SalamaSons — Ad Intelligence & AI Recommendation Loop
A home-appliances retailer in Mauritania was spending on social ads and reading the results by scrolling. I built the measurement loop: automated collection of post and ad performance into a warehouse sheet every four hours, and an AI layer that reads it and sends a recommendation per ad straight to the owner's Telegram.
Client confidentiality: no spend figures, revenue, or account identifiers appear in this case study. The build is the story.
01The Business Problem
A single-owner retail operation running paid social in a market where every ad decision is made by feel: open the ads manager, scroll the posts, remember roughly which one "did well last week," and adjust. The performance data existed — it was just locked inside platform screens, unaggregated, never compared, and never written down. So nothing accumulated: last month's lesson was gone by this month.
02My Role
Built and deployed independently: the data-collection workflows against the social platform, the sheet schema that acts as the warehouse, the scheduling and chaining, the prompt design for the advisory layer, and the Telegram delivery. Design, build, and ongoing operation.
03The Loop
the two chained workflows on the n8n canvas
(blur credentials & account IDs)assets/salamasons-workflows.png
04What I Built
Post-performance collection
A scheduled workflow pulls the published posts and their impressions, writing each into a Google Sheet — so reach stops being a number you glance at and becomes a series you can compare across weeks.
Full ad-detail extraction
The same pass collects the complete detail behind every running ad and lands it in the sheet alongside performance — creative, targeting and delivery attributes sitting in the same row as the result they produced.
A four-hour heartbeat
The cycle runs every four hours rather than daily: fast enough to catch an ad going wrong the same day it does, slow enough that the numbers have settled and the owner isn't buried in noise.
Chained AI advisory
A second workflow fires after collection completes, reads the accumulated data, and passes it to OpenAI for a per-ad recommendation — then delivers it to the owner's Telegram. The advice arrives in the app he already checks, not in a dashboard he has to remember to open.
05Engineering Decisions
06How This Differs From the Al-Raky Build
Both are commerce automation, but they solve opposite halves of the same business. Al-Raky is a pre-sale production system: it takes a photo and a price and manufactures everything needed to sell — content, media, listings, sales agents. SalamaSons is a post-publication measurement system: it takes what's already running and turns its results into advice. One creates the ads; this one judges them.