How AWS Sales uses generative AI to streamline account planning

Every year, AWS Sales personnel draft in-depth, forward looking strategy documents for established AWS customers. These documents help the AWS Sales team to align with our customer growth strategy and to collaborate with the entire sales team on long-term growth ideas for AWS customers. These documents are internally called account plans (APs). In 2024, this activity took an account manager (AM) up to 40 hours per customer. This, combined with similar time spent for support roles researching and writing the growth plans for customers on the AWS Cloud, led to significant organization overhead. To help improve this process, in October 2024 we launched an AI-powered account planning draft assistant for our sales teams, building on the success of Field Advisor, an internal sales assistant tool. This new capability uses Amazon Bedrock to help our sales teams create comprehensive and insightful APs in less time. Since its launch, thousands of sales teams have used the resulting generative AI-powered assistant to draft sections of their APs, saving time on each AP created.

In this post, we showcase how the AWS Sales product team built the generative AI account plans draft assistant.

Business use cases

The account plans draft assistant serves four primary use cases:

• Account plan draft generation: Using Amazon Bedrock, we’ve made internal and external data sources available to generate draft content for key sections of the APs. This enables our sales teams to quickly create initial drafts for sections such as customer overviews, industry analysis, and business priorities, which previously required hours of research across the internet and relied on disparate internal AWS tools.
• Data synthesis: The assistant can pull relevant information from multiple sources including from our customer relationship management (CRM) system, financial reports, news articles, and previous APs to provide a holistic view of our customers.
• Quality checks: Built-in quality assurance capabilities help ensure that APs meet internal standards for comprehensiveness, accuracy, and strategic alignment with our customers and business.
• Customization: While providing AI-generated drafts, the product allows AMs to customize and refine the content by uploading proprietary documents to match their unique customer knowledge and strategic approach.

The account plan draft assistant loads when a user tries to create an AP, and users copy and paste each section they want to use in their final plan.

Our AMs report reduced time to write these documents, allowing them to focus more on high-value activities such as customer engagement and strategy development.

Here’s what some of our AMs had to say about their experience with the account plans draft assistant:

“The AI assistant saved me at least 15 hours on my latest enterprise account plan. It pulled together a great first draft, which I was then able to refine based on my own insights. This allowed me to spend more time actually engaging with my customer rather than doing research and writing.”

– Enterprise Account Manager

“As someone managing multiple mid-market accounts, I struggled to create in-depth plans for all my customers. The AI assistant now helps me rapidly generate baseline plans that I can then prioritize and customize. It’s a game-changer for serving my full portfolio of accounts.”

– Mid-market Account Manager

Amazon Q, Amazon Bedrock, and other AWS services underpin this experience, enabling us to use large language models (LLMs) and knowledge bases (KBs) to generate relevant, data-driven content for APs. Let’s explore how we built this AI assistant and some of our future plans.

Building the account plans draft assistant

When a user of the AWS internal CRM system initiates the workflow in Field Advisor, it triggers the account plan draft assistant capability through a pre-signed URL. The assistant then orchestrates a multi-source data collection process, performing web searches while also pulling account metadata from OpenSearch, Amazon DynamoDB, and Amazon Simple Storage Service (Amazon S3) storage. After analyzing and combining this data with user-uploaded documents, the assistant uses Amazon Bedrock to generate the AP. When complete, a notification chain using Amazon Simple Queue Service (Amazon SQS) and our internal notifications service API gateway begins delivering updates using Slack direct messaging and storing searchable records in OpenSearch for future reference.

The following diagram illustrates the high-level architecture of the account plans draft assistant.

Solution overview

We built the account plans draft assistant using the following key components:

1. Amazon Bedrock: Provides programmatic (API) access to high performing foundation models (FMs) along with vector search capabilities and metadata filtering using Amazon Bedrock Knowledge Bases. We populate an Amazon Bedrock knowledge bases using sales-enablement materials, historic APs, and other relevant documents curated by AWS Glue jobs (see more on AWS Glue jobs in the item 4).
2. AWS Lambda: Supports two use cases:
   1. The async resolver Lambda function interfaces with the front-end client CRM and orchestrates async job IDs for the client to poll. This layer also handles input validations, user request throttling and cache management.
   2. Worker Lambda functions perform the actual heavy lifting to create AP content. These functions work concurrently to generate different sections of APs by using publicly available data, internal data, and curated data in Amazon Bedrock knowledge bases. These functions invoke various LLMs using Amazon Bedrock and store the final content in the AP’s DynamoDB database corresponding to each async job ID.
3. DynamoDB: Maintains the state of each user request by tracking async job IDs, tracks throttling quota (global request count and per-user request count), and acts as a cache.
4. AWS Glue jobs: Curate and transform data from various internal and external data sources. These AWS Glue jobs push data to internal data sources (APs, internal tooling team S3 buckets, and other internal services) and to Bedrock KBs, facilitating high quality output through retrieval augmented generation (RAG).
5. Amazon SQS: Enables us to decouple the management plane and data plane. This decoupling is crucial in allowing the data plane worker functions to concurrently process different sections of the APs and make sure that we can generate APs within specified times.
6. Custom web frontend: A ReactJS based micro-frontend architecture enables us to integrate directly into our CRM system for a seamless user experience.

