Analytics feeds the AI
Giant Context, the platform I build, builds a customer's website. Making the website is only half of what a marketing platform should do. The other half is knowing whether it worked, whether people arrived, read what mattered, and did what the customer hoped, and using that to make the next version better. That second half runs entirely on measurement, which is why analytics is built into the platform instead of bolted onto it. The AI needs to see results, so the platform records them itself.
Why the machine needs to see
A platform that generates marketing and never checks how it did can only keep guessing. An intelligent one has to work the other way. Generate a page, measure how it performs, and let the next decision follow from the last result. That is a loop, and it only works if the results come back to the system that made the decisions.
The measurement is the part that feeds back. Which page converts, where the traffic comes from, what people actually read before they leave. That data is what the platform reasons from when it decides what to build or change next, rather than a report the customer reads and files away. Without it, the platform can act but cannot ever know whether acting helped, which means it cannot improve, which means it is generating in the dark.
The AI needs to see results, so the platform records them itself.
Generate
Make a page
Measure
Record how it does
Decide
Next move follows
Generate
Make a page
Measure
Record how it does
Decide
Next move follows
Why it has to be owned
That measurement cannot live in someone else's system.
The easy path is to let customers wire up Google Analytics and call it handled. It is less to build. But it puts the platform's most important signal, how the marketing is performing, inside a system the platform does not control. An AI that is supposed to read results and decide the next move cannot do that against a third-party dashboard a person signs into. The results would sit in a tool built for a human to read, not for a system to query, and the loop would never close.
So the results are first-party. The platform measures the sites it builds, in its own pipeline, in its own store, in a shape its own AI can query. A customer can still turn on Google Analytics if they want a second opinion, and the cookies the platform itself needs stay few enough for one honest consent. But the results the platform acts on are its own.
The other half of what the AI knows
This is the twin of a decision I have written about before. A customer's files tell the platform what a company is, its products, its voice, its facts. Analytics tells the platform how that company is doing, against real visitors. The files are the input. The results are the feedback. An automated marketer needs both, because knowing what a business is without knowing whether its marketing works leaves you unable to decide what to do next.
How it actually works
The mechanism is plain. In the second week of the project I built one event pipeline. A single event contract, one shape that anything in the system emits and one store it flows to. A visitor loading a customer's page is an event. So is a page publishing, an account signing up, an API call landing.
The event contract
event { type, // what happened who, // user, org, or anonymous visitor where, // which surface, which page, geography when, ...}Customer site analytics is that record, queried. Visitors are page-load events over time. The map is those events grouped by the geography already stamped on them. There was no separate analytics system to build, only a question to ask of a record the platform keeps anyway. And the pleasant side effect a customer feels is that their site has analytics the moment it is live, with nothing to install. That convenience is real, but it is a byproduct. The reason the record exists is so the platform can see.
How finished is this? The measurement is built and running. The part of the platform that reads the results and decides the next move from them is still ahead of me. But I built the analytics owned and general from the second week on purpose, because a platform that will need to learn from results cannot be built on results it does not hold. Measurement is far harder to add after the fact than to plan for, so I planned for it.
Half a product
Building the intelligence to make marketing is half a product. The other half is the intelligence to tell whether the marketing worked, and that half runs on data. If the data lives in a tool you rent, the platform can generate but never learn. So the data is the platform's own. The customer gets analytics for free as a result. The platform gets the ability to learn, which was the reason all along.
Have questions?
If you're interested in my work or Giant Context, contact me!