Agentic Experience Optimisation for Online Stores

Commerce is going through its biggest shift in 25 years, and most online stores are blind to it. Shoppers are handing tasks to AI agents — find this, compare that, add to cart, check out — and those agents make buying decisions in milliseconds.

But stores were built for human eyes, not autonomous code. Agents can often find a brand yet fail the moment they try to act: broken DOM selectors, CAPTCHA walls, bot blocks, fragile checkout flows, and missing structured data quietly kill transactions a human would have completed.

None of it surfaces in Google Analytics, which can't even tell agent traffic from human — so the revenue leak is invisible, and stores lose high-intent, high-AOV agent shoppers without ever knowing it happened.

The discoverability layer is shifting in parallel: as buyers ask ChatGPT, Gemini, and Perplexity for recommendations instead of scrolling search results, brands that aren't cited simply don't exist. And quantifying any of this was nearly impossible — there was no score, no diagnostic, no way to watch an agent attempt a purchase and see precisely where it broke.

Every month without that visibility, competitors who get agent-ready first lock in a lead that compounds.

In just 12 weeks, Appening took Geck from concept to a production-ready beta — turning an invisible, unmeasurable problem into something brands can finally see, score, and fix:

• The revenue leak became visible: Brands can now watch real AI agents attempt checkout, login, and search on their own store and see exactly where each one breaks — failures previously hidden from analytics are mapped, quantified, and tied to lost GMV.
• "Are we agent-ready?" got a number: Every store gets a 100-point AXO score and a prioritized, engineer-ready list of fixes, replacing guesswork with a backlog ranked by revenue impact.
• Agent traffic stopped hiding inside human traffic: A lightweight pixel separates AI bots from real visitors in real time, giving teams their first true read on how much of their traffic — and revenue — is already agentic.
• Discoverability is measured alongside transactions: Brands can see where they're cited or missing across ChatGPT, Gemini, Perplexity, and others — fixing being found and being bought in one place.
• It holds up at enterprise scale: The platform ingests 1M+ tracking events a day with zero dashboard lag and charts that render in under 150ms — proof it was engineered to production standard.

To build a high-performance, real-time analytics suite, our engineering team designed and deployed a cloud-native SaaS architecture that processes millions of event payloads daily. The pipeline is structured as follows:

• Event Ingestion Pipeline: Lightweight Javascript tracking pixel pushes client-side traffic data via load balancers to a FastAPI collector.
• Queue & Stream Broker: Message broker pipelines Kafka and Redis buffer event traffic streams to guarantee lossless delivery.
• ClickHouse Database: Column-oriented database structure maintains highly indexed analytics events for instantaneous reporting queries.
• Headless Worker Clusters: Playwright and Puppeteer worker threads orchestrate deep-crawls and run automated script executions.