Overview

Smart Discovery is an Intelligent Virtual Assistant (IVA) that enables businesses with fast, efficient and intelligent self-service using machine learning and other technologies. A user visiting our customer’s webpage can find Smart Discovery as a floating widget style chatbot at the bottom-right corner of their webpage. Smart Discovery can help a user fetch any information easier and faster compared to the user trying to fetch it themselves on the website. Smart Discovery is easy to understand and use. It is a simple chat interface that saves the user from spending time fetching information which is readily available somewhere on the website. It reduces user frustration, saves time and helps deliver a smooth and seamless experience.

Smart Discovery is a B2B2C type product which can be used by any businesses with a large number of customers where customer service is a tedious humane task. Smart Discovery can greatly help in reducing the amount of workforce needed, better categorize queries and route them to their corresponding internal customer support teams. Smart Discovery is known to be your elderly customers’ best friend. It requires no technical knowledge and has little to no learning curve. Elderly customers can get what they’re looking for without having to search / navigate through the website just by typing their query in the chat. Smart Discovery also understands when it is not able to cater to a query and proactively asks if the user wants their conversation transferred to a live agent.

Though Smart Discovery can be implemented in any type of industry it is mainly focused for customers in the banking and credit union space. It has multiple use cases such as;

• A user can type in what they’re looking for and get instant responses which help the user reach their objective faster and even instantly in some cases.

• A user can get all the customers’ products related information just by typing the product’s name as a query

• A user can fetch branch locations and their timings, schedule appointments and in some cases even schedule a call back as per their convenience.

• A user can file disputes, raise issues / concerns and in some cases connect with a live person via chat in the same chat window.

• Smart Discovery also integrates with some of the most popular chatbot vendors in the market providing you with an AI-powered knowledge base on top of a customer’s existing chatbot vendor.

• Smart Discovery is responsive and seamlessly works on all types of devices like smartphones, tablets and laptops / PCs. Better yet, Smart Discovery also has a PWA which is a standalone desktop application that caters users without the need to get on to the website.

Introduction

Smart Discovery is a cutting-edge AI-based chatbot widget designed specifically for the banking and credit union industry. It can easily be deployed on any bank’s or credit union’s website. A user can simply start a conversation by typing what he/she is looking for using simple natural language. Smart Discovery then understands the query and intelligently responds with the appropriate information to the user in the form of a reply in chat. Some of the features available with Smart Discovery include;

• Ability for a user to converse with the IVA using just the natural language

• Ability for a user to connect or transfer the conversation to a live person in-chat without leaving the conversation

• Ability to configure or design complex process flows based on user’s responses to a query

• Ability to add or display products in the form of cards with a carousel

• Ability to design queries that could be answered as an FAQ

• Ability to have custom Welcome messages

• Ability to integrate with other third party widgets

• Ability to track performance of Smart Discovery using Automation flags

• Ability to view various performance metrics via Dashboards

Smart Discovery is a product bundled along with Academy, APT, Advisory and Automation. All these together are required to make Smart Discovery a viable solution.

Target Audience

Credit Union Members

The primary audience for Smart Discovery would be the credit union members who visit the website seeking information or assistance with their accounts, transactions, loans, and other financial services. The IVA can provide personalized and relevant responses to their inquiries, helping them find answers quickly and efficiently.

Prospective Members

Prospective members who are interested in joining the credit union may visit the website seeking information about membership requirements, services offered, rates, and other relevant details. Smart Discovery can provide timely and accurate responses to their inquiries, helping them make informed decisions.

Credit Union Staff

Credit union staff who may need to access information or assistance related to member accounts, transactions, or other operational processes. Smart Discovery can serve as a convenient and efficient tool for staff to quickly retrieve information or perform routine tasks, freeing up their time to focus on more complex member inquiries.

Non-Technical Users

Credit union members and prospective members who may not be technically savvy but still need assistance or information from the credit union website. Smart Discovery is designed with a user-friendly interface and clear instructions to cater to non-technical users, making it easy for them to interact and obtain the information they need.

Credit Union Management

Credit union management or executives who may be interested in understanding the benefits and performance of Smart Discovery, including usage statistics, user feedback, and ROI. The various tools and functionalities available are described in more detail in further sections of the document.

Use Cases

Smart Discovery has multiple use cases. The credit unions need to have a clear and comprehensive understanding of these use cases in order to better configure the IVA and get the best out of it. Let us look at the use cases in detail;

Products and Product Carousels

The products available with a credit union can be configured as a Product type experience. Experiences which are configured this way have a cleaner look and help seek the attention of the user towards the only information required. The product use case is the most important and visually appealing use case which can be set up using Smart Discovery. The product use case is the best experience a user can have to his/her query.

The credit union can have a custom image along with the product name followed by some relevant information. A product use case can also be configured with button options which redirect the user to a particular page of choice. You can also click on the blue color arrow on the Product Carousel to browse through other or related products.

