Blogs & Webinars

Session 44: AI-Augmented Marketing: Two Practitioners, Two Skills, Zero Theory

Two marketing practitioners walk through Claude Code skills they built — Varun automates performance marketing analysis across 5 data silos, Yash automates competitor ad scouting at scale.

Welcome to session 44! This week we have two practitioners — Varun Dhamija and Yash Haldankar — who each built a Claude Code skill to automate a real marketing workflow. No theory, just working tools.

Varun Dhamija is the Founder of Altiv.AI and a former P&L leader at Pearson and TimesPro with 15+ years in tech. He builds practical AI solutions for marketing and growth teams.

Yash Haldankar is a brand builder with 7 years at TimesPro, Delhivery, upGrad, and Internshala. A MICA alumnus, he's currently automating marketing workflows and building AI content pipelines with Claude Code.

Resources

Session Overview

Two practitioners, two Claude Code skills built in the real world. Varun eliminated 2-hour manual analysis cycles across 5 marketing data silos. Yash automated competitor ad library monitoring across brands — extracting screenshots, teardowns, and strategic insights from 554+ ads without touching a browser manually.

Key Takeaways

  • Free analysis changes behavior: When running /meta-analysis costs nothing in effort, you run it daily instead of weekly — and you catch problems before they compound.
  • The data is already there: Broken UTMs, missing pixels, misallocated ad budgets — they exist in your data right now. Automation surfaces them; manual analysis misses them because it's too slow.
  • A skill is just a markdown file: Both Varun and Yash distribute their Claude Code skills as .md files you drop into your project folder. No installation, no setup — just context.
  • Puppeteer MCP handles what LLMs can't: Ads library browsing requires a real browser. Puppeteer MCP opens the browser, clicks into each ad, takes the screenshot, and hands the visual to Claude for analysis.
  • White space beats copying: The motivator map Yash's skill generates shows which angles competitors are saturating — helping you find what nobody is running rather than what everyone is.

Talk 1: Varun — Performance Marketing Analysis with Claude Code

The Problem

Varun's performance marketing data lived across 5 separate systems: Meta Ads, landing page code, CRM leads, conversion events, and analytics. Manual analysis took 2 hours per cycle, ran weekly at best, and still missed things.

The Skill: /meta-analysis

The skill pulls data from all five sources and analyzes them together. In a single run, it found:

  • 57% of budget going to feed placement — a major misallocation that manual review hadn't flagged
  • Missing Meta pixel on one of three landing pages, silently killing ad optimization
  • Duplicated UTM string set by their agency — classifying all paid traffic as organic social for weeks
  • Copy outperforming creative — a counter-intuitive finding that redirected creative effort

The Meta-Lesson

When analysis is free, you run it daily. Daily analysis changes your instincts. You stop discovering problems in retrospect and start catching them before they compound.

Talk 2: Yash — Competitor Ad Scouting Automation

The Problem

Monitoring competitor ad libraries meant hours of manual scrolling through Meta Ads Library, screenshot hoarding, and subjective breakdowns. Brands iterate fast — by the time analysis was done, the landscape had shifted.

The Skill: /competitor-ad-teardown

Yash's skill uses the Puppeteer MCP to automate the browser entirely — opening ad libraries, clicking into individual ads, taking screenshots, and passing visuals to Claude for analysis. In the live demo, it analyzed 554 protein brand ads across MuscleBlaze, Whole Truth, SuperU, and Wellbeing Nutrition.

What the skill extracts:

  • Share of voice across brands for a given keyword
  • Testing velocity — MuscleBlaze launched 48 new creatives in the month; others focused on positioning
  • Motivator map — performance/absorption (MuscleBlaze), clean label (Whole Truth), celebrity aspiration (SuperU), women's health (Wellbeing)
  • White spaces — angles no competitor is running, surfaced for a new brand to own
  • Creative formats — hooks, social proof usage, offer/CTA patterns, visual composition

The output is a structured PDF report with a custom analysis framework Yash built into the skill definition.

The Meta-Lesson

When competitor teardown takes 30 minutes and runs weekly, you stop reacting to competitors and start anticipating them. Gaps become obvious. Pivots happen faster.

Q&A Highlights

  • How to distribute a skill: Drop the .md file in your project folder, open it in Claude Code or VS Code, add your own brand names — done. No installation.
  • Puppeteer MCP vs. web scraping: Puppeteer MCP controls a real browser, so it handles JavaScript-rendered pages like Meta Ads Library that standard scrapers can't reach.
  • Token costs: Both skills are designed to stay within reasonable token usage. Yash's skill analyzes all ads but only takes screenshots of the top 5 by default — configurable.
  • Local vs. cloud models: The skills work with any Claude model. Local models (via Ollama) work for parts of the pipeline but struggle with visual analysis of ad screenshots.
  • Getting started: Pick one repetitive analysis task. Write down the steps you do manually. Turn that into a skill .md file. Run it once. Refine.

Here's the entire recording of the session.