Elevate buyer expertise by means of an clever e-mail automation resolution utilizing Amazon Bedrock

Organizations spend a variety of assets, effort, and cash on working their buyer care operations to reply buyer questions and supply options. Your prospects might ask questions by means of varied channels, akin to e-mail, chat, or cellphone, and deploying a workforce to reply these queries might be useful resource intensive, time-consuming, and unproductive if the solutions to these questions are repetitive.

Though your group might need the information belongings for buyer queries and solutions, you should still wrestle to implement an automatic course of to answer to your prospects. Challenges may embrace unstructured knowledge, totally different languages, and a lack of awareness in synthetic intelligence (AI) and machine studying (ML) applied sciences.

On this put up, we present you methods to overcome such challenges by utilizing Amazon Bedrock to automate e-mail responses to buyer queries. With our resolution, you possibly can establish the intent of buyer emails and ship an automatic response if the intent matches your present information base or knowledge sources. If the intent doesn’t have a match, the e-mail goes to the help crew for a guide response.

Amazon Bedrock is a totally managed service that makes basis fashions (FMs) from main AI startups and Amazon obtainable by means of an API, so you possibly can select from a variety of FMs to search out the mannequin that’s greatest suited to your use case. Amazon Bedrock presents a serverless expertise so you will get began rapidly, privately customise FMs with your individual knowledge, and combine and deploy them into your functions utilizing AWS instruments with out having to handle infrastructure.

The next are some widespread buyer intents when contacting buyer care:

• Transaction standing (for instance, standing of a cash switch)
• Password reset
• Promo code or low cost
• Hours of operation
• Discover an agent location
• Report fraud
• Unlock account
• Shut account

Brokers for Amazon Bedrock may help you carry out classification and entity detection on emails for these intents. For this resolution, we present methods to classify buyer emails for the primary three intents. You can even use Brokers for Amazon Bedrock to detect key data from emails, so you possibly can automate your corporation processes with some actions. For instance, you should use Brokers for Amazon Bedrock to automate the reply to a buyer request with particular data associated to that question.

Furthermore, Brokers for Amazon Bedrock can function an clever conversational interface, facilitating seamless interactions with each inside crew members and exterior purchasers, effectively addressing inquiries and implementing desired actions. Presently, Brokers for Amazon Bedrock helps Anthropic Claude fashions and the Amazon Titan Textual content G1 – Premier mannequin on Amazon Bedrock.

Resolution overview

To construct our buyer e-mail response circulate, we use the next providers:

Though we illustrate this use case utilizing WorkMail, you should use one other e-mail software that permits integration with serverless capabilities or webhooks to perform related e-mail automation workflows. Brokers for Amazon Bedrock lets you construct and configure autonomous brokers in your software. An agent helps your end-users full actions primarily based on group knowledge and person enter. Brokers orchestrate interactions between FMs, knowledge sources, software program functions, and person conversations. As well as, brokers robotically name APIs to take actions and invoke information bases to complement data for these actions. Builders can save weeks of growth effort by integrating brokers to speed up the supply of generative AI functions. For this use case, we use the Anthropic Claude 3 Sonnet mannequin.

While you create your agent, you enter particulars to inform the agent what it ought to do and the way it ought to work together with customers. The directions substitute the $directions$ placeholder within the orchestration immediate template.

The next is an instance of directions we used for our use circumstances:

“You're a classification and entity recognition agent. 

Process 1: Classify the given textual content into one of many following classes: "Switch Standing", "Password Reset", or "Promo Code". Return solely the class with out extra textual content.

Process 2: If the categorised class is "Switch Standing", discover the 10-digit entity "money_transfer_id" (instance: "MTN1234567") within the textual content. Name the "GetTransferStatus" motion, passing the money_transfer_id as an argument, to retrieve the switch standing.

Process 3: Write an e-mail reply for the client primarily based on the obtained textual content, the categorised class, and the switch standing (if relevant). Embody the money_transfer_id within the reply if the class is "Switch Standing".

Process 4: Use the e-mail signature "Finest regards, Clever Corp" on the finish of the e-mail reply.”

An motion group defines actions that the agent may help the person carry out. For instance, you possibly can outline an motion group known as GetTransferStatus with an OpenAPI schema and Lambda operate hooked up to it. Brokers for Amazon Bedrock takes care of setting up the API primarily based on the OpenAPI schema and fulfills actions utilizing the Lambda operate to get the standing from the DynamoDB money_transfer_status desk.

The next structure diagram highlights the end-to-end resolution.