Quick Responses

An FAQ can have follow up questions which can be configured as quick responses and have separate flows based on the user’s selection. The image on the left is where an FAQ has a static response. The image on the right is where the FAQ responds with a follow up question in the form of a quick response to the user and routes to a different response based on the input given by the user.

Options / Buttons

FAQs can also have options / buttons 

Site Search

When a user types in a query in the widget, it first tries to search for the products that have been configured manually. Then, it searches for the FAQs that are configured. When it fails to fetch any result, it uses the built-in Search Engine capability to search for the keyword and return any related pages hosted on the domain that match with the user’s query. Smart Discovery uses Google Index Search, which is one of the most powerful search systems. To be able to use this feature the customers need to make sure that their pages are indexed properly for the page to be reachable to the Search.

Site Specific Responses

Smart Discovery has the ability to respond based on the context. The same query used in different pages of the customers website could mean different things. Smart Discovery understands this and responds with contextual information. This works by maintaining multiple sets of responses for the same query and triggering them depending on the site / domain from which the query was asked. Smart Discovery can also identify the location where it was launched. For example, Smart Discovery can identify when a user launches it on the customers’ Online Banking page and respond accordingly navigating them through the page.

For Example;

Query - I want to make payment on my credit card

Scenario 1: Site location - ent.com

Response: 

How would you like to make the payment?

<Online banking> <Mail Us> <Contact us>

<> denotes buttons, clicking on any one will present the next set of responses depending on the button clicked

Scenario - 2 - Site location - online.ent.com

Response: 

Sure I can help you with that.

Please go to ‘Menu’ and select ‘Payments’ from the list.

….

And other steps about credit card payment within online banking

Progressive Web Application

Smart Discovery also has a PWA version which is a desktop application that can be downloaded and used by the user like any other desktop application without the need to go to the webpage. This use case is beneficial for elderly users who are not tech savvy and are better at byhearting or remembering action sequences than at navigating through a webpage.

Transfer Conversations to Live Person

Smart Discovery also has the ability to transfer chats or conversations to a live person at the credit union’s customer support center. Currently this is only available for Glia and Live Person widgets.

Tools and Functionalities

Academy

Academy is the configuration tool where all the magic happens. In the context of Smart Discovery, a query that can be configured is called an Intent or an Experience. An experience comprises various elements (listed below), all of which are handled using Academy. The platform environment is the common base for all other instances. It comprises all standardized intents. A standardized intent is that which is generic enough to be given to any new customer as is, or used as a template over which customer specific customizations could be done. Every customer will have their own instance of Smart Discovery which means that every customer will have their own instance of Academy in which all the intents are configured and maintained. Below are the elements that you can find in Academy along with a brief description;

Trigger phrase

A trigger phrase is a word or phrase that enables us to derive the context of the query that is being asked. Each experience is mapped with multiple triggers. The idea being, whenever any of the mapped trigger phrases is triggered it implicitly starts the experience / intent that it is mapped to.

Experience / Intent

An experience or an intent is a group of responses, inputs or actions that are defined together in a particular structure or chronology. Whenever the experience is triggered, the chronology defined is followed and the responses are printed in the form of replies back to the user.

Package

A group of intents that can be maintained together are called a package. A package essentially consists of all the experiences that are related to each other or to a particular service.

Standard Package

A standard package has the least precedence of all the other packages. This would be something like the language, smalltalk (consists of natural language phrases that the IVA responds with), etc.

Knowledge

A group of packages together is called knowledge. The idea being that they can be shared or re-used across multiple customers.

APT (Agent Performance & Tagging)

As described above, APT is part of the bundle that comes with Smart Discovery. In APT, you would be able to search for conversations from the past and basically pick and go through each and every conversation. The Homepage of the APT tool looks like this;

Navigation

The button on the top-left corner is for selecting the application / tool that you want to use. The other tools you can navigate to using this button are Advisory and Automation, both of which will be explained in detail in the later sections of this document.

The button on the bottom-left corner is for logging out of the tool.

Member Info Pane

The pane on the right side with a blue member icon is called the Member Info Pane. When a conversation is selected, you can find details about the conversation in this panel.

The member pane displays the following information;

• Identifier - Explained below

• Environment - The name of the environment can be seen just below this identifier

• Name of the IVA - The name of the IVA is shown right after the environment name, separated by a ‘period’

• Language - The language of the conversation

• Start - The start date and time 

• End - The end date and time

• Time - Duration of the conversation

• Authentication Level - Only applicable for Smart Transactions

• Authentication Status - Only applicable for Smart Transactions

Searching Conversations

To search for a conversation, you have the following options;

• Identifier - An Identifier is a unique alphanumeric string comprising of three parts 

• For example: inquisitive_marlin_42

• It can be used to quickly identify a specific conversation

• Experience - The name of the experience that you are searching for. If there is no specific experience of choice, leaving this field as blank will show all the experiences together

• Date Picker - To select the date range through which you want to search conversations from

• Filters - There are three filter options;

• 

• No filter - No filter is applied (this is the default filter)

• Automated - Only the conversations that were automated

• Not Automated - Only the conversations that were not-automated

• For any and every combination of inputs selected for the Identifier, Experience and Duration;

• No. of results for ‘No Filter’ = No. of results for ‘Automated’ + No. of results for ‘Not Automated’

• We will look more on what Automated & Not Automated means as we go further in this document

Advisory (Insight Driven Decisions)

Advisory is the data visualization part of Smart Discovery. This is where the customer can track their IVA’s conversation volume, visitors traffic, their top products, experiences and dropoffs. These metrics help achieve insight driven decisions.

