EP 18 - Why AI Gets Your Marketing Data Wrong: GA4, Attribution and ROAS | Ahmed Elsamadisi

Fifteen years in data. He created the Activity Schema. He still can't fact-check what an AI tells him. Here's why that should worry every marketer running numbers through an agent. In this episode of Velocity: Performance Marketing Podcast by Hawky, we sit down with Ahmed Elsamadisi, Founder of Polyform and creator of the Activity Schema, the open-source event model that other companies now consult on. Ahmed has spent his career on the layer between raw data and the questions people ask of it, and he built Polyform because that layer is where AI answers quietly go wrong. Ahmed makes the case that AI is optimised to be helpful, not to be right, and that most teams have no way to tell the difference. He unpacks why the best models still get most data questions wrong, why he forbids his own AI from writing SQL, why pointing an agent at GA4 returns confident answers instead of correct ones, and where 10 to 20 percent of ad budgets quietly disappear. He also tells most brands not to fix any of it: under 5,000 dollars a month in spend, his advice is to accept that you are guessing. 🎯 Key Topics Covered: 00:00 - Meet Ahmed Elsamadisi: fifteen years in data and the Activity Schema 01:09 - Why you can't fact-check an AI's answer, and the accuracy numbers nobody quotes 04:57 - How to engineer an agent that gets it right, and why his AI never writes SQL 08:56 - Stop putting everyone in a room to define a metric: build seven instead 11:34 - Where this actually gets deployed, from go-to-market to product analytics 13:54 - How Poly works, and the moment it escalates to a human 17:45 - Kodak, Blockbuster, and why bolting AI onto legacy doesn't work 20:37 - Salesforce, MCP servers, and the CRM shift already underway 22:58 - Why he pays 5x more to leave HubSpot 24:35 - Where chat stops and the UI has to take over 27:01 - Performance marketing: attribution, junk clicks, and where budget dies 31:31 - What to actually connect, starting with raw GA4 rather than reports 32:30 - Why GA4 breaks AI agents in two different ways 33:46 - Don't fix your data under 5,000 dollars a month 34:40 - The linearity test: double the spend, do the sales double? 35:51 - What "AI native" actually means for a product 38:42 - Dashboards, daily digests, and what belongs in an email Meet Surender (Host & Co-Founder at Hawky):   / say2surender   Meet Ahmed Elsamadisi:   / elsamadisi   🔗 Polyform: https://www.polyform.to/ 🔗 Activity Schema (open source): https://activityschema.com/ About Hawky: An agentic performance marketing platform. Two always-on agents and a Copilot that buy media against your KPI, generate on-brand creatives from your winners, and answer any question about the account instantly. Other tools generate. Hawky operates. 👉 Book a Hawky demo: https://hawky.ai/get-demo 🎧 More episodes: https://hawky.ai/podcast #PerformanceMarketing #GA4 #MarketingAnalytics #Attribution #AIAgents #MarketingData #DataStack #MarketingPodcast