The answer workflow consists of the next steps:

1. A buyer initiates the method by sending an e-mail to the devoted buyer help e-mail deal with created inside WorkMail.
2. Upon receiving the e-mail, WorkMail invokes a Lambda operate, setting the following workflow in movement.
3. The Lambda operate seamlessly relays the e-mail content material to Brokers for Amazon Bedrock for additional processing.
4. The agent employs the pure language processing capabilities of Anthropic Claude 3 Sonnet to grasp the e-mail’s content material classification primarily based on the predefined agent instruction configuration. If related entities are detected inside the e-mail, akin to a cash switch ID, the agent invokes a Lambda operate to retrieve the corresponding cost standing.
5. If the e-mail classification doesn’t pertain to a cash switch inquiry, the agent generates an applicable e-mail response (for instance, password reset directions) and calls a Lambda operate to facilitate the response supply.
6. For inquiries associated to cash switch standing, the agent motion group Lambda operate queries the DynamoDB desk to fetch the related standing data primarily based on the offered switch ID and relays the response again to the agent.
7. With the retrieved data, the agent crafts a tailor-made e-mail response for the client and invokes a Lambda operate to provoke the supply course of.
8. The Lambda operate makes use of Amazon SES to ship the e-mail response, offering the e-mail physique, topic, and buyer’s e-mail deal with.
9. Amazon SES delivers the e-mail message to the client’s inbox, offering seamless communication.
10. In situations the place the agent can’t discern the client’s intent precisely, it escalates the problem by pushing the message to an SNS matter. This mechanism permits subscribed ticketing system to obtain the notification and create a help ticket for additional investigation and backbone.

Conditions

Seek advice from the README.md file within the GitHub repo to ensure you meet the stipulations to deploy this resolution.

Deploy the answer

The answer is comprised of three AWS Cloud Deployment Equipment (AWS CDK) stacks:

• WorkmailOrgUserStack – Creates the WorkMail account with area, person, and inbox entry
• BedrockAgentCreation – Creates the Amazon Bedrock agent, agent motion group, OpenAPI schema, S3 bucket, DynamoDB desk, and agent group Lambda operate for getting the switch standing from DynamoDB
• EmailAutomationWorkflowStack – Creates the classification Lambda operate that interacts with the agent and integration Lambda operate, which is built-in with WorkMail

To deploy the answer, you additionally carry out some guide configurations utilizing the AWS Administration Console.

For full directions, discuss with the README.md file within the GitHub repo.

Take a look at the answer

To check the answer, ship an e-mail out of your private e-mail to the help e-mail created as a part of the AWS CDK deployment (for this put up, we use help@vgs-workmail-org.awsapps.com). We use the next three intents in our pattern knowledge for customized classification coaching:

• MONEYTRANSFER – The client needs to know the standing of a cash switch
• PASSRESET – The client has a login, account locked, or password request
• PROMOCODE – The client needs to find out about a reduction or promo code obtainable for a cash switch

The next screenshot reveals a pattern buyer e-mail requesting the standing of a cash switch.

The next screenshot reveals the e-mail obtained in a WorkMail inbox.

The next screenshot reveals a response from the agent relating to the client question.

If the client e-mail isn’t categorised, the content material of the e-mail is forwarded to an SNS matter. The next screenshot reveals an instance buyer e-mail.

The next screenshot reveals the agent response.

Whoever is subscribed to the subject receives the e-mail content material as a message. We subscribed to this SNS matter with the e-mail that we handed with the human_workflow_email parameter through the deployment.

Clear up

To keep away from incurring ongoing prices, delete the assets you created as a part of this resolution if you’re finished. For directions, discuss with the README.md file.

Conclusion

On this put up, you discovered methods to configure an clever e-mail automation resolution utilizing Brokers for Amazon Bedrock, WorkMail, Lambda, DynamoDB, Amazon SNS, and Amazon SES. This resolution can present the next advantages:

• Improved e-mail response time
• Improved buyer satisfaction
• Price financial savings relating to time and assets
• Means to concentrate on key buyer subject

You’ll be able to develop this resolution to different areas in your corporation and to different industries. Additionally, you should use this resolution to construct a self-service chatbot by deploying the BedrockAgentCreation stack to reply buyer or inside person queries utilizing Brokers for Amazon Bedrock.

As subsequent steps, try Brokers for Amazon Bedrock to begin utilizing its options. Observe Amazon Bedrock on the AWS Machine Studying Weblog to maintain updated with new capabilities and use circumstances for Amazon Bedrock.

Concerning the Writer

Godwin Sahayaraj Vincent is an Enterprise Options Architect at AWS who’s keen about Machine Studying and offering steerage to prospects to design, deploy and handle their AWS workloads and architectures. In his spare time, he likes to play cricket together with his pals and tennis together with his three youngsters.

Ramesh Kumar Venkatraman is a Senior Options Architect at AWS who’s keen about Generative AI, Containers and Databases. He works with AWS prospects to design, deploy and handle their AWS workloads and architectures. In his spare time, he likes to play together with his two youngsters and follows cricket.