Filters

The ‘Time Range’ filter can be used to change the date range and even create a custom range by selecting the ‘Range Type’ dropdown. The ranges available by default are;

• Last Day

• Last Week

• Last Month

• Last Quarter

• Last Year

Refer the screenshot below;

Conversation Volume

This page helps visualize the following metrics;

• Conversations by Hour - A graph representing the number of conversations that started during each hour of the day

• New Visitors - Number of new visitors in the selected time range

• Returning Visitors - Number of returning visitors in the selected time range

• Total Visitors - Total number of visitors in the selected time range

• Total Conversations - Total number of conversations started in selected time range

Top Knowledge Usage

This page helps visualize the following metrics;

• Top Products - The top 20 most triggered products during conversations

• Top Experiences - The top 20 most triggered experiences during conversations

• Drop Offs - Top 20 most drop offs during an experience

Options

Clicking the ‘three dots’ gives various options such as;

• Force Refresh - Refreshes the data on the go

• View chart in Explore - Gives capability to run custom queries and analyze on go

• View Query - Can be used to introspect the SQL query being used

• Copy Chart URL - Can be used to copy the chart’s URL 

• Share Chart by Email - Can be used to share the graph via email

• Maximize Chart - Zoomed In view for better visualization

• Download as Image - Can be used to download the graph as an image

• Export CSV - Can be used to download the graph as CSV file

Automation

The Automation report shows the Automation percentage as an actionable insight. The Automation percentage is the percentage of conversations where at least a single experience is successfully triggered out of the total number of conversations and where the query is resolved without the need for human intervention. When a conversation is transferred to a live person it is considered as Not Automated. The Automation page looks like below;

Filters

The ‘Time Range’ filter can be used to change the date range and even create a custom range by selecting the ‘Range Type’ dropdown. The ranges available by default are;

• Current Month

• Previous Month

• Previous Quarter

• Current Week

• Previous Week

• Last 4 Weeks

• Last 7 Days

• Custom Date Range

Options

On the top left, beside the Time Range filter the following options are available;

• Refresh - Refreshes the Automation % on the go

• Download - Can be used to download as a CSV file

• Print - Can be used to print as a PDF

Adding Automation Flags in Academy

Defining the Automation Flag

To define the Automation flag in a new environment, follow these steps;

1. Logon to Academy for the environment in which you want to define the Automation flag

2. Go to the generic_services package under knowledge

3. Define a code named analytics_tag_experience_automated in this package

4. Define the value as $.set('tag.automated', true)

Capturing an ‘Automated’ Flag in an Experience

To capture the ‘Automated’ flag for an experience, follow these steps;

1. Go to an experience for which you want to add the Automation flag

2. Go to the last response (the ‘Text’ block) in this experience where the response is equivalent to successfully helping with a user’s query

3. For this last response, click on ‘Action Type’ under the options provided for ‘Add next response’

4. Search for the generic_services package under the ‘Attach a response, code or experience’ dropdown menu

5. Under the generic_services package, select ‘Code’

6. Under the ‘Code’ menu select the new analytics_tag_experience_automated Automation flag that we’ve just defined earlier

7. Once selected, define the ‘Action Data’ as {}

8. Click ‘Save’

Once the steps mentioned above are defined, whenever that particular response (where the action response was added) in the experience is triggered, the experience will be marked as ‘Automated’.

Capturing a ‘Not-Automated’ Flag in an Experience

To capture an experience as ‘Not-Automated’, the experience simply needs to be as is without making any changes to the workflow. The experience when triggered is automatically flagged as ‘Not-Automated’.

For example, in the card_not_working_start experience, the responses are as follows;

1. Text - We’re sorry that you’re having trouble with your card

2. Text - Let me connect you with one of our chat gurus

3. Action Type - client_transfer_live_agent_operator

In the above example, the chat is being forwarded to a live agent as per the experience workflow. All such conversations are to be marked as ‘Not-Automated’ as the response is not equivalent to successfully helping with a user’s query. To capture an experience as ‘Not-Automated’ just don’t add the Automation flag for such experiences (leave the workflow as is without making any changes). They will implicitly be marked as ‘Not-Automated’.

Finding the right Experience/Response to add a flag

An experience can have multiple linked responses, not all the responses will necessarily be needing the Automation flag. The Automation flag must only be added to the responses where we can say that the response is equivalent to successfully helping with a user’s query.

