Overview
Amazon businesses generate a tremendous amount of data across advertising, retail performance, and Amazon Marketing Cloud (AMC). But analyzing these signals often means moving between different platforms, reports, and workflows.
In this hands-on workshop, Intentwise CEO Sreenath Reddy and Product Manager Kenton Snyder demonstrate how brands can use AI tools like Claude to bring Amazon analysis, insights, and advertising execution into one place.
The session explores three key ways to get more value from AI: prompts, skills, and automation. Using practical examples, the speakers show how teams can monitor account performance, diagnose changes, uncover opportunities, work with AMC data, execute advertising optimizations, and automate recurring analysis.
Although the demonstrations use Claude, the approaches discussed can also be applied to other AI clients, including ChatGPT and Microsoft Copilot.
Building the foundation for AI-powered Amazon analysis
Sreenath explains that effective AI-powered workflows require three technical components:
- A trusted data source: A foundation that brings together fragmented advertising, retail, and AMC data, while incorporating business-specific context such as product categories, sub-brands, and unit costs.
- An interface between AI and data: Model Context Protocol (MCP) provides a standardized way for AI tools and agents to connect with business systems and data sources.
- An AI client: Tools such as Claude, ChatGPT, or Copilot that interact with the connected data.
However, the most important component remains the user. AI tools can accelerate analysis, but business expertise, context, intuition, and judgment are essential to interpreting results and making decisions.
1. Prompts: Turn business questions into actionable insights
Kenton introduces prompts as a way to translate everyday business questions into requests that AI can answer using connected Amazon data.
Rather than thinking about prompt engineering as a technical exercise, teams can start with the questions they already ask when reviewing their accounts.
Monitor account performance
The first example compares account performance over the last 14 days against the previous period. Claude combines Vendor Central sales and units sold with advertising performance, even though these metrics typically come from separate Amazon systems.
The analysis can then be broken down further by campaign type, such as Sponsored Products automatic versus manual campaigns, or by other dimensions like match type, placement, and advertising channel.
This makes it easier to identify where performance is changing and which areas deserve closer attention.
Diagnose performance changes
Once a change is identified, teams can ask follow-up questions to understand what's driving it.
For example, Claude can identify which campaigns have the greatest impact on changes in ACOS. The analysis considers both the size of the ACOS change and the campaign's current spend, helping teams prioritize issues that have a meaningful impact on overall account performance.
Identify inventory and advertising inefficiencies
One particularly useful example combines advertising and inventory data to identify products that are spending on ads despite having low stock.
Claude calculates weeks of cover using recent sales velocity and current inventory, then compares that information against advertising spend for the top-spending products.
This helps brands identify products at risk of running out of stock while still receiving advertising investment. Teams can then investigate or take action, such as pausing ads for products with insufficient inventory.
Sreenath also highlights the value of this approach for businesses with large catalogs. AI can help teams monitor long-tail products that would otherwise be difficult to track manually.
Understand the value of connecting data sources
Sreenath explains that connecting an AI client directly to Seller Central does not necessarily provide the full picture. Businesses often need to combine advertising, sales, AMC, and cross-channel data with their own product-level context.
A connected data foundation can make these signals available together while also supporting independent dashboards and data ownership.
2. Use Amazon Marketing Cloud to understand customer value
The workshop then explores how Claude can make Amazon Marketing Cloud data more accessible to teams that may not have extensive technical or SQL expertise.
Sreenath explains that AMC provides access to shopper interactions across the journey, from impressions and clicks to conversions, alongside five years of purchase data at the individual shopper level.
This enables deeper analysis of customer lifetime value, repeat purchases, Subscribe & Save behavior, new-to-brand metrics, and audience activation.
Analyze new-to-brand performance
Kenton demonstrates how to identify products with the highest new-to-brand purchase percentages and new-to-brand sales for a selected period.
Brands can take this analysis further by combining new-to-brand data with advertising performance to understand the cost of acquiring new customers.
Measure product and customer lifetime revenue
The second example examines lifetime revenue following a customer's first purchase.
The analysis looks at two dimensions:
- Product-level lifetime revenue: How much customers spend on the product they initially purchased after that first transaction.
- Customer-level lifetime revenue: How much those customers spend across the brand's broader product catalog, including complementary products and upgrades.
The measurement period can be customized, with 12 months used as an example in the demonstration.
By comparing customer acquisition cost with lifetime revenue, brands can develop a better understanding of the long-term value of their acquisition efforts.
Kenton also explains that combining multiple datasets can reveal insights that would otherwise require significant manual effort.
3. Execution: Move from insights to advertising actions
Traditionally, teams have analyzed performance in one platform and then moved to an advertising console to make changes. With AI connected through MCP, analysis and execution can happen within the same interface.
