How To Use Bland AI Conversational Pathways

Name: How To Use Bland AI Conversational Pathways
Uploaded: 2024-05-03T19:27:13.000Z
Duration: 1 h 44 min 50 s

How to Optimize AI Callers with Conversational Pathways

Introduction to Conversational Pathways

The session introduces the concept of conversational pathways as a key tool for enhancing AI callers, aiming to elevate their performance from good to great.

The video is structured into sections for better digestibility, starting with a restaurant reservation agent example provided by Bland AI.

Overview of the Restaurant Reservation Agent

The presentation will cover how to build and optimize an agent for a dental clinic after discussing the restaurant reservation agent.

Various software integrations will be explored, including cal.com for scheduling, Airtable for call storage and transcriptions, and make.com for email automation.

Setting Up in Bland AI

Users are guided on signing up for Bland and navigating to the conversational pathways section where they can create their first pathway.

A demonstration begins with creating a restaurant reservation pathway, highlighting its initial setup and potential complexity.

Functionality of the Reservation Agent

The first interaction asks if users want to make a reservation; based on responses, it either processes reservations or offers alternatives if no tables are available.

If more than eight guests are involved, the system triggers a live transfer to human agents; otherwise, it checks table availability through an API call.

Understanding Node Types in Pathways

Each card within the pathway serves as a node that dictates specific interactions; these nodes include default types like knowledge-based transfers.

The start node is crucial as it cannot be altered once set; this structure resembles other platforms like Voiceflow but allows unique configurations per node.

Key Insights on Scalability and Flexibility

Unlike traditional systems that rely on one large prompt, conversational pathways allow multiple smaller agents to connect seamlessly, enhancing scalability.

Understanding AI Chatbot Functionality

Enhancing User Interaction

The speaker emphasizes the importance of an enthusiastic tone in chatbot responses, suggesting that it makes interactions feel more alive and engaging.

Bot Builder Overview

Introduction to a bot builder that allows for either AI-generated responses (using models like 3.5 turbo or 4) or hardcoded text blocks for specific user inputs.

Conditional Logic in Conversations

Explanation of condition options where the bot remains on a card until certain conditions are met, enhancing control over user interactions.

Example provided: If a user requests a reservation, the bot can handle various inputs while ensuring relevant information is gathered before proceeding.

Customer Identification

Importance of identifying whether a user is an existing customer; this affects how accurately the bot can respond to inquiries about bookings or appointments.

Global Nodes and Knowledge Bases

Description of global nodes that allow access to knowledge bases from anywhere within the conversation flow, facilitating better information retrieval during interactions.

Current limitations noted regarding knowledge base size (approximately 30,000 characters), with future updates expected to include CSV and PDF formats.

Model Selection for Responses

Discussion on choosing between higher intelligence models for complex queries versus lower latency models for straightforward yes/no answers.

Variable Extraction in Conversations

Explanation of extracting call info into variables before moving to the next card; this feature helps retain important details throughout longer conversations.

Use Cases for Variable Retention

Understanding Conversational Pathways in AI

Importance of Availability and Variables

The initial focus is on capturing a person's availability, which significantly influences the conversation flow. This can be integrated into the first card using an "influency variable."

While this feature exists, it hasn't been widely utilized; however, it's beneficial to know that it can capture essential information like credit scores for financial services.

Fine-Tuning Models

Bland's dashboard allows users to fine-tune models by providing specific examples of sentences and pathways to follow based on conditions.

Fine-tuning is particularly useful for complex decision trees where nuanced changes are necessary, such as directing conversations based on credit score thresholds.

This process resembles hardcoded logic, ensuring that certain responses lead down predetermined paths.

Connecting Cards and Pathways

The second half of using Bland’s conversational pathways involves connecting different cards together to create a seamless flow in conversations.

Users can define conditions for each pathway, determining when to proceed based on user responses (e.g., appointment scheduling).

Differentiating Customer Types

An example from a dentist clinic illustrates how pathways differentiate between new customers and existing patients, guiding the conversation accordingly.

These pathways function similarly to chatbots with yes/no options leading through distinct flows tailored to customer needs.

Building Effective Conversational Pathways

Establishing a strong foundation allows users to build 90% of effective conversational pathways within the system.