For example, in the account_number experience, the linked responses are as follows;

1. account_number_mode_get

2. account_number_mode_set_check

3. account_number_mode_set_digital

4. account_number_start

The automation flag must not be added for all the 4 linked responses that you see above. It must be added only for end-responses. The above experience starts with account_number_start followed by account_number_mode_get. The end response for this experience will be based on the input captured from account_number_mode_get.

Therefore, account_number_mode_set_check and account_number_mode_set_digital are the end-responses for this experience. Hence, the ‘Automated’ flags must be captured at the end of both account_number_mode_set_check as well as account_number_mode_set_digital

Experiences with Multiple Flows

Some of the experiences could have multiple branches where one of the branches would mark the experience as ‘Automated’ whereas the other would mark it as ‘Not-Automated’. 

For example, an experience could have 4 different responses wherein one of the responses returns some information to the user whereas another responds by connecting the conversation to a live agent. In this example, the former response would be marked as ‘Automated’ while the latter would be marked as ‘Not-Automated’.

Basically, all the scenarios where the response connects / transfers the chat to a live agent must be marked as ‘Not-Automated’, meaning that there will be no automation flag captured for such a response which will be implicitly marked as ‘Not-Automated’. However, in the other response in which information is shared with the user, the automation flag must be used and must be captured as ‘Automated’.

Identifying Where to Add a Flag using Workflows

One simpler way to identify where the ‘Automated’ flag could be added for an experience is by looking at the experience’s paper workflow (the workflows that we have defined in excel sheets). In the paper workflows, wherever the response is marked as a ‘Text’ type with response tags ending with the suffixes ‘_link>’ and ‘_info>’ in blue, are the responses where you must add the ‘Automated’ flag in Academy.

Integration with Third Party Widgets

Smart Discovery seamlessly integrates with some of the leading vendors in the market that provide widgets. If a credit union already has a third party widget for live chat, Smart Discovery can be added as a preliminary layer to this, filtering out all the conversations that can be automated by directly serving them with the information they need without the need for them to connect with a live person. Only conversations in which no experience is triggered as well as experiences in which transferring to a live person is specifically configured as part of an Action type response. Some of the vendors Smart Discovery currently integrates with are;

Glia

Glia is a leading customer engagement platform that offers a suite of communication and collaboration tools, including an AI-powered chatbot, to help businesses provide exceptional customer service and support. Glia's chatbot is designed to interact with customers in a conversational manner, leveraging machine learning and natural language processing (NLP) technologies to understand and respond to customer inquiries and requests in real-time.

Live Person

LivePerson is a customer engagement platform that offers a range of tools and features to help businesses provide exceptional customer service and support. Some of the best features of LivePerson's chatbot product include its advanced natural language processing (NLP) capabilities, seamless omnichannel support, and intelligent routing. The chatbot is designed to understand and respond to customer inquiries in a conversational manner, simulating human-like interactions. It can be deployed across various digital channels, such as websites, mobile apps, and messaging applications, providing a consistent customer experience. The chatbot also has the ability to intelligently route inquiries to the appropriate human agent or department, ensuring that customers are connected to the right resource for assistance. Additionally, LivePerson's chatbot product offers customization and branding options, allowing businesses to tailor the chatbot's appearance and tone of voice to match their branding, providing a personalized experience for customers. The product documentation for LivePerson's chatbot would provide comprehensive instructions and best practices to help businesses successfully implement and optimize the chatbot for their customer service and support processes.

Eltropy

Eltropy is a leading communication platform that provides businesses in the financial industry with tools to engage with customers through various channels, including an AI-powered chatbot. Some of the best features of Eltropy's chatbot product include its robust automation capabilities, personalized interactions, and industry-specific focus. The chatbot is designed to automate repetitive tasks, such as answering frequently asked questions and providing account information, saving time for both customers and agents. It also delivers personalized interactions by leveraging customer data and preferences to tailor responses and recommendations. Eltropy's chatbot product is specifically tailored for the financial industry, including banking and credit unions, with industry-specific language, compliance features, and integrations with financial systems. The product documentation for Eltropy's chatbot would provide detailed instructions and best practices for businesses to successfully deploy and utilize the chatbot, enabling them to provide efficient and personalized customer engagement in the financial space.

Widget & User Interface

The user interface of a credit union's chatbot assistant can be described as a seamless and intuitive platform that enables customers to interact with the chatbot in a user-friendly manner. The chatbot interface may be accessible through various digital channels, such as the credit union's website, or messaging applications (PWA), providing a consistent user experience across devices by dynamically adjusting to every screen size. The interface features a chat window or chatbox where customers can type or speak (only available in PWA) their inquiries and requests in a conversational manner. The chatbot may respond with text, images, buttons, or other interactive elements, depending on the capabilities of the chatbot. The interface may have a personalized touch, displaying the customer's IVA name of choice, for a unique customer experience. It may also provide options for customers to escalate to a live agent or access additional support resources. Overall, the user interface is designed with simplicity, ease of use, and customer-centricity in mind, providing a convenient and efficient channel for customers to interact with the credit union and obtain the information and assistance they need.

