A year ago I was doing everything manually. Keyword research for Etsy listings, weekly review of Meta Ads data, competitor analysis, product description writing. Repetitive tasks that followed the same logic every time and produced the same type of output.
Today most of that runs automatically. Here’s the exact stack I use and how each piece fits together.
Why Make.com
Make.com is a no-code automation platform — it lets you connect apps and build workflows without writing code. It has a visual interface where you chain “modules” together: trigger something, do something, send somewhere.
I use it for everything that fits the pattern: input arrives → process it → output goes somewhere.
Make.com’s strength over alternatives like Zapier is the complexity it can handle. Zapier is better for simple two-step automations (when this happens, do that). Make.com is better for multi-step workflows with conditional logic, API calls, data transformation, and loops.
All my automations run on Make.com’s free or lowest paid tier. The cost is negligible.
Why Claude AI
Claude is Anthropic’s AI model. For text generation and analysis tasks — which is what I use AI for — Claude Sonnet is my default choice.
The reason is instruction following. When I write a prompt that says “produce output in this exact structure with these exact sections at these exact lengths”, Claude follows it consistently. Other models drift — they add sections I didn’t ask for, change the format, produce variable-length outputs that break downstream processing.
For automations where the AI output needs to be consistent and parseable, consistency matters more than raw capability.
The Four Automations I Run
1. Weekly Meta Ads Report
Trigger: Every Monday 9am Step 1: HTTP request to Meta Graph API → pulls 7 days of campaign data (spend, impressions, clicks, CTR, CPM, conversions, ROAS per campaign) Step 2: Data passed to Claude Sonnet with a structured prompt → produces a 5-section report: executive summary, key metrics, best campaign, worst campaign + fix, 3 actions for next week Output: Report delivered by email, dashboard at louvrlabs.com/report updated
Cost per run: ~€0.007
This is the automation that saves the most time. An hour of manual work every Monday reduced to 20 minutes of reading and executing pre-decided actions.
2. Etsy Listing Optimizer (ListingBoost)
Trigger: User submits a product description via the ListingBoost interface Step 1: Input processed through a keyword research module → identifies primary and secondary keywords, validates search volume via Etsy autocomplete patterns Step 2: Data passed to Claude Sonnet → produces an optimized title (140 characters, primary keyword front-loaded), 13 non-overlapping tags, and a structured description Output: Formatted listing copy delivered to the user interface
This is the core of ListingBoost. What takes a seller 45-90 minutes of manual research and writing per listing takes the system about 30 seconds.
3. Etsy Listing Audit
Trigger: User submits an existing Etsy listing URL Step 1: HTTP request to fetch listing data Step 2: Analysis module checks: primary keyword position in title, character count, tag overlap, buyer intent language presence, description structure Step 3: Claude Sonnet produces a scored audit with specific recommendations for each issue found Output: Audit report with score and prioritized fixes
4. Competitor Research (Internal)
Trigger: Manual — I run this when researching a new product category Step 1: Fetch top-ranking Etsy listings for a target keyword Step 2: Extract titles, tags, and pricing from each listing Step 3: Claude Sonnet analyzes patterns — which keywords appear consistently, what price points are common, what gaps exist in the market Output: Research summary with keyword patterns and positioning opportunities
This one isn’t automated in a time-triggered sense — I run it on demand when I need it. But it replaces 3-4 hours of manual competitor research with 10 minutes of setup and a 5-minute read.
The Prompt Engineering That Makes It Work
The quality of AI output depends almost entirely on prompt quality. A vague prompt produces vague output. A specific, structured prompt produces specific, structured output.
Every prompt I use in production follows the same format:
- Role definition: “You are a [specific type of expert] specializing in [specific domain]”
- Task description: Exactly what I want produced, in what format, at what length
- Structure specification: The exact sections, headers, and constraints for the output
- Context injection: The data or information the model needs to do the task
- Output constraints: Format, length, tone, what to include and explicitly what to avoid
The Meta Ads Report prompt, for example, specifies that the executive summary must be no longer than 3 lines, that the “3 actions” section must contain exactly 3 actionable items with specific campaign names from the data, and that recommendations must be based on the data provided rather than general best practices.
Without those constraints, Claude would produce a good but generic report. With them, it produces a report specific enough to act on directly.
What This Stack Can’t Do
Automations handle repeatable tasks with consistent logic. They don’t handle ambiguity, novel situations, or anything requiring judgment about things the model wasn’t trained to evaluate.
The Meta Ads Report tells me what happened and suggests what to do. It doesn’t know whether I’m planning a product launch next week that should change the budget strategy. It doesn’t know that a supplier issue is affecting my inventory and I should pull back spend on one product. Context that exists outside the data it has access to doesn’t exist for the automation.
I make the final calls. The automation does the data processing and preliminary analysis that makes those calls faster and better informed.
Getting Started With This Stack
If you want to build something similar, start with a single use case — the most time-consuming repetitive task in your business — and build one automation for it.
Make.com has a generous free tier. Claude API access is cheap for most use cases (the Meta Ads Report costs me €0.007 per run). The upfront investment is time, not money.
The compounding effect of automating one task well is that it frees up the attention to identify and automate the next one. Start simple, build incrementally.
→ louvrlabs.com — e-commerce automation as a service