Crawl, walk, run: 28+ tactics for evolving your AI agent

Crawl—Your simple Q&A agent should be answering questions and automating basic tasks

At the crawl phase, you already have or are in the process of implementing a simple Q&A agent. The best part of a simple Q&A agent is that it’s accessible to most businesses—even small and low-volume companies. Tools like customGPT, Microsoft CoPilot, ChatBase, and GPT Builder can help teams with minimal technical knowledge launch agents quickly. When done well, a simple Q&A agent can take a lot of tedious, manual tasks off your customer success (CS) team's hands and help customers get the answers they need quickly. 

How to get your Q&A agent to start answering questions

To get started, just upload a few documents and start asking questions. Customers can use the agent on your website to find answers to frequently asked questions or to guide them through searches. In fact, I've noticed that customers found more success by embedding a simple Q&A agent within a customer's existing search flow using the Voiceflow APIs.

Most AI tools have something called a Knowledge Base. This is how you upload your documents which are sourced by your AI assistant to answer user questions. It may seem like magic, but it is using a technology called Retrieval Augmented Generation (RAG). Understanding how to format your data well is the key to getting good responses.

How to add actions to your simple Q&A agent

There are loads of problems that a simple Q&A agent can solve for you. Automating tasks with a simple agent frees your CS, sales, and operations teams from menial tasks—expanding your simple AI agent has the potential to give everyone more of their valuable time back. 

Whether you're using an agent to route support tickets to the customer success team or even suggesting personalized content to users—here’s four areas we recommend focusing on to push a simple AI agent to its full potential. Remember, you don’t have to do everything on this list, but we recommend starting with customer support tasks (or those that will address high-volume requests) and moving down this list of priorities. 

1. Customer service and support

• Automating ticket submission: Generate and submit support tickets automatically based on customer issues and queries. This could include categorizing queries, prioritizing the ticket based on urgency, and routing it to the right department or support agent. (Here’s a tutorial on how to send a ticket to Zendesk).
• Collecting and processing feedback: Use AI to collect and categorize customer feedback from your website or app. Automatically create follow-up actions or leads in your customer relationship management (CRM) based on user feedback with tools like Zapier or using the API directly.

2. Sales and marketing

• Generating, qualifying, and capturing leads: Automatically create leads in a CRM from interactions on social media, email, website chatbots, and more. These can be based on specific triggers or interest shown by potential customers. Then capture basic customer information (e.g. name, email, interest) from a conversation and add it as a new lead in your CRM.
• Automating follow-ups and engagement: Send personalized follow-up messages or emails to leads or customers based on their interaction history or behavior, encouraging them to take the next step in the sales funnel. Send a templated email to customers to confirm their inquiry or action (e.g. order confirmation, appointment booking, etc.)

3. Operational efficiency

• Automating data entry tasks: Use AI to update customer records in your database or CRM—from adding new contact details to existing user profiles or updating a customer's status based on their latest interaction.
• Automating routine tasks: Automate scheduling appointments, send reminder emails, or update task statuses in project management tools through integrations with platforms like Zapier.

4. Personalization and user engagement

• Personalizing content: Suggest content, products, or services to users based on their previous interactions, queries, or preferences.

Automating event triggers: Set up AI to trigger specific actions or notifications based on user behavior or milestones (e.g. sending a congratulatory message on a user's anniversary with the service, scheduling a reminder, or booking an appointment).

Walk—Your complex Q&A agent should be leveraging customer data

If you’re in the walking stage of AI maturity, you likely have more complex queries from users that require context and information. You already have a simple Q&A agent that you want to evolve to complete more complex actions. You require integrations with existing databases, CRMs, and other third-party services to provide accurate and actionable responses from your agent. And you have the moderate technical ability on your teams to make it happen. 

This complex agent can complete simple actions ideal for mid-sized businesses or those with higher volume looking to automate a high percentage of their support queries. 

Here are three factors that set a complex agent apart from a simple one:  

1. Deep data integration: A complex assistant is adept at drawing from a wide array of data sources, including internal databases, CRM systems, and external APIs to gather all relevant customer information. That way, the agent understands the full context of each query, including past customer interactions and preferences.
2. Intelligent query routing: This assistant typically routes inquiries to the most appropriate pathway. Whether it requires specialized knowledge, access to specialized knowledge bases, or intervention by specific departments, a complex assistant ensures that each query is addressed by the most qualified entity.
3. Ability to complete advanced actions: Moving beyond basic actions, a complex assistant can perform sophisticated tasks that require multiple steps and verifications within business systems. Think multi-step account changes, or problem-solving support queries without human intervention.

‍

How to optimize your complex AI agent based on your industry

At this stage, you’ll want to tailor your approach based on your industry, since the areas of opportunity are so different. Below, we’ve honed in on software, telecommunications, retail, and financial services and shared a unique to-do list for each one. If you’re not in those industries, the use cases below still act as inspiration. Take what’s useful and leave what’s not. 