If the customer is using a third party widget, the position and appearance of the chatbot would differ. The widget could have a different style, behavior or could even come up later as a floating icon. If the widget is a third party widget, Smart Discovery would not look like a distinguishable product but will merely look like an integrated service within the same or existing widget. In such cases, Smart Discovery only awakens upon triggering an intent or for fallback responses, which completely will depend on the way the customer configures Smart Discovery on their webpage.

For any customer using the Interface’s Smart Discovery widget, to use Smart Discovery a user simply needs to logon to the user’s credit union page and lookout for the floating widget icon at the bottom-right of the screen. The widget will pop-out prompting the user of its presence as soon as the page finishes loading. Below is an example of a customer’s instance;

After a while, the pop-out prompt is minimized with just the icon. Hovering over the icon looks like this;

In this example, the name of the customer’s IVA is Carl. You can click ‘Open Chat’ or on the chatbot icon to start a conversation.

Before you can start a conversation, as a user, it is mandatory to give consent for recording the conversation. The conversations are recorded for quality and training purposes, on which you will learn more about, in the next section.

To start the conversation, click on ‘Accept’. After clicking on ‘Accept’, you will be shown the customer’s customized welcome message. Each customer will have their own version of the welcome message.

You can start by typing your query in the input box which says ‘Please enter your question here’ or by selecting the option buttons which are available;

Every customer can choose what options they want to display at the beginning of a conversation. These welcome options could usually be the most visited intents or experiences for that customer.

For example, let us say I’m one of the credit union’s customers and I want to know the routing number for which I usually have to search through the credit union’s website or have to call the credit union’s customer service. However, if my credit union has Smart Discovery in place, I would rather get this information much quickly by just performing the following steps;

• Go to the credit union’s website

• Open Smart Discovery from the bottom-right corner

• Type in ‘What is the routing number?’

• Get the response within a second without having to wait

Within just a second, I have the information that I was looking for. Using Smart Discovery is such a simple and straightforward process that takes no getting used to.

You can continue the conversation even further by asking Smart Discovery a new question or by ending the conversation. Some experiences also have follow up questions and nested flows based on the users input.

Release & Maintenance

Smart Discovery also comes with maintenance tools like our Support Portal which can be used by the customer to log issues and change requests. An employee from the credit union can logon to the Support Portal via their user accounts and log / create tickets. They can attach screenshots of the issues or even describe the change or feature they are looking for (if any). As per the Service Level Agreement, Interfac.ai responds based on the priority and severity of the issue reported. Once a fix / change is developed, it is usually deployed during the scheduled downtime windows which are agreed with the customer.

Knowledge Update Services

Knowledge Update Services which is also abbreviated as KUS is a framework designed to improve the product release cycle and customer experience. It is available only to customers who have chosen Managed Service. KUS helps with easing project management by bringing multiple customers into a single basket which will be catered with all the latest product releases. While customers who are not yet induced into the KUS program could be on different versions with not all of them having the set of all the latest features or enhancements developed. Below are the kind of activities that will be done via the KUS program;

1. Improving existing knowledge

• Adding new ways people ask existing questions

• Optimizing existing workflows / experiences to perform better

2. Adding new knowledge

• Adding new workflows / experiences

• Extending the existing integration to support new workflows

• Implementing new integration to support new workflows

3. Any new product developments / enhancements

• Backend / frontend changes

• Widget improvements

• NLP / parser updates

• Improvements to the tools (APT, Advisory & Automation)

Release Cycle

The KUS program comprises two minor cycles and one major cycle. The current release cycle planned for the KUS program for Smart Discovery includes the following activities;

Minor Cycle - 6 weeks

The Minor cycle will consist of only New Intents, Inscopes and could sometimes even include bug fixes. A single major cycle will take place by the time two minor cycles are completed. The minor cycles will be represented as ‘x.a’ and ‘x.b’, with ‘x’ being the release number.

• Data Analysis - 1 week - The data analysis done on the tagging reports for each customer which is required for collating New Intents to be created along with the Inscopes that are to be trained.

• Artifacts - 1 week - Creation of artifacts and collecting the required inputs from customer resulting to the Artifact completion

• Configurations & Testing - 2 weeks - Getting the new intents configured and tested in the platform environment. The Customer's environment will be derived from this platform environment.

• UAT Release & QA - 1 week - Planning for UAT release and sanity testing. Must be deployed only during the Run & Maintenance window.

• Production Release & QA - 1 week - Planning for production release and sanity testing. Must be deployed only during the Run & Maintenance window.

Major Cycle - 12 weeks

The Major cycle will comprise all the new features, enhancements, platform level changes, product improvements and so on. 

