Introduction
Amazon advertising analysis can be time-consuming. For many brands and agencies, weekly performance reviews still involve exporting reports, building pivot tables, comparing campaign data, reviewing search terms, identifying anomalies, and then writing up the findings.
In this live workshop, Kenton Snyder, Product Manager at Intentwise, showed how Amazon teams can use Claude, connected to their Amazon data through Intentwise AI Gateway, to diagnose ad performance faster and turn recurring analysis into repeatable workflows.
The session focused on a practical question: how can AI help teams understand what changed, why it changed, and what action to take next?
Moving From Manual Analysis to AI-Assisted Diagnosis
Kenton opened by explaining the traditional workflow for Amazon ad analysis. A team might export campaign-level data, keyword data, search term data, and product-level data, then compile everything into large spreadsheets or dashboards.
That process can work, but it takes time. It also makes follow-up analysis harder because each new question may require more data cleanup, another pivot table, or additional manual review.
With AI tools like Claude, that workflow starts to change. When Claude is connected to Amazon data, teams can ask natural-language questions such as:
- Why did ACOS change over the last 14 days?
- Which campaigns are hitting budget caps while still performing efficiently?
- Where are we seeing new keyword or search term opportunities?
- Which campaigns need bid or budget adjustments?
- Where are we underinvesting?
Instead of starting with a spreadsheet, the analysis can start with a business question.
Context Is What Makes AI Analysis Useful
One of the biggest takeaways from the workshop was that AI needs brand-specific context to provide useful recommendations.
By default, AI tools are general-purpose. They can analyze data, but without context, their recommendations may be generic. To make the analysis more relevant, Kenton showed how to create a Claude project with a brand context file.
That context file can include information such as:
- Company overview
- Target performance metrics
- Annual revenue and spend goals
- Target ACOS by campaign type
- Key ASINs and campaigns
- Known problem areas
- Seasonality trends
- Competitive context
- Recent account changes
This gives Claude a better understanding of the account before it begins analyzing performance. For example, if a sponsored brands campaign has a different target ACOS than a sponsored products campaign, Claude can reference that target when deciding whether performance is healthy or concerning.
Setting Up a Claude Project for Amazon Analysis
Kenton recommended using Claude Projects as a way to keep context consistent across sessions.
Instead of re-explaining the brand, goals, and account structure every time, teams can upload a markdown context file once and use it as the foundation for future analysis.
He also recommended using markdown files because they are easier for AI tools to process and use fewer tokens than heavier file formats.
For teams connected to Intentwise AI Gateway, Claude can query Amazon Ads, Seller Central, and Vendor Central data directly through the MCP connection. For teams without an MCP connection, similar workflows are possible by uploading CSV exports manually, but they are less automated and harder to schedule.
Diagnosing ACOS Changes With Claude
The first live example focused on diagnosing ACOS movement.
Kenton prompted Claude to identify which campaigns drove ACOS changes over the last 14 days and determine whether the issue came from bids, new keywords, or external changes.
Claude then queried campaign-level and search term-level data through Intentwise AI Gateway and summarized what changed. It identified campaigns that hurt performance, patterns around click and order movement, possible competitive pressure, search term changes, and recommended next steps.
The key value was not just that Claude summarized performance. It also connected the performance change back to possible drivers and suggested where to investigate next.
Finding Budget and Scaling Opportunities
The second example focused on finding campaigns that hit their daily budget at least three days in the prior week while keeping ACOS below a target threshold.
Claude identified campaigns that were budget constrained but still efficient, then recommended which ones could potentially scale with more budget.
This is a strong example of how AI can support weekly optimization. Instead of manually filtering for budget-capped campaigns, comparing ACOS targets, and reviewing spend patterns, teams can ask Claude to find the opportunities directly.
Turning Analysis Into a Recurring Agent Workflow
After walking through one-off prompts, Kenton showed how to make the workflow recurring.
Using Claude Co-Work, he created a weekly ad analysis project with a context file and an agent prompt. The agent prompt instructed Claude to run a set of weekly checks, including:
- ACOS anomalies
- Campaigns hitting budget caps while performing well
- Keyword harvesting opportunities
- Revenue declines
- CPC changes
- Underinvested campaigns with strong conversion rates
The output was designed to be short and actionable. Claude was instructed to summarize each check, identify whether there was a flag, and provide the top recommended actions ranked by likely impact.
Kenton also showed how the workflow could send results to Slack, making the analysis easier to share with a team. The same approach could also be used with email or simply kept inside Claude.
Memory Helps Track Trends Over Time
Another key part of the recurring workflow was memory.
By saving key results from each weekly analysis, Claude can reference past findings and start to identify trends over time. That means the workflow can become more useful as it runs repeatedly.
For example, if the same campaign keeps showing up as a concern, or the same account trend appears several weeks in a row, the AI can call that out in future summaries.
This moves AI from a one-time analysis tool to a lightweight performance monitoring system.
Why an MCP Connection Matters
Kenton explained that MCPs are one of the best ways to connect AI tools to live business data.
An MCP acts like an API for AI tools, allowing Claude or ChatGPT to access the data needed to answer a question without requiring manual CSV uploads every time.
Intentwise AI Gateway connects Amazon Ads, Seller Central, and Vendor Central data into tools like Claude and ChatGPT. This allows teams to ask questions that combine ad performance, sales, inventory, finance, and account context.
The benefit is not only convenience. It also helps with data stitching. Connecting campaigns, keywords, products, inventory, and sales data across Amazon systems is difficult to do manually. An MCP connection gives AI a cleaner path to the right data.
When Manual Uploads Still Work
For teams that are not yet connected to an MCP, Kenton explained that they can still use a similar process by uploading reports manually.
The recommended approach is to create a Claude project, upload a context file, then manually provide the relevant reports each time the analysis needs to run.
This can still be helpful, but it is less scalable. The workflow will not be fully automated, and Claude may struggle more when it has to interpret and stitch together several CSV files on its own.
Security and Data Access Considerations
The session also covered security considerations when connecting AI tools to business data.
Kenton noted that teams should understand both the security of the MCP connection and the plan they are using with their AI provider. In general, teams should look for no-retraining agreements and, when needed, zero-data-retention options.
For business use cases, he recommended avoiding consumer-level AI plans when working with sensitive business data and instead using team, enterprise, or contained environments where data protections are stronger.
Key Takeaways
The biggest takeaway from the session is that AI can help Amazon teams move from manual reporting to repeatable performance diagnosis.
For teams getting started, the recommended steps are:
- Create a brand context file
- Use Claude Projects to keep context consistent
- Connect Amazon data through an MCP when possible
- Start with practical weekly analysis questions
- Turn repeatable checks into an agent prompt
- Schedule the workflow to run automatically
- Deliver results where the team already works, such as Slack
- Use memory to track recurring trends over time
AI does not remove the need for human judgment. But it can reduce the time spent gathering data, comparing reports, and writing summaries. For Amazon brands and agencies, that means more time focused on deciding what to do next.
