Hands-on insights from projects we actually run: data analytics, conversational analytics, AI workflow optimisation, ads agents, and MCP in marketing.
Apps are growing a second front door for agents. The protocol is the easy part; whether your data survives an agent reading it is the hard part.
Andrii Krutko··6 min
"AI saved us 200 hours" is an input, not a return. Capture the baseline first, then prove what those hours became in GA4 and BigQuery.
Andrii Krutko··6 min
MCP connects an AI agent to Google Ads, BigQuery and GTM. The hard part isn't the protocol — it's having data clean enough to ask questions of.
Andrii Krutko··6 min
LLM 0.32 gave Claude server-side WebSearch, WebFetch, CodeExecution and MCP. Less glue to maintain, and the same old question: is the data underneath any good?
Andrii Krutko··9 min
Waiting on engineering to change a schema is why analytics projects stall. Migrations put the shape of your GA4 data back in the analyst’s hands.
Andrii Krutko··7 min
There is no best AI model — only the best one per task. A routing matrix for marketing: generation, analysis, automation, and what each should cost.
Andrii Krutko··7 min
Your GA4 BigQuery export is a raw, event-level data model. Feature engineering — now doable with an LLM — turns it into signals your reports hide.
Andrii Krutko··7 min
OpenAI split GPT-5.6 into three tiers. Route models by job — cheap tiers for volume, reasoning for money decisions — or your cost per campaign multiplies.
Andrii Krutko··6 min
Everyone knows what they spent on AI tools. Almost nobody knows what those tools earned. A GA4 + BigQuery framework to close the gap.
Andrii Krutko··9 min
Real examples of AI agents running Ads reports, warehouse queries, and tag management through MCP — plus what we learned shipping our own tracking stack this way.
Andrii Krutko··6 min// first articles shipping soon
We're writing up what we've learned running analytics and AI workflows in production.