We’ve entered an era of chatbots. Interaction as a platform is here thanks to some major moves by Facebook, Microsoft, Telegram, and Salesforce, etc. Yeah yeah, we get it already. This is old news, anything new, you say.

Well, the brand new news is about how these bots are being received. The early response by the users has been lukewarm at best. They undoubtedly caught peoples’ attention, however, after some time, numerous users were not too excited by the unintelligent conversations they were having. It is where Natural Language Processing (NLP) comes to the rescue.

A quick dive into learning different aspects of chatbots

Determine availability of your bot

Working hours and channels

Should your Bot be available for 24/7?

If you’re faced with a bulky load of calls and messages in your org that can be dealt with by bots then this technique makes sense. Some companies can improve their administration levels by serving customers on these broadly deployed platforms. But typically, the harder issue — is serving the customers right and connecting them with agents directly when required.

There are two significant factors you should consider to determine the availability of your Bot.

(i) Display chatbots based on team availability

Depending on team size - not everyone needs chat flows live on their site constantly. It should have an option, that can give you the alternative choice to make chat flows that respect your team’s availability schedule. If people want their user to only interact with a bot in which they can pass off to a human immediately, set your chatbots to only display during business hours when your entire team is there for instant chatbot-to-agent handoffs.

(ii) Display chatbots around the clock

Out of the pool of problems your customers have, there are some that are best suited for a talk with a human. Although that’s not as common as ‘reset my password. Team’s time is valuable, so save them for the justified issues. Let the chatbot take care of the simpler jobs like collecting basic info etc. Salesforce gives these entities: Text, DateTime, Date, Money, Number, Person, Location, Organization, Percent, Boolean, and Object (standard Salesforce or custom). You can form your custom entities as wanted.

Better ROI

It should be noted that the ROI is straightly associated with the extent to which the bot is used in the organization. The immense advantage of the bot is that it can be scaled for a minimal cost, so the more justified use of the bot, the better the ROI.

Benefits of Einstein Chatbot in percentage

These are ways bots can benefit your business:-

Automatically resolve top customer issues:

Keeping your top clients happy with your product or service is non-negotiable. Any brand that wants to improve and hold their customer base will prefer to rely on NLP and connected CRM data. It won't just lessen the costs for the company but also help make the customer experience more authentic. To help its customers take benefit of the advances in AI, Salesforce has introduced Einstein Bots—AI-powered and CRM-connected chatbots—built natively on the Salesforce Platform.

Collect customer data:

Consistent handoff to human agents when needed - Bots are supposed to be sharpened than humans, at least when it comes to particular tasks, They are designed to do some functions very well, and also can rapidly get lost when asked complex questions, While humans can understand complicated requests and non-standard language easily so it would be better if they hand off to human agents when needed. you don’t need to forget that bots can’t solve everything yet. Customer service with some uncommon questions being asked is still performed better by a real person.

Effectively connect with business process:

Integrate bots to existing business processes to start actions by agents side - Organizations are already effectively using chatbots as an integral part of their customer contact strategy since chatbots are currently so advanced that they can successfully connect with business processes to automatically finish actions on agents' behalf.

Build a CRM-powered Einstein chatbot

Now get ready to deliver an exceptional customer service experience around the clock. you simply just need to build your bot once and then deploy it across multiple channels like SMS chat and more.

Here how it works: Begin by identifying the most frequent types of questions your customers ask. For instance, you may notice that your service agents often address inquiries related to general information, queries, and shipping issues. These are the areas where your chatbot can provide valuable support.

Service Concole > Service KPIs > Dashboard

Identify the most frequent questions your customers have.

Set Up Einstein Bot:

Now, it's time to build your personalized chatbot. Utilize the Einstein Bot Builder to create conversational flows that seamlessly integrate with your Salesforce data and utilize natural language processing to accurately interpret and address your customer's requests.

Create your custom chatbot.

Test and Deploy:

Next, it's time to take it for a spin. Einstein lets you preview your bot and make sure everything is working properly before deploying it. Over time your bot will continue to become more accurate with every conversation.

Einstein lets you preview your bot and ensure everything is working well before deploying it.

Track Performance:

And finally, Track your bot's performance with built-in analytics. Measure metrics such as response time average handle time and other key factors to make sure things are on track.

Einstein Bot Builder > Banking Bot ver.3 > Performance

Follow your bot's performance with built-in analytics.

Build a team to build your bot

The sort of a team that is required to build your bot is one that has articulated inspiring and differentiated vision. That can design experiences that are effective and delightful.

The following staff are must-haves:

Product Owner: Identifies business use case for bots, convinces partners for investment & ensures all team members are working towards predefined problem statements. These following five are the main concern of the Product owner while building a bot:-

What’s the issue and how will you overcome it?
What is the Bot task?
How will you design the bot?
How will you know if the bot is effective?
How will you launch the bot?

System Administrator: Configures einstein bot dialogs in bot builder, coordinates bots with channels, incorporates bot with external systems, versions your bots & manages.

Content Writer: One of the intriguing aspects of content writing for chatbots is the desire to move beyond writing for persuasion. The vital points while designing the conversations for your bot for advertising and communications (and some technical writing) are traditionally about conveying a brand voice and preferably soft-selling a product or service.

