Manifesto

# An agent cannot operate a system *it cannot observe.*

![Josele Pérez, CEO and co-founder of Trackingplan](/images/699487c44d72b2ee9745211c_josele-perez.avif)**Josele Pérez**, CEO & Co-founder4 min read

For years, digital analytics teams have been dealing with a familiar problem: **data can be technically available without being trustworthy.**

Tracking breaks silently. Implementations drift. Consent changes what gets collected. Platforms behave differently than expected. And analysts are left figuring out whether a change in the data reflects a real change in the business — or something that broke underneath it.

This has always been a problem.

But AI is about to make analytics *worse* before it makes it better.

But AI is about to make analytics *worse* before it makes it better. The first wave of AI in analytics is making analysts more productive — faster queries, faster reports, faster analysis.

Useful, yes.

The bottleneck was never ~~asking faster.~~It was knowing whether the answer *could be trusted.*

AI makes it dramatically easier to produce answers from data. It can reason over enormous amounts of data, connect signals, explain patterns, and produce answers in seconds. But it doesn’t automatically make those answers correct. The bottleneck was never asking faster; it was knowing whether the answer could be trusted.

A wrong analytical answer doesn’t look broken. *It looks like an answer.*

A wrong analytical answer doesn’t look broken. It looks like an answer. It has numbers. Context. An explanation. It can sound intelligent. And increasingly, it can trigger an action.

That changes the cost of bad data.

A broken tracking implementation used to distort a dashboard or a report. Now, the same broken signal can flow into attribution, automated bidding, forecasts, AI models, copilots, and agents — making decisions on behalf of your business.

one broken signal**purchase event, missing on iOS since the last release**

-   nowAttributiondecides
-   nowAutomated biddingspends
-   nowForecastsplans
-   nowAI modelslearns
-   nowCopilotsrecommends
-   nowAgentsacts

That’s why data quality moves from being a technical hygiene problem to **a strategic prerequisite for AI.**

## On where the industry looks

Yet most data-quality tooling starts downstream, where data is already structured, accessible, and easy to query. By the time a number reaches the warehouse, the decisions that determine whether that number is trustworthy have already happened.

Web · Apps · Server-side · Tag Manager

RequestDataLayerPixelsCAPI payloadConsent stateGTM release

Where trust is decided

→

providers

GA4MetaGoogle AdsAdobeAmplitudeYour CDP

→

dashboards & warehouse

LookerBigQuerySnowflakeTableau

Where most tooling starts

TrackingplanDownstream data-quality tools

**Trackingplan has spent years sitting exactly where data breaks:**

the hit · the dataLayer · the SDK · the pixel · the CAPI payload · the consent state · the GTM release

**That’s where the truth of a number is actually decided, before it becomes the number everyone relies on.**

Observability was only the foundation.Agency is the *destination.*

We believe trust cannot be added to data at the end of the pipeline.

We believe the industry needs to move closer to the source.

We believe no one can fully understand the quality of a number without understanding how that number came to exist.

Because an agent cannot reliably operate a system it cannot observe. Observability was only the foundation; agency is the destination.

**That is the layer Trackingplan is building.**

![Josele Pérez](/images/699487c44d72b2ee9745211c_josele-perez.avif)

**Josele Pérez** CEO & Co-founder, Trackingplan[Read more about Josele →](/team/josele-perez)

## Your turn.  
Try a question.

Your digital analytics agent.

[Book a demo](/demo)