• Analysis & Planning - 2 weeks - Analysis of backlog items and prioritization of features that are to be implemented during the release

• Specifications and PRD Creation - 1 week - Getting in touch with the BAs and gathering requirements for features coming from customer feedback. Documenting those requirements in a PRD with clear instructions and process flow diagrams

• PRD Review & Overview to Engineering - 1 week - Internal review with Product team and then present to the Engineering team. Clarify doubts and changes to PRD if required

• Development & Quality Assurance - 4 weeks - Engineering team to get the features developed and ready for QA in development instances. QA to test and give a go-ahead to move items to UAT

• UAT Release & QA - 2 weeks - QA to test changes in UAT and revert to developers as needed and give a go-ahead to move items to Production

• Production Release & QA - 2 weeks - QA to test changes in Production and revert to developers as needed. BAs to intimate the customers after the release

The KUS programme will be run and managed as a joint effort between the Product, Project Management and Engineering teams. Although the programme is handled by the Project Management team, there are multiple teams that they’ll need to work with. Such as;

• Product Team - For planning deliverables for each release and set priorities

• Business Analysts - To send out Artifacts and get the required inputs from the customer

• Configuration Team - To get the new Intents configured on platform environment

• QA Team - To get the intents tested as per the requirements

• Engineering - To get the backend configurations done whenever required, as well as for development of new features for the major cycle

• Delivery Team - To plan and release deployments during the appropriate maintenance windows

New Intent Addition

CEA offline team will identify the New Intents from the offline tagging which will be done on a daily basis and these new intents will be added to the new intent sheets every day for the respective clients at the end of the month they do a Pareto Analysis and Identify the Top 5 Intents per client for that particular month. This data will be shared to the product team by a respective SPOC from the CEA Offline team.

The CEA team will collate all the data from the CEA working sheets and after performing checks and analysis(like Pareto), they will feed the data to ‘SD New intent & Inscope sheet - New Intents Identified tab’ of the month. The CEA team has to notify the product team once the data is filled in.

After the New Intents are submitted to the product, product(Smitha) will perform a quality check on the data.

Quality check will primarily be focused on accuracy of data. There will be spot checks on if new intents are generated properly. Order and importance of new intents should match the raw data, also new intent queries should be matched correctly against user intent, query category, sub category, and count of queries observed.

After all these Quality Checks, the product will pass or fail the report. If it fails, it’ll go back to the CEA team for improvement and further actions. If passed, the product will then work on the next step.

Quality check criteria:

• Queries should be mapped correctly against user intent, query category and sub category.

• Count of the queries should be accurate

• To ensure that queries reported are not random but top 5, as observed in raw data based on count.

• Queries are not mapped forcefully against the user intent list, if a query does not have a match in user intent list, a new entry has to be added and then has to map against it.

Inscopes

A query which is a miss, but was supposed to be in the scope of the intents configured comes under the Inscope training. Any intent / experience that was supposed to be triggered during a conversation must ideally be working if the proper trigger phrases are defined. When an intent is a miss and comes under the purview for an inscope, the query is added as part of the trigger phrase. After this activity is done, the same query would be a match and will now trigger the required experience. Inscope activity currently happens on a weekly basis.

Live Tagging

While a user initiates a conversation, and when no intent is triggered for a user’s query, the site search kicks in. If the site search doesn’t match any results, then the conversation could be forwarded to a Live Tagging team where a notification is sent in runtime and a member from the Live Tagging team would deliberately look at the user’s query and proactively trigger the experience which matches the user’s query. All such conversations handled by the Live Tagging team will also be used as input for CEA teams’ analysis for the New Intents and Inscopes for that particular period. Live Tagging is currently only used for Glia widgets.

Implementation

To implement Smart Discovery for a new customer there are a series of steps that need to be followed once the agreement is made with the customer. The first step is to set the scope of Smart Discovery for the customer. This happens by collecting the use-cases, products to be configured, intents that IVA must respond to, name of the IVA, color theme and various other requirements using an Artifacts document. This artifacts document is shared with the customer in the form of an excel sheet where the customer responds to various questions and fills out the required fields that are required for a smooth and successful implementation.

Artifacts

Artifacts are used to collect information from the customer in order to configure the intents on Academy. We collect information about the credit unions’ products, details like the routing number, what name they’d like to have for the IVA and so on. The format of the Smart Discovery artifacts document looks like SD Artefacts - Master template.

The document essentially covers the following categories to capture the following aspects;

• General Information - The IVA logo, IVA banner, Time Zone, Introduction line, Color code for UI and the logo

• System Information - Live chat integration (if any or planning to), Widget preference, URLs for UAT and Production, PWA preference

• Products - Product categories, Product names, descriptions, synonyms, image, buttons to be configured, If customer uses a third party widget we support then we capture only those product features that can be configured

• Product Properties - Mapping properties to products along with value types and values

• System Flows - Welcome message, Default fallback experiences, ATM & Branch Locations (API & URL versions)

• FAQs and Informational Experiences - List of informational experiences to start with

• PWA Requirements - Customer logo and IVA logo

• Widget Integration - Details regarding the third party widget integration

• Blockers & Suggestions - Capturing roadblocks and suggestions for implementation

Artifacts collection and completion usually takes about 2 to 3 weeks and is crucial for starting the implementation process for any new customer.

