If GA4 exports to BigQuery for you, you already collect the raw material for a predictive preloading model. Most teams miss this because someone set up the export for another reason, often so finance could build a custom revenue dashboard, and no one went back to ask what else the dataset could do.

What’s sitting in that export

A standard GA4-to-BigQuery export includes, for each event, the page, timestamp, session ID, traffic source, device category, and enough sequencing information to rebuild the full path a visitor took through your site. This is a behavioral dataset. It records what tens of thousands of people did on your pages, in your navigation structure, arriving from your channels.

A next-page prediction model needs exactly this training data, from your own visitors. You need no synthetic data, no third-party panel, and no industry benchmark.

Why it sits untouched

In most organizations the data has one job: feeding a handful of dashboards that someone checks once a month, if that. Someone built the pipeline, and no one built a second use for it.

“Train a predictive model on our navigation data” sounds like a data science project with a data science team’s timeline. If the export exists and is clean enough, a usable next-page prediction model takes a few weeks of focused work, not multiple quarters. Foresight rests on that premise: you already paid to collect this data, so putting it to work should cost little.

What putting it to work looks like

Foresight starts by checking that your GA4-to-BigQuery export exists and holds enough history. If it does, the data that produces a monthly PDF for finance also trains the model that decides what to preload for the next visitor who matches a pattern your past visitors showed. We collect nothing new and ask nothing new of your users. You use the dataset twice instead of once.

If the export doesn’t exist yet, it becomes step one. It’s a prerequisite, and once configured it takes about a week to produce the first useful batch of data.

The cost of not doing this

You paid to collect this data, and it does one job when it could do two. That gap separates what your BigQuery export costs to maintain from what it returns.

An audit answers in five minutes whether your setup has what it needs.