Matt Hudson, Founder of BILDIT, has spent enough time inside large retail organisations to know where mobile app growth actually stalls. It is rarely the product. The app works; the engineering is competent. The problem is that nobody in the building has a strong enough reason to care whether anyone uses it.
Matt made this argument on the latest episode of Branch‘s How I Grew This, and it is worth taking seriously because it reframes a question most retailers treat as technical into one that is fundamentally about organisational design. When bonuses are tied to channel-level performance, when email teams own their open rates, and web teams own their traffic, the app becomes everyone’s second priority.
If the mobile app does not improve return on ad spend, nobody will do anything with it. The implication runs further than it first appears. Attribution that cannot connect app engagement to measurable outcomes means the app remains peripheral in planning conversations regardless of how good the product is. Store teams will not promote something that does not help their numbers. Email teams will not route into an experience that does not show up in their reporting. The app sits in a corner accumulating respectful neglect.
The channel conflict problem
The web versus app split that persists in many retail organisations is less a strategic disagreement than an accounting artefact. Separate budgets and separate attribution models create teams with an incentive to defend their own traffic, and that defence regularly produces worse outcomes for the customer.
However, the customer does not know or care which internal team owns the surface they are using. If the app is the fastest path to conversion on mobile, routing users away from it introduces friction at precisely the moment the brand needs to remove it.
The uncomfortable operational consequence is that internal competition can be structurally built into the organisation. Solving it requires changing what gets measured and what gets rewarded, which is often harder than shipping a feature.
When an app is actually worth building
For smaller retailers still deciding whether a mobile app makes sense, several signals matter: catalogue depth, physical retail presence, loyalty mechanics. But the variable that matters most is purchase frequency.
A customer who returns regularly benefits from an app in ways that are concrete and measurable — easy login flow, reduced checkout friction, faster time to purchase. This is not a demographic question about younger vs older shoppers. Loyal customers use whatever surface creates the least resistance, and for high-frequency retail relationships, a well-built app is usually that surface.
The returning customer segment is where most of the revenue actually lives. It is smaller in volume, concentrated in value, and more sensitive to friction than acquisition metrics tend to suggest. An app that removes one unnecessary step from a repeat purchase pays for itself in ways that are difficult to see in channel-level reporting but show up clearly in cohort analysis.
AI as a distribution question
Summarisation in search results is already reducing click-through to retail sites, and the question of whether a brand’s products appear accurately in AI-generated responses is now an operational one.
AI systems are not trying to find a relevant link. They are trying to produce a response that appears reliable and correct, which is a meaningfully different objective that rewards different kinds of content.
Backlink-heavy SEO instincts are poorly matched to this environment. What reads as trustworthy to a model doing inference is closer to what reads as trustworthy to a careful human reader: forums, FAQs, human-validated sources, content with a clear structure and a specific answer to a specific question.
Product descriptions written for keyword density fail that test. FAQ-style content tends to pass it, partly because conversational queries are themselves structured as questions, and a model evaluating correctness finds it easier to work with content that mirrors that structure.
Automated agents crawling or browsing a site behave differently depending on their purpose — some pull raw HTML, some simulate a browser, some attempt to mimic human sessions. Understanding which is which, and what each is doing, is the baseline for making any informed response to AI-driven traffic rather than simply being subject to it.
Data as the underlying constraint
Product data has become a distribution asset. Descriptions, attributes, structured metadata, Q&A content determine whether a product appears correctly in AI responses, not just how it ranks in traditional search. Most retail organisations still treat this as a housekeeping problem. However, it is closer to a channel problem, and it compounds until the gap between what a brand sells and what AI systems can accurately represent becomes wide enough to matter in revenue terms.
The broader pattern is that friction accumulates wherever organisational incentives, technical architecture, and data quality diverge from how customers actually behave. Retailers tend to discover this one surface at a time, and each discovery tends to reveal the same underlying misalignment wearing a different set of clothes.
Tune in below to listen to the episode, or check out all the episodes here.