New Instance

Implementation Steps

Once a customer signs up for Smart Discovery the Implementation team gets in touch with the customer and captures the required information by sharing the Artifacts document. Once the information is captured, the Implementation consultant needs to follow the below steps;

1. Get in touch with the development team to create a new instance for the customer along with separate instances for Dev, UAT and Production

2. Get in touch with the Product team before promising any timelines on new feature requests or enhancements

3. Get the required backend configurations and setup done to get started with the setup

4. Implement the IVA logo, IVA banner, Introduction line, Color code for UI and the logo

5. Collate the required accesses and URLs to apps and landing pages

6. Get the tools in place (APT, Advisory and Automation)

7. Get the configuration team to start configurations on Academy-Dev Instance;

1. Import the system packages from shared knowledge base, other language and essential packages

2. Get the backend configuration (if any) from team to start with experience configuration

3. Get in touch with the Configuration team to start configuring Products and their properties

4. Get configurations done for FAQs and informational experiences along with adding Automation flags

5. Get the synonyms defined and starred

6. Add the list of common triggers and tokens for an experience

8. Get the configurations tested on Dev and later moved to UAT and further perform internal UAT testing

9. Once UAT testing passes, then get the customer to test it on UAT for any final changes or suggestions

10. Plan and schedule for Go-Live along with the tools (APT, Advisory and Automation)

The usual timelines for getting the configurations done is about 1 week from the time of receiving the completed Artifacts document. This is followed by another week's time for Quality Assurance along with another one to two weeks for UAT release and feedback from the customer.

Academy Configurations - FAQs

There are a series of configurations and steps to be followed while creating experiences for a new instance;

1. All experiences are to be created under the services_banking_client package

2. The experience name must contain the activity achieved by the experience in simple words using ‘_’ as the separator. For example, routing_number, download_mobile_app etc.

3. RIght after a new experience is created using the above convention, you must also create a synonym with the same name in plain language without any identifiers. For example, ‘Routing Number’ for routing_number and ‘Download Mobile App’ for download_mobile_app. 

4. This synonym must be star marked to set it as the default identifier for that experience.

5. Define another synonym including all the synonyms / related terms which mean the same using ‘ ‘ as the identifier. By defining all the synonyms here, the system helps tag this experience as a suggestion whenever the user’s query has one of the identifiers. For example, For Routing Number we usually create a second synonym with identifiers such as; 

1. ABA ACH IBAN Transit Routing Bank Identification Swift Number Code rout routing num number

6. Next is to set up the Activity. Every experience would have a ‘Start’ activity. In this activity we would add all the trigger phrases in a tokenized manner. For example for routing_number the Start activity would have trigger phrases such as;

1. @token.enquire @token.routing_number

2. @token.routing @token.number

3. @token.enquire_want @token.routing_number

4. @token.enquire_what @token.routing_number

5. What is your routing number?

7. The regular query is added as-is and star marked as the default trigger. The In-scope training also needs to happen in the trigger phrases under the same Start activity and will be added using an ‘@token’

8. Next is to configure the responses. The first response to be configured is the _start. The name of the response must be the experience name followed by the _start identifier. 

9. You can create multiple groups of responses, each group with different types of responses. If the customer signed up for a PWA, then the first response must have the identifier ‘voice’ as the first word of the sentence, followed by the actual response in the next line. This is the default convention when defining a response for PWA.

10. Import generic_services package from the platform environment and make the required customizations to the experiences. A customer could have a personal as well as a business type across all / some of the services they offer. For example, personal_account, personal_banking_loans, personal_credit_cards, further_assistance, business_account, business_banking_loans and business_credit_cards etc.

11.  If the customer uses a third party widget, for example glia, you must also have a separate client and client_menu packages.

12. The client package will have all the customer specific properties which will be required for various purposes. For example the client_contact_livechat_hours_specifier is a property which is required for live chat transfers (if the customer supports live chat). The value for the property must be entered in a specific format. For example, M-U 09:00-17:00

Academy Configuration Guide

Smart Discovery Standardization

Standardization is the activity of bringing the knowledge acquired from working with multiple customers over time to come up with a generic use-case instance for Smart Discovery which could be used as the baseline version or the standard version for any new upcoming customer.

