Adscalr
Ad intelligence for people who buy media. It turns competitor activity, audience language and performance data into the next ad to ship.
Status: web platform in private preview. Built by one media buyer for his own accounts.
What it does
Six systems that feed each other.
Competitor ads, decoded daily
The Meta, TikTok and Google ad libraries are pulled each day. Vision AI breaks a static ad into about twenty structured fields with design, copy and strategy scores from 1 to 10. Video is read frame by frame for hook, pacing and CTA placement.
A market map on five axes
Competitors are clustered on price, quality, speed, trust and innovation, so you can see which message nobody in your category is making.
Audience language from five sources
Reddit, Amazon reviews, the App Store, Google Play and forums, captured as the exact phrasing people used. Four to six personas per project, each backed by at least three real quotes and mapped onto the five awareness stages.
Creative with an attention prediction
Static ads, UGC concepts and motion storyboards in four native formats. Statics get a saliency prediction from DeepGazeIIE, a model built on eye tracking data, and a copy critic scores them against six direct response principles.
Testing with real statistics
Composite scoring across six metrics with Bayesian shrinkage, Benjamini-Hochberg false discovery control at q = 0.10 in pattern mining, and Thompson sampling to pick what to test next.
Automation with four safety catches
Rules fire on eight metrics and can only pause or kill. A learning phase lockout, a ROAS floor, a minimum of three active ads under CBO and a frequency ceiling stop a rule from killing a winner. Every kill is reversible for 24 hours.
What it will not do
The boundaries are the useful part of a claim.
Nothing here is real time
Ad libraries are scraped roughly daily, pacing alerts run every five minutes, and theme analysis is cached for 24 hours.
It advises, it does not decide
Creatives are never auto rejected or auto improved, and nothing reaches Meta without a person approving it. Budget plans and language targeting are recommendations, and estimated values are labelled as estimates.
Video rendering is not built
UGC concepts and briefs are real output. Turning them into finished video sits on the roadmap and is not shipping today.
Where it stands
Private preview. 478 API routes and 47 database tables, built end to end by one person in about three months, across three ad platforms. Access happens through a conversation for now.