The AI behind a NLP based Bot:

Einstein Intent - The Einstein Intent API arranges unstructured content into user-defined labels to understand what users are attempting to say. Use the API to explore the text from emails, chats, or web forms to:

Figure out which products prospects are interested in, and send customer requests to the appropriate salesperson.
Route service cases to the right agents or departments, or offer self-service alternatives.
Be aware of customer posts to present personalized self-service in your community.

Einstein NER:

Identify the Entities in a Body of Text: A component that automatically updates fields in CRM forms. The technique is similar to word classification, as users can classify words into pre-made categories (e.g., name, organization, URL, date, or phone number).

Setting up intent for your NLP Bot

Intents are the intentions of the end-user. There is four-level of placing intent in your NLP Bots.

1) Enable Dialog Intent

(i) Identify a list of Intents that you might want the NLP model to cover.

(ii) Tip: Start with a subset (10-15 intents), then start building more intents into the bot.

Einstein Bot Builder > Jana : ve.2 > Dialogs > Transfer to agent > Dialog Details

Here, provide bot some utterance examples for the sorts of questions people may ask when they want the conversation to Transfer To Agent.

2) Add Utterances for Intent

(i) Input utterances to the intent

(ii) For best performance, each intent requires 150 utterances to train

(iii) Use of Intent Set from AppExchange

Tip: Avoid the Relatively Infrequent Term Confusion "RITZY" problem

Einstein Bot Builder > Jana : ve.2 > Dialogs > Transfer to agent > Dialog Intent

That's how the points mentioned above will appear in the Einstein Bot Builder.

3) Build AI Model

(i) Train the model after inputting the utterances for the intents

(ii) Review the Model Result under the "Model Management Page"

(iii) For Intent with low correctness (<80%), improve training data by adding more examples

Tip: Consider grouping intent that are ambiguous to each other

Einstein Bot Builder > Jana : ve.2 > Model Management

When the model finishes training, the dashboard will display a summary of the model intents and accuracy.

4) Test, Deploy & Iterate

(i) Prepare a list of utterances to test the Bots for different skills

(ii) Deploy the Bots in production for data collection

(iii) Use the data collected for iterate and improve the intent model for the Bot

Tip: Building an NLP Bot is an iterative process, plan for continuous review, and labeling of utterances gather from Bot in prod for continuous improvement.

Setting up NER for your NLP Bot

NER is a subtask of information extraction that seeks to locate and classify named entities in text into predefined categories it is essential to set up NER for NLP Bot because Chatbot NER helps you add more utilities to your Bot and ease out the process of detecting entities.

Out of the box vs custom entities

1) System Entity:

(i) Pre-trained Entities from NER Model

(ii) Predefined Entities: Object, Text

2) Custom Entity:

(i) Value Entities

(ii) Regex Entities

Roadmap: NLP Bots

Path to smarter NLP Bots,The key benefits of NLP Bots:-

More positive results through accurate interpretation.
Identify user input failures and resolve clashes utilizing statistical modeling.
Use comprehensive communication for user responses.
Learn quicker to address the development gaps.
Accomplish natural language ability through lesser training data inputs.
Capability to repurpose input training data for future learning.
Offer straightforward and fundamental corrective measures for addressing false positives.
Intent V2

Next generation of text classification that supports multiple languages and OOD

Multilingual Intent:-

Multilingual to Support Global Customers:

(i) New languages including French, Spanish, Portuguese, German, Italian, UK English are available in the summer20 Release

(ii) More languages including Chinese, Japanese are going to roll out in coming releases.

Automatic OOD Classification:-

Handle OOD Dialogue Without Training Data:

(i) Automatic classification of utterances into OOD if the intent is not defined

(ii) Common challenges in Bots to respond to customers OOD utterance appropriately

(iii) Customers do not need to provide additional training data for OOD utterances

Input Recommender:

Setup and Improve Languages Models Faster, It requires proper command with a large number of samples.

Feature: Recommends new inputs to train your NLP model from your chat transcript data

Why is that important: Reduce the work to come up with utterances for intent training

Use Cases: Setting up a new bot intent, improving the performance of an existing intent.

Get excited with Salesforce Einstein Bots for Service

Their quick start info provides you what you require to take the first steps to set up Einstein Bots in your Contact Center. For more info: Visit here

Resources and References

https://help.salesforce.com/s/articleView?id=sf.bots_service_new_bot.htm&type=5
https://help.salesforce.com/s/articleView?id=sf.bots_service_intro.htm&type=5
https://www.salesforce.com/blog/category/it/
https://www.youtube.com/watch?v=FEzUl3e7SbY
https://help.salesforce.com/s/articleView?id=sf.bots_service_model_mgmt.htm&type=5
https://help.salesforce.com/s/articleView?id=sf.bots_service_activate_bot.htm&type=5
https://help.salesforce.com/s/articleView?id=sf.bots_service_whats_an_entity.htm&type=5
https://help.salesforce.com/s/articleView?id=sf.bots_service_dialog_about.htm&type=5
https://help.salesforce.com/s/articleView?id=sf.bots_service_train_bot.htm&type=5
https://help.salesforce.com/s/articleView?id=sf.bots_service_best_practice.htm&type=5
https://help.salesforce.com/s/articleView?id=sf.bots_service_analyze_bot.htm&type=5
https://www.chatcompose.com/telegrambot.html
https://azure.microsoft.com/en-in/products/bot-services/

Hungry for more info? If you are interested in reading more of Salesforce, check out these amazing Blogs.

Salesforce Chatbots - New Frontier of Innovation