Back 2018, Andy Carvell, Co-Founder and CEO at Phiture, was already asking app teams an uncomfortable question: are you working in silos? More than half the room at his Business of Apps Berlin back then session put their hands up. Seven years later, in 2025, he asked the same question in the same room. The hands went up again.

The difference is that in 2018, siloed growth was inefficient. Today, it is structurally incompatible with how digital marketing has evolved, and with AI rising to an ever more omnipresent force in the workplace, working in silos is uncompetitive, if not outright dangerous.

Andy has spent the past several years running a podcast called Brave New Digital World, talking with marketing leaders about how they are adapting to an ever-changing environment. Drawing on the points and ideas that keep surfacing across those conversations as well as his experience at Phiture’s steering wheel, the argument he made in Berlin was simply if provocative: the teams that succeed today are not just better at optimising channels; successful teams are scrapping them altogether.

At a glance

• Signal loss is permanent. Privacy regulation and platform opacity have degraded the targeting toolkit, and the teams still waiting for precision to return are falling behind those building systems that don’t depend on it.
• First-party data is only valuable when it travels. Behavioural and intent data collected inside the product should inform acquisition creative, app store positioning, paywall logic, and retention — not sit inside product analytics alone.
• Funnel learnings move in both directions. Insights from engaged users should shape upper-funnel messaging. Strong upper-funnel creative should feed back into store assets and onboarding. The system gets stronger when nothing stays siloed.
• AI changes the scale of what is testable, not just the speed. Running 500 experiments across 100 markets in a month is not a faster version of what a human team does — it is a different category of activity, with compounding organisational learning as the output.
• Cross-functional team design is now a growth variable. When insight, experimentation, and execution sit inside the same pod rather than across separate teams, learnings move faster and the whole funnel improves.

Why the old model stopped working

The erosion of targeting precision was not a temporary disruption; it is the new environment. Privacy regulation, the degradation of third-party identifiers, and the opacity of platform algorithms have collectively shifted what marketers can actually control. Lookalike audiences have become less reliable. Attribution on mobile has lost granularity.

Simply put, the signals that used to power precise audience targeting are weaker, and they are not coming back.

Web-to-app has been one intelligent response to this — a way to recover measurement precision that mobile attribution lost after iOS 14. But it is a workaround rather than a solution. The deeper shift is that growth can no longer be engineered primarily through external targeting. The edge has moved inward, toward what teams know directly about their own users.

“The winners are the teams that stop mourning the loss of how things used to be and start building integrated growth systems that don’t depend on precision targeting from third-party data.”

First-party data as the connective tissue

The behavioural, intent, and preference data collected inside a product is now the most valuable growth input most teams have — but (and that’s a big but) most teams are not using it properly. It sits in product analytics, informing product decisions, and rarely travels anywhere else.

That is the silo problem in its most common form. The data exists. It is just not accessible where it would do the most good.

There is big difference between collecting first-party data and federating it, as Andy explains. Collection is table stakes. Federation — making that data actionable across acquisition, app store positioning, paywall logic, and retention flows — is where the actual value sits.

An example should help make things clearer. For Audacy, a mobile radio app with heavy in-car listening, Phiture identified drive-time commuting as the dominant use case, confirmed it with in-app surveys, and built acquisition creatives around it. The result was a 7% uplift in app store conversion. For Adobe Acrobat in Japan, product data revealed that PDF editing was the primary behaviour among engaged, high-value users. A Custom Product Page built around that use case drove meaningful conversion increases on the store.

Neither of these required new data collection. The insight was already there. What changed was where it was applied.

Why engagement is your key to sustainable scale

Source: Business of Apps via YouTube

When the funnel learns from itself

The same logic runs in both directions. Lower-funnel data should shape upper-funnel decisions, but upper-funnel learnings should also feed back down.

When Meta creative performs unusually well for a particular message, that signal should inform app store screenshots and onboarding. When app store assets are redesigned and conversion improves, those assets are worth testing in paid channels.

Adobe went through exactly this cycle — a revamped creative strategy on the store produced results strong enough to justify deploying the same approach in performance marketing on Meta and Apple Search Ads.

The most striking example of cross-funnel data federation Andy describes comes from HBO Max. Their CRM team was sending push notifications based on what was new in the catalogue, personalised only to country level. Meanwhile, inside the product, a recommendation engine was already doing sophisticated user-level personalisation based on watch history. The two systems were not connected.

Phiture worked with HBO Max’s engineering team to expose the recommendation engine as an API endpoint, making it accessible to Braze (HBO Max’s push provider) at send time. Push notifications went from country-level targeting to individual-level personalisation overnight. Engagement improved significantly and the system is still running.

The same lesson again: the data and the algorithm were already there; the only thing missing was the connection.

“If you take anything away from this, it’s break down silos. Silos are bad.”

What AI actually changes about experimentation

The AI conversation in growth marketing tends toward the generic. Andy’s version of it is more specific, and more useful.

The real change AI introduces is not speed. It is the scale of the option space that becomes explorable. A manual team, however expert, can run a finite number of experiments in a month. An AI system running ASO optimisation across 100 markets simultaneously is not doing what a larger team would do — it is doing something a team of any size could not do.

Again, an example imposes itself. Phiture built Press Play, an automated ASO testing system for the Google Play Store, to test this directly. For a games company with multiple titles in soft launch, the system ran hundreds of experiments across markets, generated creative variants, and accumulated learnings. One finding — that icons with significantly more visual detail drove higher conversion for an aircraft combat game — would not have come from a human design team. The creative violated conventional design logic. It was cluttered. The art team objected. But users responded to it with a 40% conversion rate uplift, apparently because the density of detail communicated immediately that the game was not a flight simulator.

Want to speak at Business of Apps?

That learning then transferred across four other titles, each of which saw double-digit conversion improvements. The value was not just the optimisation of one asset. It was the organisational knowledge that accumulated from running at that pace and scale.

“The best ASO team in the world can’t actively optimise for 100 markets and run 500 experiments in a month. But you actually can do it with AI.”

The prerequisite is data infrastructure. Reliable event tracking, clear success metrics, and a reinforcement loop that gets smarter with each iteration. Without those, AI generates volume. With them, it compounds. Garbage in, garbage out — or in many cases, nothing in, nothing out.

How the best teams are now organised

Phiture restructured its own team around this argument. It used to have dedicated service lines for ASO, user acquisition, CRM, and paywall optimisation. Those teams were dissolved. In their place, cross-functional pods now hold all of those capabilities internally, each operating as a small, full-funnel growth unit.

The shift was not primarily about efficiency; rather, it was meant to remove the handoff problem. When insight from one part of the funnel has to travel across team boundaries to influence another, it usually doesn’t. When the same team owns the whole funnel, learnings move faster and the system gets smarter.

This is the operational structure Andy sees among the app businesses performing best right now. Small, agile teams running the full growth stack, with AI accelerating experimentation and first-party data providing the signal that connects acquisition to retention to monetisation.

Some of the most effective are launching marketing before the product is finished, using early conversion data to validate whether the product should exist at all before committing to a full build.

The silo model persists because it made sense when channels were independent and targeting was precise. Neither of those things is true anymore. The teams still organised around it are optimising a system that the market has already moved past.