Superweek 2026 – Day Four

AI Isn’t the Shortcut. It’s the Amplifier.

By Day Four, something interesting happened.

The tone shifted from “what is AI doing to us?” to“what are we actually doing with it?”

The hype had cooled. The pattern had emerged.

AI is not replacing analytics. It is exposing weak analytics.

Day Four was about that exposure.

Inconceivable Questions Aren’t the Problem

Krista Seiden – From Inconceivable to Intentional

Krista opened with a familiar tension: clients are asking questions that feel inconceivable to answer.

But here’s the uncomfortable inversion:

If clients can conceive of the question, it’s not inconceivable.

It’s inconceivable to us because we understand the hazards of the data. They don’t.

That’s not a criticism. It’s an opportunity.

Client imagination is often a prioritisation engine. If they’re asking it, it matters.

Another strong reminder:

Never assume high traffic means high success.

High traffic often just means high noise.

Clean beats volume. Every time.

Krista also touched on AI-driven anomaly detection; which, after Siavash’s deep dive yesterday, now feels incomplete unless paired with post-detection triage and activation logic.

The real practical takeaway:

If you feed an LLM messy, inconsistent, poorly named events, it will not magically infer structure.

We used to standardise event names for ML models. Now we must standardise for LLM reasoning.

Clean taxonomy is no longer a “nice to have”. It’s prompt engineering for your data warehouse.

One of the best points of the talk:

Individual session analysis has never scaled and never will.

The future is:

• Identify dominant journey clusters

• Train models on aggregated behavioural paths

• Surface divergence

• Apply segmentation to replay only where necessary

And a subtle but important comment:

It is permissible to favour a platform for what it doesn’t do.

Restraint is a feature.

When AI Creates a Million Ads

Juliana Jackson – The Haystack Paradox

Juliana tackled the other side of the AI coin: creative explosion.

Generative AI now enables infinite personalisation at infinite scale.

Which creates infinite mediocrity if misused.

The real discipline here isn’t production. It’s creative intelligence.

The problem:

• Creative assets explode in cardinality

• Measurement struggles to keep up

• Attribution remains linear while behaviour is not

Funnels assume linear journeys. They never were.

Juliana proposed aligning creative as a channel dimension; but cardinality quickly becomes a structural constraint. You must categorise to analyse. Which means assumptions. Which means prior knowledge.

This is where Matt Gershoff’s Day Three point on interactions becomes critical:

More variables ≠ better decisions.

Applying AI-driven micro-optimisation across every campaign variable is:

• Expensive

• Risky

• Dependent on solid foundations

Her most pragmatic guidance:

Never launch from zero knowledge. Run pre-flight checks. Monitor in-flight. Analyse post-flight qualitatively.

AI doesn’t remove the need for judgment. It makes mistakes at scale.

Use it in low-risk environments first. Fix the high-risk structural issues before automating marginal gains.

Data Activation Is the Point

Balázs Vajna – Getting Data to Earn Money

Balázs cut through dashboard culture.

Most organisations use data to report.

Very few use it to activate.

Activation requires:

• Systems integration

• Trigger logic

• Ownership

• Feedback loops

Data that sits in slides is theatre.

Data that changes spend, creative, pricing or audience allocation earns its keep.

This connected directly to Nicolas’ Day Three framework: direct activation (systems) and indirect activation (people).

Reporting is not value.Change is value.

Who Are You Really Working For?

Dan Truman – The Quiet Part Out Loud

This session landed differently because it wasn’t about tools.

It was about incentives.

Most analytics doesn’t fail because it’s wrong. It fails because it works for the wrong stakeholder.

The vendor.

The internal champion.

The budget owner.

The loudest voice.

Rarely the end user.

Rarely the long-term health of the business.

This week repeatedly surfaced a pattern:

Vendor-centric narratives dominate. User-centric thinking is harder. Business-centric thinking is rarer still.

The uncomfortable question is not “is this correct?”

It’s “who does this serve?”

Google Analytics: From Urchin to Probability

Timo Aden – A Look Into the Future

Timo delivered historical context.

UTM = Urchin Tracking Monitor. Urchin 5 in 2003 cost $5,000/month. Google bought it in 2005.

The evolution matters.

GA is no longer the product. BigQuery is the product. The UI is a preview layer.

In a consent-driven world, we no longer measure reality. We measure probability distributions.

Expect tighter integration with MMM solutions like Meridian. Expect GA to move closer to CDP territory.

And yes, the UI is too complex.

Education is now infrastructure.

Quality Data In, Intelligent AI Out

Peter Meyer

Peter reinforced what Krista began:

AI amplifies whatever you give it.

Garbage in, confident garbage out.

Data quality maturity is now directly tied to AI value extraction.

You cannot shortcut governance and expect intelligent output.

Day Four, Distilled

AI is not the hero of this story.

Structure is.

Taxonomy. Data quality. Signal design. Activation logic. Commercial clarity. Human oversight.

Across sessions, the pattern held:

• Clean > volume

• Foundations > scale

• Control > circumvention

• Activation > reporting

• Judgment > automation

By Day Four, Superweek felt less like a technology conference and more like an accountability audit.

AI didn’t lower the bar.

It raised it.

Day Five awaits.