Introduction
Most Amazon teams already have repeatable processes for managing ads, budgets, inventory, keyword performance, and day-to-day optimization. The challenge is that many of those processes still live in static documents, spreadsheets, checklists, or in someone’s head.
In this webinar, Kenton Snyder, Product Manager at Intentwise, walked through how brands and agencies can turn their Amazon SOPs into AI-powered workflows using tools like Claude, ChatGPT, and connected data sources through MCPs.
The big takeaway: AI becomes much more useful when it understands how your business actually operates.
Why generic AI outputs fall short
Most teams have experimented with AI in some capacity. But out of the box, AI responses can feel generic. That is because the model does not automatically understand your brand, your market position, your performance thresholds, or the way your team makes decisions.
For example, your AI tool may not know that:
- Your brand is a challenger brand in a competitive category.
- Your hero ASIN was out of stock during the same period last year.
- You have specific ACoS or ROAS thresholds for campaign analysis.
- You avoid harvesting branded or competitor terms in certain workflows.
- You manage budgets differently across campaigns or product lines.
Without that context, AI can still provide answers, but those answers may not reflect how your team actually works.
SOPs are the bridge between your team and AI
Kenton explained that SOPs can act as a training layer for AI agents. In the same way you would onboard a new team member by giving them process documentation, you can give an AI agent the same instructions.
That might include:
- How to review daily ad performance.
- How to identify campaigns over or under target.
- How to flag high-spend campaigns with zero sales.
- How to evaluate search terms for harvesting.
- How to check inventory before making ad recommendations.
- How to summarize findings for the team.
The SOP does not need to be overly complex. It simply needs to explain the steps your team already follows, the thresholds that matter, and the decisions the AI should help support.
How to prepare SOPs for AI
One practical recommendation from the session was to convert SOP documents into markdown files. Markdown is easier for many AI tools to read and can reduce the amount of context needed for the model to understand the documentation.
Teams can start with whatever format they already use, including Google Docs, Word documents, PDFs, Notion pages, or internal checklists. From there, they can convert the SOP into a cleaner structure that AI tools can reference more effectively.
Kenton also recommended keeping SOPs in a living source of truth, such as Notion, Google Drive, Airtable, or another system the team already uses. The key is to avoid creating a separate process just for AI. Instead, teams should connect AI to the workflows and documentation they already maintain.
Setting up an AI agent for recurring analysis
During the live demo, Kenton showed how to create an Amazon performance analyst agent that could run a daily SOP-based analysis.
The agent was given brand context, including the example brand’s market position, category, pricing, seasonality, historical issues, and data source. It was then given SOP files that outlined the steps it should follow.
From there, the AI agent was asked to create a recurring task that would:
- Pull campaign performance data.
- Compare recent performance against trailing averages.
- Check spend pacing against budget.
- Flag campaigns with ACoS or ROAS issues.
- Identify high-spend campaigns with no sales.
- Check inventory health for key ASINs.
- Summarize the findings in Slack.
This turns a manual reporting process into a repeatable workflow that can run automatically and deliver insights where the team already works.
Feedback loops make the agent smarter over time
One of the most important parts of the workflow is feedback.
Kenton showed how teams can review the AI-generated output and provide corrections or additional context. For example, if the AI uses the wrong budget assumption, the user can update the agent with the correct monthly budget. If the AI recommends harvesting a branded term, the user can tell it not to include branded terms in future recommendations.
That feedback helps the AI agent improve over time and become more aligned with the team’s operating model.
Human review still matters
The session also covered where AI agents fit today. While AI can already support analysis, recommendations, and reporting, Kenton recommended keeping a human in the loop before allowing AI to take independent action.
The best starting point is to use AI for read-only analysis, recurring reporting, and recommendations. As the team builds confidence in the outputs, they can decide whether to move toward more autonomous workflows.
Managing token usage and model selection
Kenton also shared practical guidance on token usage. For many structured SOP-based tasks, a high-end reasoning model may not be necessary. If the SOP clearly outlines what the agent should do, a more efficient model like Claude Sonnet may be enough for recurring analysis.
He also noted that MCPs can help reduce token usage compared to uploading large CSV files, because the AI can request only the data it needs rather than processing an entire export.
Where Intentwise AI Gateway fits in
In the demo, Kenton used Intentwise AI Gateway to connect Amazon and Walmart data into AI tools like Claude and ChatGPT. This allowed the AI agent to access the performance data needed to complete the SOP-based analysis.
The broader workflow is not limited to AI Gateway, but AI Gateway makes it easier for retail media teams to bring commerce data into AI-powered workflows without relying on static exports.
Final takeaway
AI is most powerful when it is connected to your business context, your data, and your way of operating.
By turning SOPs into AI-ready workflows, Amazon teams can move beyond generic prompts and start building agents that understand their processes, monitor performance, surface issues, and improve over time.
For brands and agencies managing complex Amazon and Walmart operations, this is a practical first step toward more scalable, AI-powered commerce workflows.
I can also make this more SEO-focused, more executive/strategic, or shorter for a clean on-demand landing page section.
