cost of not embracing digital transformation.”

AI Chatbots today, are known to automate 80% of first-tier queries in support, improving lead generation, and improving a brand’s visibility with a wide language-spectrum of audience.

It has led to a 30% savings in OpEx in support, saved support agent’s time to empower them to work on critical business matters and improve CSAT.

However, understanding that business requirements and use-cases that originate from it are subjective, help in understanding the extent of healthy and non-disruptive automation. This is the first tangible understanding of ROI from chatbots.

For instance, the most readily identifiable ROI is saving operational expenses in support. The percentage of automation and first contact resolutions achieved by bots directly helps in determining the reduced need for agents to perform repetitive tasks.

In turn, key measurements of success should be decided and agreed upon as the basis of deployment.

2. Is the Chatbot solution built for end-to-end enterprise needs?

End-to-end platforms and services from chatbot companies are pivotal for bot life-cycle management and unified engagement view.

Conversational AI elements are in silos today, and 43% of CXOs around the world say it is one of the major barriers towards adoption.

Platforms today allow developers to build a bot and deploy on particular channels. Enterprise needs, however, require end-to-end lifecycle management of the bots, which involves provisioning the scope of the bot, training, deployment, analysis and reinforcement.

Some tools and frameworks allow enterprises to build chatbots and deploy on particular channels. However, enterprises need a wholesome and robust service to not only have bots deployed but to provision its scope, test interactively, deploy and manage their experience throughout the lifecycle, including latent learning and upgrading.

Another key requirement is to have a mechanism to analyse engagement and emotive experience of the users to allow for successful reinforcement of the knowledge and CX that the bot offers. Does the vendor offer a custom analytics dashboard to track important metrics to draw these conclusive insights?

If platforms do not offer such stand-alone solutions, can the solutions be tailored to fit the business needs?

How feasible are the integrations?

How easy is it to handle the bot’s lifecycle during development and post-deployment?

These important questions, if addressed, help companies save massive costs and time.

3. Is the conversational agent truly natural and intelligent?

Perhaps the greatest source of scepticism in CXOs about Conversational AI is its language understanding.

In a study conducted in the US, 9 on 10 CXOs confounded that bots need to be smarter before they could trust them.

Language understanding to understand the intent of the customer-query, eliminating erroneous conversational flows to avoid misinformation, retaining critical information from chats, and offering a positive emotive experience can all be reduced to the intelligence of the bot’s NLU (Natural Language Understanding) engine.

The following are a few key considerations to make while assessing the language understanding of bot platforms -

a. Context-Aware Dialog Management

Conversational Design Robustness leads is key for realistic and useful Customer Experience.

The need for bots to align with the foibles of a human while conversing is key to creating a wholesome conversational design that doesn’t break while automating business queries.

Dialogue managers come in handy while tracking the state of any conversation with the bots.

We digress from one matter to another. The ability of the dialogue manager to track context and initiate a flow particular to that context is key to delivering a useful experience.

Also read: “Make conversations flawless with a Dialogue Manager”

Their ability to further reset the context as the user changes his nature of the query, while also retaining information from the previous context, is what allows the user to have a natural conversation with bots.

Failing to retain context and information forces the bot re-initiate redundant conversational flows and repeat questions. This can be a major cause of ire of your customer as conversations relapse.

b. Sentiment Analysis for Positive Emotive Experiences

Positive emotive experiences and acute personalization come from using critical data points to help create empathetic experiences.

The key to creating loyalty in today’s world of complex customer experience standards is not only offering prompt and round-the-clock service with consistent information across channels. It is also determined by offering positive emotive experiences.

The customer doesn’t just want to be treated as another account, rather as an individual with personalized experiences.

In light of this realization in recent studies, NLP has made great strides towards capturing cognitive signals from conversations to analyse user-engagement on new levels and rediscover value-exchange from insights that are otherwise covert from the business-view.

Moreover, it is important to use these signals to adapt bot responses to user emotions and create an empathetic customer experience.

The market is rife with Chatbot solutions that offer superficial and synthetic experiences and there is no real mechanism for businesses to understand their customers’ sentiments. CXOs need to find the right vendor that incorporates these aspects into their solution.

c. Fluently handling out of scope queries.