1. Software (SaaS)  

• Diagnosing and troubleshooting user problems: Use your AI assistant to pull up a customer's product usage or to provide personalized troubleshooting instructions based on their past issues logged in your CRM or product database. It can route the query to specialized support paths based on the product type or issue severity, using an internal knowledge base to guide the customer through a resolution. 
• Managing tickets: Generate detailed support tickets based on customer issues with relevant context pulled from various data sources. Use your AI agent to assign tickets to the most appropriate human support agents based on expertise or availability.
• Providing product information and updates: Your AI agent can access up-to-date information to provide answers to product questions, suggest products that meet the customer's needs, or inform about the latest updates and how to apply them.
• Managing billing and subscriptions: For questions or actions related to billing, surface specific billing details, explain charges, or modify subscription plans according to the customer's request.

2. Telecommunications

• Offering plan recommendations: In a sophisticated Q&A system, your agent can suggest the most suitable mobile or broadband plans by analyzing user consumption patterns, available plans in the database, and current promotions.
• Resolving network issues: By connecting your troubleshooting guide, network monitoring tools, and your AI agent, you can provide users with real-time solutions to connectivity issues, including personalized actions based on account information.
• Automating bill inquiries: Answer complex billing questions by integrating your assistant with billing platforms to pull up individual customer histories, and explain charges or credits in detail.

3. Retail 

• Offering product recommendations: Integrate your complex AI agent with your ecommerce platforms and customer purchase history databases to offer personalized product recommendations.
• Checking available inventory: Allow customers to inquire about product availability in real-time by integrating with inventory management systems, including alternative suggestions if products are out of stock.
• Tracking order status: Collect information on a customer's recent orders and generate a summary for them. Connect with tools like Loop Returns and Shopify to get information on a customer's recent purchases and generate a status summary.

4. Mid-sized banks and financial services

• Onboarding new digital accounts: Guide customers through the process of submitting necessary documents, verifying their identity online, and setting up their new account without needing to visit a branch. Make it seamless for users to get started with your services.
• Offering financial health checkups: Offer customers an automated financial health assessment tool that integrates with their account information and transaction history. By analyzing spending patterns, savings, investments, and financial goals, this system can provide personalized advice. Your AI assistant can answer complex questions about financial products that suit their needs, such as savings accounts with the best interest rates, credit cards with beneficial rewards, or investment products that match their risk tolerance.
• Monitoring and alerting users of fraud risk: Integrate AI with your transaction processing systems and fraud detection algorithms. When potential fraud is detected, the system can automatically alert the customer via their preferred communication channel and then guide them through a verification process to confirm or deny the flagged transaction. If fraud is confirmed, the system can initiate a dispute resolution process, temporarily lock the account, or take other predefined actions to protect the customer's assets. 

Run—Your sophisticated AI agent should be completing increasingly complex actions

We all need something to strive for. A sophisticated agent goes beyond simple API calls and is able to receive and format data, as well as complete complex actions on behalf of the user. Examples of this include executing account changes on behalf of the user, triggering specific workflow within a product, and fixing user issues or executing a purchase on their behalf.

Lately, our teams at Voiceflow have been experimenting with increasingly complex AI agents. We’re prototyping an assistant that downloads the user’s project file, reviews the code, identifies the issue, and provides a step-by-step report to fix the issue. The agent has also written code snippets for the user. We’re even exploring how the agent might execute those changes to the project itself on the Voiceflow platform. 

If you’re looking to create industry-leading AI agents, then the ideas below should be added to your product vision board. Needless to say, this kind of AI agent requires a high level of technical ability to build and iterate. 

Dynamically and contextually solving problems with your AI agent

• Dynamically offering services: Tailoring service offerings based on the customer's usage patterns, preferences, or feedback.
• Managing complex accounts: Your AI agent could handle multi-faceted requests such as merging accounts, adjusting service levels, or applying nuanced billing adjustments.
• Problem solving: Troubleshooting and solving technical or service-related issues that require an understanding of the customer's specific setup, history, and requirements.
• Automating decision making: Equipping your agent to make informed decisions on behalf of the business, such as approving discounts or customizing offers based on predefined criteria and customer data.
• Personalizing recommendations: Generating tailored advice or suggestions based on a complex analysis of the customer's past interactions, preferences, and available options.

• Providing high-quality, contextual responses: Consistently offer responses that are deeply informed by the customer's context. A level of service that mimics human understanding and intuition, thereby significantly enhancing customer satisfaction and engagement.

Taking your AI agent from baby steps to marathons

From simple to complex, AI agents have the potential to make our jobs easier by automating tasks and dynamically responding to customer requests. AI is evolving constantly, so our goal for this article is to help you envision how your agents could expand into the future. 

‍

This isn’t an exhaustive list of possibilities. But regardless of which use cases you add to your wishlist, it’s always good practice to expand your agents with thoughtfulness and planning. Your agent can always do more, but AI teams are still made up of humans. Keep your capacity in mind as you expand. Currently, every AI tool still requires human oversight—we don’t advise letting your AI agent run off into the sunset just yet. 

‍