AI Chatbots - Redefining Business Processes

Yes, chatbots offer multilingual capability, however, it differs between NLP and guided bots. For guided chatbots, there is no issue in deploying the bot in any language as you can create the flows in as many languages as you want. NLP chatbots, however, have a critical dependence on public datasets and a lot of third-party tools in that particular language. It is not impossible to create an NLP bot in any language you desire; it only depends on how much data you have while creating the bot.

Reducing the error rate in a chatbot conversation requires you to be correctly recording all relevant metrics. The bot must be able to store all the relevant information: when a user drops off, when an issue is taken to completion, when a user wishes to transfer to a live agent, etc. Once you have all of this data, you can see where a flow breaks off and why, and fix that flow, thereby reducing the error rate in that chatbot.

While doing a disposition analysis you will find that roughly 20-30% of all the issues that a user has do not fall into any particular disposition bucket. These issues require a human touch, and therefore cannot be handled by a chatbot. Assuming you have a sizeable amount of data and a well-designed bot, you can handle around 70-80% of your chat volumes.

Data security in chatbots depends on the type of chatbot you have and what type of integrations and functionalities that the bot will allow. For example, if you allow login through the bot, you need to ensure that there is two-factor authentication, encrypted communication between the login systems, etc. From a GDPR perspective, ensure that you are entirely transparent about what data you will be storing. Besides, you should check your chatbot’s logs to ensure that you are not recording any data that isn’t allowed.

Chatbots definitely offer metrics and reports that will allow them to learn and even reorganize themselves as often as you need. This all depends on whether you have designed a chatbot that is recording all of the relevant information and an active learning platform that is intelligently reorganizing the chatbots’ options, as often as your business requirements dictate.

This would depend on the complexity and the type of freedom of conversation you wish to give to your end-user. Building a chatbot itself could take anywhere between a few weeks and a few months, but the period used to train the bot depends on how complex the conversations will be within the bot.

Both buying and building have their pitfalls. Buying a standard and an easy-to-deploy bot will not provide you with the end-to-end automation value that a customized chatbot can bring, as it will not integrate with all the back end systems you have. But building a fresh chatbot can be expensive and time taking, and has a poor ROI. We recommend using a framework as a jumping-off point, and pulling in all the APIs you need into that bot so that it brings real value.

The metrics that measure chatbot success are business metrics that you took into account while creating the chatbot. For example, if you wanted a chatbot that can help elevate your customer service experience, you should look at your Customer Satisfaction Score before and after deploying the chatbot and not any specific metrics such as the number of clicks or broken flows that the chatbot may offer.

The rapid prototyping framework/bot, can pick up flows in any format like CSV, Excel, flowcharts, etc. and can take these flows directly from systems and applications like Lucidchart as well. Lucidchart has available APIs that can be used to integrate with the bot directly.

Our chatbot solution has the ability to integrate with any external API. Just like with any other application, API integration is done through an authentication token and any data/parameters that need to be sent to the API will be collected from the user session or as user input.

The NLP engines that we use and integrate with, have the ability to interpret most dialects of the given language. The training phase for the bot will include various utterances and slang that could be encountered, which can address the variations in user questions.

As per market research analysis done by Forbes, over 80% of the customers would like to take matters into their own hands, provided the experience of the bot is rich, relevant and addresses their issues quickly. We follow certain guidelines and design principles to ensure that the interaction with the chatbot is a great experience for the customer. For example, no customer should have to go through more than 5 clicks in a guided bot flow to resolve their issue or hand off to a live agent.
We also recommend a transition phase, where customers who prefer to talk to live agents will still have the option to call customer care directly and those who do not wish to be in the waiting line can switch to a bot. This allows customers to ease into a new medium of interaction and issue resolution.

Our solution is built on a microservices based architecture and the frond-end/UI can be easily customized to meet your requirements and brand guidelines.

The voice bot handles multiple intents similar to how an NLP bot would. We would define the flows, train each of the questions and the corresponding intents. Based on the user voice input, the bot would convert the input to text, map it to one of the trained intents and output the pre-defined respond through the bot’s voice.

A. We track both direct customer feedback and scoring on the chatbot experience and relevance of the response as well as observed behavior like customer drop off rate and transfer to live agents.
Some of the metrics we tracks to determine the success of the chatbots interactions are:
a. Number of users pre and post chatbot deployment
b. Actions performed by the bot
c. Number of users who are transferring to a live agent
d. Number of users who are dropping off the chatbot flow mid-way
e. CSAT and rating of the response relevance

Yes, we can integrate with AWS, Azure and Google Cognitive Services for NLP and Speech to Text capabilities. We also evaluate and recommend which of the available Cloud and on-premise options will work best for your use case, requirements and constraints. We are tool and technology agnostic and believe in leveraging the best in class Cognitive Services that are available as the need arises. We have done similar integrations in the past for our clients.