The next step involves asking for reservation information while ensuring all necessary variables (date, time, number of guests) are captured before proceeding.

Importance of Conditions in Conversations

A critical point is that without proper condition settings, conversations may progress without obtaining essential information needed later (e.g., reservation details).

Understanding Higher Intelligence Models and Their Functionality

Key Concepts of Data Extraction

The discussion highlights that higher intelligence models do not require fine-tuning or extracting call information into variables for data retrieval. This indicates a more streamlined approach to data handling.

The concept of "temperature" in model responses is introduced, suggesting it influences the creativity and accuracy of the model's output during conversations.

Temperature Settings and Their Impact

A high temperature setting can lead to less accurate questions being asked by the model, as it may generate overly nuanced responses instead of straightforward ones.

It is noted that adjustments to temperature settings are typically made downwards (lowering), as a setting of 0.5 is sufficient for natural conversation flow without compromising specificity.

Static Text Limitations

Static text cannot incorporate dynamic variables such as user names, which limits personalization in interactions. Instead, using placeholders with lower temperature settings allows for tailored greetings.

Call Transfer Mechanisms

When a caller requests information about reservations, if the number exceeds eight guests, the system will redirect them through a transfer node path to another agent or service.

The transfer node allows customization of what the agent communicates before transferring the call, enhancing user experience.

Global Nodes and User Intent Recognition

Implementing global nodes enables continuous monitoring for specific user intents (e.g., wanting to speak with a human), allowing seamless transitions in conversation flow when needed.

Utilizing Webhooks for Dynamic Data Retrieval

For scenarios involving fewer than eight guests, webhooks facilitate real-time data access regarding table availability based on user input during conversations.

Webhooks send API calls containing essential reservation details like date and time; correct formatting is crucial for successful communication with external systems.

Importance of Correct Formatting in API Calls

Understanding Variable Types and API Calls in Reservation Systems

Variable Types and Their Importance

The format of variables (string, integer, boolean) is inherently emphasized by their type; no additional formatting is necessary.

The webhook card allows for the extraction of information from previous conversations without needing to repeat variable definitions across multiple cards.

Technical Aspects of API Calls

A straightforward API call can be made to a CRM, requiring the correct URL and body that includes all relevant variables discussed earlier.

Response data can be complex; it may include whether a reservation was successful and other available time slots if the desired slot is booked.

Handling Reservation Success or Failure

If a reservation is successful, the system confirms availability for the requested number of guests at the specified time.

In case of unavailability, alternative options are presented to users, such as different days or times.

Conditional Logic in Reservations

The system uses boolean logic to determine next steps based on whether a reservation was successful (true/false).

Users can set conditions for various outcomes; if certain criteria are met (e.g., reservation success), specific actions will follow.

User Experience Enhancements During Processing

Implementing speech prompts during API calls helps manage user expectations while waiting for responses from potentially slow processes.

This feature allows for conversational engagement with users while they wait for booking confirmations or alternatives.

Final Steps After Webhook Responses

Upon confirming a reservation's success, users receive confirmation details about their booking before concluding the interaction.

If no reservations are found, users are informed about unavailable slots and offered alternative dates/times instead.

How Does the Bot Handle Oddball Questions?

Understanding Global Prompts

The concept of Global Prompts is introduced, which serves as an overarching guide for conversation flow, similar to a large prompt used in previous models.

Each specific prompt is nested under this global prompt, ensuring consistency in responses related to the bot's identity and context (e.g., working for a restaurant).

The global prompt is appended to all individual prompts, maintaining coherence throughout the conversation.

Managing User Input

If a user provides an unavailable time slot, the system can loop back on itself rather than creating multiple new paths. This simplifies handling unexpected inputs.

The looping mechanism allows users to be prompted repeatedly until they provide a valid time that works within available slots.

Differences Between Looping and Conditions

A discussion arises about whether there’s a difference between using loops versus conditions; it appears both can serve similar functions in managing user queries.

Loops are particularly useful for knowledge-based questions where users may have follow-up inquiries, keeping them engaged until their questions are fully addressed.

Future Considerations

The technology discussed is still evolving; features like self-looping may change as developers learn from real-world applications.