Business intent automation should also include out-of-scope queries as fluently as those that are in scope.

We read at the beginning of this article how the scope and the extent of automation in support, sales and marketing can be limited and subjective to business needs.

The conversational agent also needs to be aware of the context that it does not handle, to promptly inform customers that they will need human assistance without wasting any time trying to classify the user query as one within its scope of handling.

d. Learning Management

Post the deployment of an AI chatbot is when it truly goes through the litmus test, as various scenarios and measures of learning are discovered in the process.

The bot’s need to continuously add to its knowledge through a mechanism to identify errors and reinforce its domain and language understanding is paramount to its sustainability in the long term.

  • Active learning allows for interactively testing the bot and correcting its responses to feedback to NLU model. Bot trainers must be availed with a way to verify bot responses that have a low confidence score and be able to validate or correct its responses and feedback to the model.
  • Flagging failed sessions post-deployment: The platform must incorporate a mechanism to collect sessions that the bot wasn’t able to handle due to its inability in correctly understanding the intent of the user query. Once these sessions have been collected, there should be a way for a support agent to inform the NLU with the correct scope of that particular conversation.
  • Reinforcing the learning by making available a mechanism to re-train the bot on live data collected from chats can significantly boost its performance.

e. Handling Multilingual Conversations naturally to eliminate Lossy Translation

Omnichannel Multilingual Chatbots need native NLP understanding for scalable business-use.

Conversational AI tech today is designed for English and relies on Machine Translation (MT) to deliver multilingual conversations. Translation leads to misrepresentation and loss of information in spoken language. It also corrupts the data which is a huge blunder for any Deep Learning-based system.

As a result, critical information can be overlooked, customers can be misunderstood, and business-value can be lost in translation. Needless to say, it can lead to irreparable damage to a brand’s reputation.

A quintessential measure of the robustness of the underlying language technology is how the bot handles non-English inputs. If it does resort to machine translation for delivering multilingual conversations, it is indicative that the solution is not viable.

The right approach towards delivering multilingual conversations is to have an NLU that is built on top of a native NLP stack. Native NLP stacks are built ground-up on individual language models to once and for all eliminate the need for a translation that proves to be lossy.

4. Does the chatbot company offer an intuitive human-in-the-loop AI?

Human-handoff is key for holistic customer service automation.

Conversational agents are known to solve 80% of the customer-service queries. However, when there is a need for a human touch, can the system route to the support agent?

Further, can it retain enough information to contextually route the conversation to the right department, and empower the support agent with information on his fingertips?

It is important to forward the context of the conversation to allow the agents to interject and the conversation without disrupting the flow or wasting too much time of the customer.

Otherwise, we’ve relapsed to making the same mistake of having our customers repeat their issues over and over again.

Having to repeat info, and being transferred from agent to agent, is in fact on of the biggest red flag support can get from their customer.

5. Compliance to Security, and Scale

Conversational AI must not only conform with security standards but also ensure scale.

Security is a key concern for enterprises and it includes security of data and transactions that a chatbot has with users.

The solution’s compliance with security standards like GDPR, HIPPA, SOC etc, are considered table-stakes for CXOs to make their decision. Other important security practices must involve encryption, authentication and authorization measures.

Logs and any other sensitive data must be stored and maintained from all system interactions across different levels.

Key Take-aways:

  1. The path towards Digital Transformation using omnichannel multilingual Conversational AI requires a critical assessment of business needs and technological capabilities.
  2. Conversational AI is not a technology to supplant human agents but to work in coalition to provide better customer experience and business bottom-line and transform the methods of value-exchange.
  3. Security is a key concern for enterprises and it includes security of data and transactions that a chatbot has with users.
  4. AI Chatbots in the market are built on platforms that are essentially built for English and resort to translation for offering multilingual bots. Business value can be lost in translation, and brands’ reputation can suffer.
  5. Conversational AI elements are in silos, and enterprises need an end-to-end platform or service to allow manage the entire life-cycle of a bot, pre and post-deployment.



Adarsh Srivastava

Shaping category design and marketing for differentiated B2B SaaS