--- title: "AdsPower - AI-Powered Ad Campaign Optimization SaaS Platform" description: "SaaS platform for AI-driven ad campaign optimization" locale: "en" canonical: "https://portfolio.josedacosta.info/en/achievements/adspower-plateforme-optimisation-ia" source: "https://portfolio.josedacosta.info/en/achievements/adspower-plateforme-optimisation-ia.md" html_source: "https://portfolio.josedacosta.info/en/achievements/adspower-plateforme-optimisation-ia" author: "José DA COSTA" date: "2016" type: "achievement" slug: "adspower-plateforme-optimisation-ia" tags: ["Symfony 3.2", "Angular 4/10", "TypeScript", "Azure ML Studio", "NLTK", "Google AdWords API", "Bing Ads API", "Facebook Ads SDK", "MySQL", "GitLab CI", "Vagrant", "Electron"] generated_at: "2026-04-23T15:45:28.527Z" --- # AdsPower - AI-Powered Ad Campaign Optimization SaaS Platform SaaS platform for AI-driven ad campaign optimization **Date:** January 2016 - December 2018 **Duration:** 3 years **Role:** Technical Project Manager & Co-Founder **Technologies:** Symfony 3.2, Angular 4/10, TypeScript, Azure ML Studio, NLTK, Google AdWords API, Bing Ads API, Facebook Ads SDK, MySQL, GitLab CI, Vagrant, Electron ## Project Overview **Content:** AdsPower was an **early-stage startup** building a SaaS platform for AI-driven ad campaign optimization. The product used **ML classification models** (Azure Machine Learning Studio) to classify user behaviors, analyze search intent via Google SERPs, and **automatically optimize bids** based on conversion rates. The core concept: ingest thousands of campaign data points - keywords, bids, quality scores, click-through rates, conversion data - feed them into **supervised classification models**, and output **actionable optimization recommendations**. The project was directly inspired by **Optmyzr** and **Dolead** (French performance marketing platform), both analyzed in depth during our competitive research. ## Objectives & Context **Content:** Business objective: enable Google AdWords advertisers to automate their campaigns with AI, detecting non-performing keywords and adjusting bids in real-time. Context: the market was dominated by Optmyzr (US) and Dolead (FR). We aimed at a more ML-first approach with SERP scraping and intent classification. Stakes: compete with established platforms, build a product/R&D team, manage dependency on Google/Bing/Facebook APIs. Risks: API quotas, breaking changes, competition, imprecise classification models. ## ML Approach & Technical Pipeline **Content:** Three-step ML pipeline: (1) Data Collection via Google AdWords, Bing Ads APIs, and an in-house SERP Scraper built with Goutte and CasperJS to fetch structured search results. (2) NLP Processing with NLTK for tokenization, stemming, stop words removal, and TF-IDF vectorization. (3) ML Models on Azure ML Studio: supervised Gradient Boosting classification for relevance prediction, unsupervised k-means clustering for keyword segmentation, and Google Prediction API for audience segmentation. Architecture: Symfony 3.2 backend with 4 bundles (Core, Api, Admin, QueryBuilder), Angular 4 then Angular 10 frontend with Highcharts and Syncfusion. Infrastructure: Vagrant Ubuntu 16.10, MySQL utf8mb4, Memcached, GitLab CI/CD, Electron desktop + Cordova mobile. ## Team & Organization **Content:** Scrum team of 5: 1 CEO/founder, 1 CTO/technical co-founder (my role: technical project manager and co-founder), 2 freelance backend/frontend developers, 1 UX designer. Organization: 2-week sprints with Sprint Planning (Trello), development, code review (GitLab MR), automatic staging deploy (GitLab CI), Sprint Review (product demo), Retrospective. 18+ code repositories across 3 major versions. 40+ mockup pages. Daily coordination and external freelancer management. Commercial interactions with early beta testers, weekly product feedback. ## Results & Achievements **Content:** Functional SaaS platform in production with automatically managed campaigns. Validated proofs of concept for search intent classification and negative keyword detection. Multi-platform applications: Angular web SPA, Electron desktop (Mac/Windows/Linux), Cordova mobile (iOS/Android). Operational ML pipeline with Azure ML Studio. Integration with Google AdWords, Bing Ads, Facebook Ads APIs. Personally: first experience co-founding a tech startup, managing a technical team, and R&D in machine learning applied to digital marketing. Validated my Technical Project Manager skills. ## Project Outcome **Content:** The project ended in late 2018 after 3 years. The startup did not find product-market fit at the level required to raise significant funds. Lessons learned: strict regulatory constraints of Google/Facebook APIs, fierce competition in the US market, and the difficulty of differentiating a technical product against established players. These lessons fed my subsequent projects (Groupe Pichet, then ACCENSEO). ## Critical Reflection **Content:** Strengths: modern technical architecture for the time, motivated team, ML-first approach differentiated from purely heuristic competitors. Weaknesses: technical over-investment before product-market fit validation, excessive dependency on third-party APIs (changing Google quotas), and lack of commercial power to face Optmyzr and Dolead. Lessons: validate the market before over-building. A tech startup does not survive without commercial traction, even with excellent ML models. Favor quick POCs over complete platforms before having the first paying customers. This project was a school of entrepreneurial lucidity that served me at ACCENSEO later.