Comment on page
Seamless Integration of ChatGPT into Smart Replies
This service is currently in beta testing, please connect the ChatGPT app before you start using the ChatGPT integration with Smart Replies
Picky Assist's smart replies are also powered by an AI model. However, ChatGPT, being a generative AI, is far superior. Therefore, integrating with smart replies allows you to channel queries to ChatGPT, which can provide more natural and improved responses.
Picky Assist Smart Replies is a proprietary solution developed by Picky Assist that is both easy to configure and cost-effective. It operates based on a semantic indexing method, which uses the confidence score of the given text. In other words, businesses provide all possible combinations of questions and answers to Picky Assist. Then, based on the confidence score match, the system selects the best answer and sends it back to the customer. To maintain the context of the conversation, we use Intent keywords.
In contrast, ChatGPT operates on a large language model that is pre-trained with a vast dataset and is capable of understanding user input in a more natural way, generating equally natural replies. With ChatGPT, you generally don't need to create any sets of questions and answers. Instead, you need to provide a context for your process, and the system will automatically construct the best answer based on the question and the training data you've supplied.
Picky Assist's ChatGPT integration operates based on the categories in Smart Replies. You can enable or disable ChatGPT for each category, which provides better control over when and how you route customer questions to ChatGPT.
For each category, you can add potential keywords or questions that a user is likely to ask. You can then configure your ChatGPT with System Data (instructions to ChatGPT) and Assistant Data (training data).
For instance, if you've added keywords like 'pricing', 'price', 'rates', 'rate', 'discounts', 'charges', 'charge', 'billing', 'bills', and 'cost' as questions in your Smart Replies against a category, the system will route the conversation via ChatGPT whenever it detects any of the given keywords in the user's question.
To handle follow-up questions effectively where the keywords may not be found - for example, "What is the price of the iPhone 14 Pro Max?" followed by "Do you have any discounts?" - you need to use Picky Assist's Intent feature. It's also necessary to enable "Passing Message History" to ChatGPT for better response construction.
Go to Smart Replies Train page where you can see the "ChatGPT" button against each category as shown below
A popup will appear on the screen as shown below where you can configure the ChatGPT for the respective category
You can see the following options to configure.
System behavior guides ChatGPT on how it should interpret and respond to the training data. For example, you need to carefully set this as ChatGPT will consider these instructions as inputs to generate the output. Here are a few examples of system data:
You are a chatbot providing assistance for the Tours & Travel Industry. Limit the response only to the Tours & Travel Industry in the USA and generate output in a maximum of 200 characters always.
In the above instruction, we are directing ChatGPT to do the following:
- 1.Operate as a chatbot within the Tours & Travel Industry.
- 2.Limit responses to inquiries related to the tours and travel industry and restrict responses to the context of the USA.
- 3.Generate output that does not exceed 200 characters.
You are a chatbot specializing in providing pricing information for a retail electronics store. Only provide answers that relate to the price and availability of electronic products within the store. Please use polite and professional language
You are a chatbot for an online bookstore. Your primary function is to assist users in providing pricing and discount details for books in the store. Ensure your responses are accurate, up-to-date, and concise.
You are a chatbot for a health insurance company. Your role is to provide information about various insurance plans, pricing, and policy details. Keep responses concise, factual, and always remind users that final decisions should be made after consulting with a qualified insurance advisor
Remember, these instructions help ChatGPT create the most relevant and helpful responses for your specific use case.
This is where you are giving training data to ChatGPT to generate an output based on your business data, see a few sample training data examples
Using the "Attribute" feature its possible to pass the data that are saved in the Contact CRM.
Picky Assist provides 4 plans 1) Basic plan which costs $14 per month including 2000 contacts, 10000 interactions, and no access to API 2) Pro plan which costs $29 per month and includes 10000 contacts and 500000 interactions, API access. 3) Ultimate plan costs $49 per month and included 20000 contacts, and 1000000 (1 million) interactions 4) API Only plan costs $34 per month and is suitable only for developers, the API only plan doesn’t give access to any of Picky Assist product and services just able to send and receive WhatsApp messages through API & Webhook. Apart from API-only plan all plans support all features however Ultimate plan only supports connecting Multiple WhatsApp accounts and each additional channel costs $19 per month. The ultimate plan includes add-ons like Zoho CRM, WhatsMail, Facebook Lead Ad form, and Zapier, and in the Pro plan add-ons are offered as paid and the Basic plan doesn’t have access to any of the add-ons, WhatsApp Web Automation plan is not included in any of the plans and is offered as paid adon, WhatsApp Cloud API is included in all plans for free. For any doubts about the pricing send an email to [email protected] or compare all plans https://pickyassist.com/en/compare-plans
Our hotel proudly offers five unique banquet halls to cater to different events and varying guest numbers.