Some of the goals for Smart Discovery Standardization are as follows;

• Designing and implementing a standardized platform environment for SD

• Creating a super set of experiences with standardized workflows

• Creating a super set of Glia configurations which are standardized and configurable

• Design and requirements of Academy features for handling a standardized platform

• Phase wise plan for rollout of concluded features

Intent Standardization

• List out all the experiences for all clients (The SD - Experience set only has a few, not all)

• Go through each env/client's Academy

• Look into the experiences defined in the 'services_banking_client' package

• Investigate missing experiences from the 'SD - Experience set' sheet vs client's Academy experiences and add them to the sheet (Live tagging team should be able to help us with an updated list, check with them)

• Come up with titles for experiences based on what the experience is used for

• CS offline team (Configuration team) or QA team should have a list of all the experiences defined for a client. They can help us with the experiences list with titles

• Add categories to the final list of experiences

• Pick a category of experience and look up all the experience flows defined for each client individually

• Dive into the env for each client and try to trigger that experience

• Observe which client has the best flow OR come up with a best flow based on the analysis

• The QA or Config team should be having the flows for each client's experiences. Reviewing the flow there will be easier than having to go through each clients' landing page for analysis.

• Come up with the best experience flow for that experience

Standardized Intent Set

The standardized intent set is a union of all the most commonly triggered or used intents across all the customers. The first version of the standardized set can be found in the SD - Intent Standardization sheet.

The list of standardized intents include;

1. Apply Membership

2. Find Member Number

3. Find Account Number

4. Money Transfer

5. Account Balance

6. Account Closure

7. Customer Support

8. Routing Number

9. Loan Payoff

10. Working Hours

11. Register for Online Banking

12. Online Banking Login Issue

13. Rates and Fee

14. Reset Password

15. Find an ATM or branch

16. Order or Reorder a card

17. Digital Wallet

18. Digital Banking

19. Tax Documents

20. Dispute Transactions

21. Report Fraud

22. Skip a Loan Payment

Standardization of Platform Environment

Customizable Workflows

In Academy, a user must be able to configure a custom / a different workflow of an experience for different customers. Users must be able to select for that particular customer if the standardized version of the workflow must be used or if a custom workflow must be followed for that particular experience. When selecting for a custom workflow, the user should be able to use the standardized workflow as a preset which can be editable.

Properties

A property could be anything like response options defined for an input, info response for an input, link response for an input etc. For example, when ‘Mobile App’ experience is triggered we generally respond with Android and iOS buttons and upon responding to which we show a link to that corresponding option which was selected. These options could be configured as properties. As in, ‘Mobile App’ will be the property, whereas whether the property is set as ‘Valid’ or ‘Invalid’ for a customer depends on whether the customer needs that particular experience.

Property Values

The value which is set for a said property and for a particular customer (if the property is enabled) will be the property value for that customer. A property value could be the same for two different customers but a customer cannot have two different values for the same property.

Pricing

Smart Discovery’s pricing is dependent on the number of customers for the credit union billed on a monthly basis. An example of this pricing strategy could be seen below;

The standard pricing and order terms could be like in the image above.

Limitations

Smart Discovery comes with some of the following limitations;

• It is information only, so the user cannot transact on the conversation (Possible with Smart Transactions)

• It does not have a built-in Live Chat interface that can connect you to a person from the CU within the same chat. We currently support other third party integrations for this

• It cannot understand or help with all the queries. The typical fallback would be to perform a site search or to transfer chat to a live person.

• The chat is confined to the webpage and cannot be opened / moved to a separate tab / window

• 
  

Technical Architecture

Security & Compliance

Security

Enterprise-grade security and encryption at the core with in-built systems for protecting sensitive user data and information.

Privacy

Customizable data protection framework to define the sensitiveness for different types of data such as private, public or protected.

Data Protection

Multiple methods for masking and redacting PII and other sensitive information shared with the AI assistant.

Third Party Audits

Smart Discovery is SOC2 Type 2 attested with Trust Service Criteria (TSC) - Security, Confidentiality and Availability. The SOC2 Attestation of interface.ai addresses the third-party risk concerns by assessing policies, internal controls, and procedures that directly relate to the AICPA’s Trust Services Criteria. The Trust Services Criteria (TSC) has checked interface.ai on various parameters which were successfully validated by us. This prestigious certification has affirmed that interface.ai is a secured enterprise for businesses whose regulators, auditors, compliance officers, business partners, and executives require documented standards.

Data Encryption

interface.ai as a platform helps build AI assistants with data encrypted both in transit and at rest.

Audit Logging

Enterprise grade user activity audit logging for all Apps and Platform Components.

Access and Authentication

Authorization combined with authentication, logging and access reporting eliminates security risks. Granular control to manage teams, roles and accesses.

Taking Consent from Users

Compliance is an important aspect for Smart Discovery as the users are mostly the bank’s or credit union’s customers. Though not mandatory, it is likely that users could type in sensitive information while in a conversation. Also, the conversations are later masked and used for quality improvement and training of Smart Discovery. For these reasons, Smart Discovery starts with a Consent pop-up which asks for consent from the user before starting the conversation. A user has to click on the ‘Accept’ button for us to be compliant, only after which the conversation starts. The consent pop-up looks like this;

Although the widget is placed on the credit union’s website, because it is Interface.ai which records this data, the link to the Privacy Policy redirects to Interface.ai’s privacy and policy. The user cannot start the conversation unless he/she clicks on the ‘Accept’ button. Hence, every recorded conversation is a conversation for which the user has consented.