While foundational concepts will likely remain stable, specific functionalities might not persist over time due to ongoing development.

Building Conversational Pathways: A Practical Example

Application in Real Scenarios

The focus shifts towards applying conversational pathways technology in practical settings such as voice assistants for dental clinics or local businesses needing appointment scheduling.

Demonstration of Voice Interaction

A live demonstration showcases how these conversational pathways function through an actual call scenario with a dental clinic.

Sample Conversation Breakdown

Appointment Scheduling and Automation Insights

Initial Appointment Setup

The caller provides their name and phone number, confirming the best contact for appointment reminders.

The appointment is scheduled for a regular checkup to address a toothache, with confirmation sent to the caller's email.

Acknowledgment of seamless communication during the call, highlighting effective customer service.

Importance of Interruption Threshold in Conversations

Discusses how interruption thresholds vary based on audience demographics; younger audiences may prefer quicker responses while older individuals might find it disruptive.

Emphasizes the need for balance in response timing to accommodate different speaking paces and styles.

Handling Information Capture in Automated Systems

Highlights an instance where an email was not initially captured but could be retrieved later, showcasing system flexibility.

Notes that some users prefer text confirmations over emails, indicating varying preferences in communication methods.

Benefits of Automation in Dental Practices

Automation can significantly reduce receptionist workload by handling appointment reminders efficiently.

Discusses how automating appointment confirmations can enhance operational efficiency within dental practices.

Overview of Software Integration for Call Management

Introduces three software options (Make, Airtable, IAL), suggesting a focus on Make first for managing calls and appointments.

Describes the process of using webhooks to trigger actions post-call, including recording management and data entry into Airtable.

Post-processing Call Data

Explains how each call is recorded and stored systematically for easy access later.

Understanding Post-Call Processing

Importance of Post-Call Data Handling

The module's primary function is to extract data from the call, emphasizing the need for processing outside of the call itself.

This approach minimizes workload during calls, allowing agents to focus on immediate customer needs without overloading them with tasks.

By gathering available dates and other essential information post-call, it allows flexibility if customers change their minds about appointments.

Data Collection Process

Key data collected includes appointment date, caller's name, requested service, notes, and email address when available.

The process involves straightforward JSON parsing to format the output into a structured block for further use.

Integration with Airtable

Connection to Airtable is simplified; users can add or update records easily by signing in with their Make account.

Various attempts at capturing data are logged within Airtable for review and troubleshooting purposes.

Customer Service Tracking

The system also logs customer service issues raised during calls, including complaints about staff behavior or service quality.

This feature is particularly beneficial in industries like real estate where tracking customer feedback is crucial.

Appointment Management Automation

Automations create new appointment requests in Google Calendar and send confirmation emails containing relevant details formatted using HTML.

Utilizing GPT for HTML formatting enhances email aesthetics by generating visually appealing templates based on existing designs.

Cancellation Processes and Additional Features

Managing Cancellations

To cancel an appointment, users simply locate the booking through a JSON block and update it accordingly in Google Calendar.

Advanced Functionalities

How to Create a Date Range for Availability Verification

Overview of the Process

The process involves sending a web hook with a start date variable, which is captured and formatted using an OpenAI module to create two variables representing the current date range.

The system utilizes the current date, time zone, and start date to output a range start and end, allowing quick adjustments to the availability range.

Flexibility in Date Ranges

A standard 72-hour range is implemented; this means if a customer requests an appointment on Monday but it's full, they can easily check availability for subsequent days without additional API calls.

This method streamlines scheduling by taking the earliest available date and extending it over 72 hours for better flexibility.

Integration with Cal.com

The integration with cal.com allows synchronization of multiple calendars (Google Calendar, Outlook, Notion), simplifying management of various scheduling tools.

An API call is made to cal.com using event type data along with specified time parameters to retrieve available slots efficiently.

Web Hook Response Mechanism

After retrieving available slots from cal.com via HTTP request, a web hook response sends this data back into Bland for processing.

The entire sequence takes approximately three-fourths of a second, ensuring minimal impact on latency during operations.

Conclusion and Future Directions

The discussion wraps up by inviting feedback on whether viewers want more detailed breakdown videos about complex components like post-call processing.