1 The Sapphire Hall: Our largest hall, Sapphire can comfortably accommodate up to 800 guests in a theater-style seating arrangement or 500 guests for a banquet dinner. Ideal for large gatherings, weddings, and corporate events.
2 The Emerald Hall: A mid-sized option, the Emerald Hall can host up to 500 guests theater-style or 300 guests for a banquet dinner. Perfect for medium-sized events and functions.
3 The Ruby Hall: The Ruby Hall is designed for smaller, intimate events, accommodating up to 200 guests theater-style or 150 guests for a banquet dinner.
4 The Pearl Hall: Pearl is perfect for small gatherings or business meetings, with a maximum capacity of 100 guests theater-style or 75 guests for a banquet dinner.
5 The Opal Boardroom: Designed for corporate meetings, workshops, or small private gatherings, the Opal Boardroom can comfortably accommodate up to 50 guests in a theater-style arrangement or 30 guests for a round-table discussion.
Each banquet hall comes with state-of-the-art AV equipment, Wi-Fi connectivity, and offers various seating arrangements as per your event requirements. Catering options are available with customized menus to suit your taste and budget. Please email our event coordinators at [email protected] for booking and pricing details. You can also check out our banquet halls and their amenities on our website at https://ourhotel.com/banquet-halls.
Always provide contextual, precise information in the training data. Avoid including all the processes here; instead, create different categories and save the relevant information for each one. Adding too much training data will increase the usage of ChatGPT tokens, leading to a higher cost for data completion.
This feature allows you to pass the message history to ChatGPT for more contextual information. This way, the chatbot can respond more naturally and accurately by referencing previous messages. By default, our system only passes System Data, Training Data, and User Questions to ChatGPT. This may be sufficient and relevant in many cases; however, if businesses aim for more accurate and natural answers, we recommend passing the last 10 messages to ChatGPT. We offer options to pass the last 'X' number of messages or messages from the last 24 hours, as shown below.
Please note that enabling this feature may increase your costs as it consumes more input tokens.
This feature assists you in routing the request either to the Connector or the ChatGPT inbuilt App. You can direct the request to the Connector for performing advanced automation, such as sending a reply only if the user matches a tag or any dataset configured in the Picky Assist CRM, etc. In contrast, the ChatGPT inbuilt app simply processes the request by providing an answer back to the user.
Temperature is a parameter that instructs ChatGPT on what sampling temperature to use, within a range of 0 to 2. Higher values, like 0.8, will make the output more random, while lower values, like 0.2, will make the output more focused and deterministic. See the below example for a better understanding
User: "What are your pricing plans?"
If the bot uses a low-temperature setting of 0.2 (more focused and deterministic), the response might be quite straightforward and stick to the core facts:
Bot: "We offer three plans: Basic at $20 per month, Premium at $50 per month, and Enterprise at $100 per month. All plans have different feature sets."
On the other hand, with a higher temperature setting of 0.8 (more creative and diverse), the bot might provide a more varied response:
Bot: "Our pricing plans are like a treasure trove, designed to suit everyone's needs. The Basic Plan, which is like the golden key, is available at just $20 per month. Our Premium Plan, the sparkling diamond, can be yours for $50 per month. If you're looking for the full royal treatment, our Enterprise Plan at $100 per month is the crown jewel. Each of these plans comes with its unique set of features!"
For a Messaging chatbot that needs to reply within a 250-character maximum, it might be best to use a lower temperature value such as 0.2 or 0.3.
The reason is twofold:
- 1.Deterministic and Focused: A lower temperature value makes the generated responses more deterministic and focused, ensuring that the bot sticks to providing concise and relevant information, fitting within the character limit.
- 2.Avoid Unnecessary Creativity: Higher temperature values tend to make the AI more creative and produce more varied outputs, which might cause responses to exceed the 250-character limit.
Remember, temperature settings might require a bit of tweaking and testing to optimize for your particular use case. It's a good practice to experiment and observe the responses to find the ideal setting.
If you prefer the request to be processed by Picky Assist's built-in AI, you have the option to disable ChatGPT for each category. This can be done by simply turning the status off as shown below.
Please note that when ChatGPT is enabled, the answers configured for each question in the smart replies will not be sent to the user. Instead, the system will wait for a response from ChatGPT to provide a reply.