Yes, these bots can sit on portals and CRMs like Salesforce, etc. They can be integrated with the required systems through APIs as well.

The customer agent’s training material is used for the initial training phase of the bot. We support this IT on both models – in house and maintenance – based on the requirements.

Yes, the rapid prototyping framework/bot, can be branded and customized to match the brand guidelines of a customer.

Yes, chatbots can be used and are very relevant in the education and learning space. For example, guided bots can be used to address the questions and issues faced by students, parents, and university staff. NLP bots can be leveraged to help students and teachers get answers to their questions from various resources and course material.

Yes, we can integrate chatbots with the systems and tools like JIRA and SNOW through APIs that they expose.

Guided bots can be created to address the users in any language, including Finnish. If there are any NLP or voice components in the bot, we can use models that work well on the Finnish language and further train the bot on Finnish datasets to improve accuracy.

Yes, chatbots can be put into a multitude of platforms including messaging platforms like WhatsApp, Messenger, Skype, Workchat, Microsoft Teams, etc.

Yes, we can capture both direct customer feedback and scoring on the chatbot experience and relevance of the response including the observed behavior like customer drop off rate and transfer requests to live agents.
Some of the metrics we can gather and track are:
a. Number of users pre and post chatbot deployment
b. Actions performed by the bot
c. Number of users who make transfer requests to a live agent
d. Number of users who drop off the chatbot flow mid-way
e. CSAT and rating of the response relevance

There are no language limitations when we develop guided chatbots. With NLP and Voice chatbots, some languages may not work with a high level of accuracy at the beginning. Therefore, we recommend a training period for the bots before deployment which depends on the use case and the language(s) required.

Yes, this can be done.

In companies where a certain level of AI is already implemented, we typically conduct an audit of the current state, perform a gap analysis, and recommend possible improvements and new opportunities. Specifically for chatbots, we can leverage a lot of the AI that has already been implemented and house it in an interface that can be consumed by users more conveniently to improve the adoption of UI by users. If there are certain Cognitive Services already in place, we can further leverage them in the implementation to ensure the best ROI.

Yes, this can be done.

In the beginning, we recommend a manual training phase instead of an automated one so that we have some control over what we want the chatbot to learn. Once we incorporate the automated self-learning workflow, regular cadence with evaluating the metrics along with the necessary alert mechanisms is what is essential to identify the incorrect or “bad” flows learned by the bot. Based on the learning and gaps identified, we can adjust the weights and flows for the bot.

In such use cases, a chatbot could be used internally to traverse through the repair documents or manuals. Instead of a customer/support officer needing to look through the manuals themselves, a chatbot can do it for them, thereby saving time and allowing them to respond to requests more quickly.

Chatbots are very powerful for internal uses also. Chatbots can be used as windows to automate a significant range of processes, from onboarding new employees/customers to automating the accounts payable/receivable systems among other business functions.

Yes, we utilize some open-source AI algorithms, and we have created some of these internally. The algorithm implementation depends on the use case. We also use cloud and on-premise cognitive services as need be.

Please contact us for pricing and options.

The chatbot will deal with the majority of dispositions while the live agents are offline, and then the live agents can pick up the query with the forwarded chat history when they are available. If live service agents are always required, then the chatbots will at least reduce the number of live agents that need to be staffed.

We can cater to the Lao language in a guided bot flow. Having historical data is not a prerequisite, however, it is hugely important while creating a system that is well informed and is able to deal with all the dispositions.

First, you would need to understand all the metrics, data, dispositions, and active learning model and put this all in a centralized portal. This portal should be connected with the current chat application, and be able to take in live data from the bot to enable things like reporting and re-prioritization of intents.

Definitely! In fact, chatbots SHOULD trigger any sort of downstream automation workflow. RPA is a great way to implement this automation and bring real business value.

This will depend on if it’s a guided, NLP, or hybrid approach, languages, and what back-end integrations need to be established. Please contact us for pricing.

A script would be necessary in the back-end in order to mimic what the live service agent is doing. This script would essentially be some sort of RPA tool or system integration that would trigger the same workflow that a live agent would.