Bamboo spends 5-7 years building its root system underground before it grows 90 feet in a single season. To anyone watching, it looks like nothing is happening. To the bamboo, everything critical is happening. The businesses planting data trees now are in the root phase—and most of their competitors don’t even know a race has started.
Your tracking infrastructure choice determines your AI capabilities in 3-5 years. GTM gives you a reporting feed. A server-side pipeline—one that streams raw first-party events directly into BigQuery—gives you a growing data asset. The gap between those two outcomes compounds every month.
Why GTM Is Standing Between You and AI Readiness
The AI analytics market is growing at 34.6% CAGR and is projected to reach $4.3 billion by 2025. AI is becoming the primary intelligence layer for commerce—personalizing experiences, predicting buying intent, optimizing ad spend automatically. Every major platform is moving in this direction.
Here’s the problem: 80% of AI projects fail, and 70% of those failures trace directly back to poor data quality (Gartner and IBM, 2023). The most common reason isn’t bad AI models. It’s bad data feeding them.
And GTM produces exactly the kind of data AI cannot work with.
GTM’s sandboxed JavaScript runs on ECMAScript 5.1—a standard from 2011. AI code tools consistently fail to generate working GTM code because they were not trained on it. GTM’s architecture of tags and triggers is locally coherent but cross-functionally incompatible: it cannot provide the consistent schemas, clean hierarchies, and unified semantics that AI automation requires.
Translation: the data architecture that GTM produces is structurally incompatible with the AI era—not because of any single limitation, but because of how the whole system was designed.
GTM was built in 2012 to simplify JavaScript snippet management. AI was not part of the design brief. It shows.
What GTM Actually Gives You vs. What AI Needs
When a conversion fires in a GTM-managed setup, that event travels through GA4’s processing pipeline before it reaches you. By the time it’s available for analysis, it’s been sampled, aggregated, and reformatted to GA4’s standards—not yours.
What you receive is a report. What AI needs is raw data.
- GTM → GA4 export: Processed, sampled, delayed, formatted by Google. You cannot train a model on it.
- Server-side pipeline → BigQuery: Raw, real-time, structured by you. Every event, full fidelity, immediately available for AI use.
One bad data stream breaks the entire AI pilot. Fragmented architecture—the kind that a typical GTM stack produces across platforms—is the single most common cause of failed AI implementations.
You cannot build an AI capability on top of data you don’t fully own, in a format you didn’t define.
You may be interested in: Your WooCommerce AI Recommendation Plugin Is Trained on the Wrong Data
The Data Tree in Action: LMBK Surf House
LMBK Surf House is a WordPress-based hospitality business. Every booking search, package view, inquiry, and reservation event flows through a server-side pipeline directly into BigQuery as raw, structured, first-party data.
That’s 1.6 million events per month—and growing.
Each month adds another layer to what is becoming a compound data asset: booking patterns by season, guest behaviour flows, conversion path analysis, abandonment signals. In three years, LMBK will have a dataset that can train a predictive booking model, power personalised recommendations, and identify high-value guest profiles before they even make a first inquiry.
A GTM-locked competitor in the same space has GA4 exports. Processed, sampled, formatted by Google. They cannot build the same thing—not with what they have.
The gap isn’t a technical edge. It’s a compounding moat. And it’s being built right now, one event at a time.
WordPress.com announced official AI-powered plugin development support in December 2025, with AI as a fundamental technology layer. The platform serving 43.5% of all websites (W3Techs, 2024) is moving toward AI-native operations. The businesses with owned event data in BigQuery are positioned for that future. The businesses dependent on GTM reports are not.
You may be interested in: GTM in 2026: A Platform Designed Before AI That Was Not Built to Evolve
The Infrastructure Decision Is Made Now, Not Later
There is a timing dimension to data assets that doesn’t apply to software licenses or marketing tools. You can switch your email platform next year and lose nothing. You cannot switch your tracking infrastructure next year and recover the data you didn’t collect this year.
The AI models that will power commerce in 2028 will be trained on data collected between 2024 and 2026. The businesses that started collecting raw first-party events now will have training data. The businesses that waited won’t.
This is not a future problem. It’s a now problem disguised as a future one.
The question isn’t whether AI will matter to your business. The question is whether you’ll have the data to use it when it does.
How to Actually Plant a Data Tree
Planting a data tree means one thing technically: raw first-party events flowing into BigQuery from a server-side infrastructure you control.
Not GA4 exports. Not GTM with a BigQuery connector. Raw events, structured by your schema, streaming in real-time from your own infrastructure.
The architecture looks like this:
- A WordPress event occurs (purchase, booking, form submission, product view)
- A lightweight plugin captures the event data from WordPress hooks
- Events batch and send via API to a first-party server on your subdomain
- The server validates, enriches, and routes simultaneously to all destinations—including BigQuery via Streaming Insert API
- Raw event data lands in BigQuery in real-time, structured, owned by you
No GTM. No GA4 middleman. No Google reformatting your data before it reaches you.
Transmute Engine™ is a dedicated Node.js server that runs first-party on your subdomain—data.yoursite.com, not a third-party server. The inPIPE WordPress plugin captures events and sends them to Transmute Engine via API. Transmute Engine then routes simultaneously to GA4, Facebook CAPI, Google Ads, BigQuery, Klaviyo, and more—streaming raw events into BigQuery as structured first-party data at $89–$259/month. No GTM expertise required. No developer needed to set it up.
Key Takeaways
- The AI analytics market is growing at 34.6% CAGR. AI commerce tools are coming. The data infrastructure gap between businesses is widening now.
- 80% of AI projects fail due to poor data quality. GTM produces processed exports, not raw events—structurally incompatible with AI training and automation.
- A Data Tree is a compound asset. 1.6M events/month in BigQuery becomes a personalisation engine, a predictive model, and a competitive moat over 3-5 years.
- The infrastructure decision cannot be made retroactively. Data not collected now cannot be recovered later. This is a timing problem, not a technology problem.
- WordPress is moving toward AI-native operations. Businesses with owned BigQuery data are positioned for that shift. GTM-locked businesses are not.
A Data Tree is Seresa’s term for a growing, owned first-party event dataset—typically stored in BigQuery. Like a tree, a data asset requires time to develop before it delivers value: the events you collect today become training data, personalization signals, and predictive models in 3-5 years. Every month of clean event data makes the asset more valuable.
GTM’s sandboxed JavaScript runs on ECMAScript 5.1—a 2011 standard. AI tools consistently fail to generate working GTM code because they were not trained on it. More fundamentally, GTM delivers processed GA4 exports, not raw first-party events. AI systems require consistent schemas, clean hierarchies, and unified semantics that GTM’s architecture cannot reliably produce.
AI readiness requires raw first-party events flowing into a data warehouse like BigQuery directly from a server-side pipeline—not through GA4’s processing layer. The data must be structured by your schema, owned by you, and collected in real-time. A dedicated server-side pipeline streaming to BigQuery is the correct architecture.
Data assets compound over time. Businesses collecting raw events into BigQuery today will have 3-5 years of training data when AI commerce tools mature. Businesses that waited cannot recover that historical data retroactively. The infrastructure decision is permanent in its consequences—data not collected now is lost permanently.
No. GTM offers a BigQuery export, but it delivers processed GA4 data—not raw first-party events. A server-side pipeline can stream raw events directly to BigQuery via the Streaming Insert API, bypassing GTM entirely. This produces clean, schema-controlled, first-party event data that is actually usable for AI applications.
When AI becomes the primary commerce intelligence layer, the businesses that built data trees will have something to work with. Plant them now. Start at seresa.io.