Kenton demonstrates this by identifying the 20 keywords with the largest increase in spend over the previous seven days. He then asks Claude to reduce bids by 10% on the five keywords with the largest increases.
However, the speakers emphasize that AI-powered execution needs appropriate safeguards.
Two essential safeguards for AI actions
1. Require explicit confirmation
AI should propose changes before implementing them. In the demonstration, Claude presents the selected keywords, their current bids, and the proposed new bids. No changes are made until the user explicitly confirms the action.
2. Maintain the ability to roll back changes
Teams should have a straightforward way to undo actions if an agent makes an unintended change or something goes wrong.
These safeguards help businesses benefit from AI-driven execution while retaining control over their advertising accounts.
4. Skills: Make analysis consistent and repeatable
Prompts are useful for answering individual questions, but teams often need the same analysis performed repeatedly with consistent outputs.
Sreenath and Kenton introduce skills as a way to package detailed instructions for complex tasks into reusable workflows.
A skill defines how a task should be performed, which data should be considered, and how the results should be presented. This helps prevent the inconsistencies that can occur when asking an AI tool to perform the same task repeatedly.
Create reusable account-performance analysis
Kenton demonstrates a pre-built skill called What's Changed, which analyzes account performance over the previous seven days.
The skill contains detailed instructions that Claude follows whenever it runs. These instructions specify the metrics to include and the format of the output.
Compared with a basic prompt, the skill produces a more tailored report, including metrics such as shipped revenue and percentage changes.
Build deeper, visual reports
The demonstration then moves to a more detailed version of the skill, which identifies:
- Top-level account performance changes.
- Campaigns driving changes in sales and spend.
- Products contributing to changes in sales and spend.
The results are presented in a consistent visual format, making it easier to review performance and investigate changes.
Kenton also explains that users do not need to write lengthy skill instructions from scratch. Claude can help create a skill by working through the desired task and its requirements step by step.
Apply skills across teams
Sreenath highlights the value of skills for standardizing workflows across an organization. Examples include weekly and monthly business reviews, performance reporting, and QBR or WBR deck creation.
By defining the process once, teams can make recurring analysis more personalized, consistent, and repeatable.
5. Automation: Scale recurring analysis and reporting
The final section explores how prompts and skills can be scheduled to run automatically.
Kenton recommends evaluating potential automation opportunities using three questions:
- Is the task recurring? Does it happen daily, weekly, or on another regular schedule?
- Is the expected output defined? Does the task produce data or a report in a consistent format?
- Can the process be explained step by step? Could you give someone clear instructions to complete the task in the same way every time?
If a task meets these criteria, it may be a good candidate for AI-powered automation.
Automate weekly performance monitoring
Kenton demonstrates setting up a recurring task in Claude to monitor account performance across four areas:
- Products experiencing increases or decreases in total sales.
- Campaigns experiencing changes in spend or ROAS.
- Products running low on inventory.
- Important performance changes that need attention.
The task is scheduled to run every Monday, giving the team a regular overview of account performance at the start of the week.
Deliver insights where teams already work
The demonstration also shows how automation can connect to tools outside Claude.
Using a Slack MCP connection, Kenton configures the findings to be delivered as a personal Slack message rather than requiring users to log into Claude to retrieve them.
The same concept can be applied to other tools, such as email or ClickUp, depending on where a team manages its daily work.
This makes recurring insights more accessible and helps teams start their week with a clear view of what needs attention.
The three orchestration levers for AI-powered Amazon workflows
Sreenath concludes by bringing the three concepts together:
- Prompts provide agility: Ask ad hoc business questions, investigate performance, and initiate actions.
- Skills provide personalization and consistency: Turn complex tasks into repeatable workflows tailored to an individual, team, or business.
- Automation provides scale: Schedule recurring prompts and skills so important analysis happens reliably without requiring manual intervention each time.
The speakers emphasize that every business has unique requirements. The questions teams ask, the metrics they prioritize, and the workflows they automate should reflect their own business context.
Conclusion
AI can do more than summarize Amazon performance data. When connected to a trusted data foundation, it can help brands bring together fragmented signals, investigate performance changes, uncover customer and product-level opportunities, and execute advertising optimizations.
By combining prompts, skills, and automation, Amazon teams can move from manually collecting and analyzing data toward more personalized, consistent, and scalable workflows.
The key is to keep people involved in the process. AI can accelerate the work, but human expertise and judgment remain essential to deciding what actions make sense for the business.
Intentwise helps brands and agencies build these AI-enabled workflows on their own data and use cases. Teams interested in getting started can explore an AI enablement session with the Intentwise team.