Data management

Our account plans draft assistant uses an Amazon Bedrock out-of-the-box knowledge base management solution. Through its RAG architecture, we semantically search and use metadata filtering to retrieve relevant context from diverse sources: internal sales enablement materials, historic APs, SEC filings, news articles, executive engagements and data from our CRM systems. The connectors built into Amazon Bedrock handle data ingestion from Amazon S3, relational database management systems (RDBMS), and third-party APIs; while its KB capabilities enable us to filter and prioritize source documents when generating responses. This context-aware approach results in higher quality and more relevant content in our generated AP sections.

Security and compliance

Security and Compliance are paramount to AWS when dealing with data regarding our customers. We use AWS IAM Identity Center for enterprise single sign-on so that only authorized users can access the account plans draft assistant. Using Field Advisor, we use various internal authorization mechanisms to help ensure that a user who’s generating APs only accesses the data that they already have access to.

User experience

We built a custom web frontend using a micro-frontend approach that integrates directly into our CRM system, allowing AMs to access the account plans draft assistant without leaving their familiar work environment. The interface allows users to select which sections of APs they want to generate, provides options for customization, and notifies users to create their APs on time through Slack.

Looking ahead

While the account plans draft assistant has already demonstrated significant value, we’re continuing to enhance its capabilities. Our goal is to create a zero-touch account planner that sales teams can use to generate a full AP for a customer, incorporating best practices observed across our customers to provide sales teams best-in-class strategies to engage with customers. This would include:

•  Deeper integration with our bespoke purpose-built planning tools and assistance with account planning, such as automatically generating value maps and stakeholder maps.
• Enhanced personalization to tailor content based on industry, account size, and individual user preferences.
• Improved collaboration features, so that multiple sales team members can work together on refining AI-generated plans.
• Expanded use of recommendations to provide what next? ideas to our sales teams to better serve our customers.

Conclusion

The account plans draft assistant, powered by Amazon Bedrock, has significantly streamlined our AP process, allowing our AWS Sales teams to create higher quality APs in a fraction of the time they currently need. As we continue to refine and expand this capability, we’re excited to see how it will further enhance our ability to serve our customers and drive their success in the AWS Cloud.

If you’re interested in learning how generative AI can transform your sales function and its processes, reach out to your AWS account team to discuss how services such as Amazon Q and Amazon Bedrock can help you build similar solutions for your organization.

About the Authors

Saksham Kakar is a Sr. Product Manager (Technical) in the AWS Field Experiences (AFX) organization focused on developing products that enable AWS Sales teams to help AWS customers grow with Amazon. Prior to this, Saksham led large sales, strategy and operations teams across startups and Fortune 500 companies. Outside of work, he is an avid tennis player and amateur skier.

Vimanyu Aggarwal is a Senior Software Engineer in AWS Field Experiences (AFX) organization with over 10 years of industry experience. Over the last decade, Vimanyu has been focusing on building large-scale, complex distributed systems at various Fortune 500 organizations. Currently, he works with multiple teams within the AFX organization to deliver technical solutions that empower the $100 billion sales funnel. Outside of work, he likes to play board games, tinker with IoT, and explore nature.

Krishnachand Velaga is a Senior Manager for Product Management – Technical (PM-T) in the AWS Field Experiences (AFX) organization who manages a team of seasoned PM-Ts and a suite of sales products, using generative AI to enable the AWS Sales organization help AWS customers across the globe adopt, migrate and grow on the AWS Cloud in line with their business needs and outcomes while bolstering sales efficiency and productivity and reducing operational cost.

Scott Wilkinson is a Software Development Manager in the AWS Field Experiences (AFX) organization, where he leads a cross-functional engineering team developing tools that aggregate and productize data to power AWS customer insights. Prior to AWS, Scott worked for notable startups including Digg, eHarmony, and Nasty Gal in both leadership and software development roles. Outside of work, Scott is a musician (guitar and piano) and loves to cook French cuisine.