Who wins as value moves up the tech stack? A golden era for emerging B2C platforms

We are now two and a half years into the AI infrastructure buildout, catalysed by the launch of ChatGPT. We have a good degree of conviction over who we believe the winners are in the AI infrastructure layer of the new technology stack. However, as the cost of AI systems comes down and we move from enterprises experimenting with AI to deploying AI inference at scale, value moves up the stack. The key question for investors at the moment is who the winners will be in the application layer. These are the companies building on top of accelerated compute and delivering AI-infused productivity to us as users. The battle to become the user interface for AI is fierce and disruption is a certainty, not a likelihood.

Companies that are well positioned have a handful of critical features in common. Firstly, distribution advantage matters. We are seeing this play out in a number of B2C platforms like Meta, Spotify, Netflix and Shopify. These companies already own the consumer, they live in our pockets, and they sit on a body of unique, proprietary, real-time data – essentially a goldmine for AI use cases.

This cohort of companies moved quickly to adopt AI internally, leading to shrinking cost bases, and they are now focused on infusing AI across their platforms to create new revenue streams, leading to enhanced user engagement which turbocharges the network effects they enjoy.

Take Spotify as an example: in the past two years, operating expenses as a percentage of revenue has declined from c.30% to c.20%. The company has reduced headcount in favour of automation and is now printing cash. On top of this leaner architecture, AI features are being layered to draw in new customers: AI-fueled daylist and playlists are increasing customer searches by 2000%, for example, which in turn deepens user engagement and stickiness – Spotify’s AI DJ customers return on average 25% more. This feeds better data on 640 million users, which further improves recommendations and the platform experience.

Crucially, these companies are innovating on a unified software platform. How a platform is architected really matters for AI in order to be able to ingest, store and utilise data at massive scale at speed. This cannot be said for many legacy enterprise software companies.

We view Spotify as an earlier-stage Netflix, but who are the earlier-stage Spotifys of this world we are backing? Mercado Libre, Doximity, Affirm, Lemonade and Upstart to name a few. These platforms span sectors, from healthcare to financial services to ecommerce, but they, like Netflix and Spotify et al, possess a unique distribution footprint, unique proprietary real-time datasets, and a unified software platform.

Mercado Libre – Latam’s digital ecosystem

Mercado Libre has transformed its dominant Latin American e-commerce platform into a multifaceted digital ecosystem – integrating marketplace, payments, logistics and lending. Picture Amazon ten years ago, but on a larger scale and with a uniquely expansive logistics footprint. With more than 100 million unique buyers on its platform, Mercado Libre’s expansive distribution network generates a vast, real-time data stream on consumer purchasing behaviour and merchant inventory – far surpassing that of its fragmented rivals and giving the company a lasting competitive edge.

The company leverages this proprietary data to personalise the shopping experience and inform financial offerings: AI-driven features now automate product review summaries, tailor notifications for abandoned carts, and equip sellers with dynamic pricing and instant answers to customer questions. These AI enhancements have measurably improved the customer journey, driving conversion and retention: unique active users on the company’s marketplace grew 25% year-on-year (yoy) last quarter, with Fintech active users growing 21% yoy.

Such growth demonstrates the power of a unified platform in which marketplace transactions flow seamlessly into fintech services – most notably Mercado Pago digital wallets and credit programmes – creating a self-reinforcing flywheel. In effect, Mercado Libre follows Amazon’s playbook yet supercharges it with AI: its extensive reach and integrated services generate data-network effects that allow machine-learning models to refine product recommendations, optimise logistics and sharpen credit underwriting in real time, pushing value creation ever further up the stack towards the application layer.

Affirm – Real-time finance at the point of sale

For a growing swathe of Americans, revolving credit-card debt has become the default way to make ends meet – stacking households with punitive interest charges and paving the way for alternatives like Affirm. Almost 37% of US adults carry a credit-card balance – an astonishing share given today’s punitive interest rates. Affirm offers a remedy: shoppers receive their goods immediately and repay the cost over three to 18 months, interest-free.

While the service began with big-ticket purchases, it is now increasingly used for smaller items as well. Tackling this widespread consumer pain-point has fueled Affirm’s rapid growth over the past decade, bringing the company to scale. Today it operates one of the largest “buy now, pay later” (BNPL) networks, embedded in 300,000-plus merchants – from Shopify to Amazon – and serving more than 21 million active customers. Affirm exemplifies the power of AI-driven fintech, delivering intelligent credit decisions at the precise moment of purchase.

This widespread distribution gives Affirm two key assets: a trusted brand at checkout and a constant flow of transaction-level data. Every time a shopper opts to split a payment, Affirm’s platform performs an instant, AI-driven credit decision – assessing the individual’s risk based on thousands of data points beyond a static FICO score. In fact, Affirm’s approach “underwrites every extension of credit” in real time using not only the consumer’s Affirm history but also their up-to-the-moment indebtedness elsewhere. This dynamic use of proprietary, real-time data differentiates Affirm from legacy credit card models and shifts value to the application layer (where the lending decision happens) rather than the credit infrastructure beneath.

Critically, Affirm has invested in AI and machine learning to sharpen its edge. The company’s fraud detection and loan underwriting now employ transformer-based AI models, boosting accuracy and reducing losses. Internally, AI automation parses contracts and streamlines merchant onboarding, allowing Affirm to scale operations without proportional headcount. These AI-driven productivity gains translate into better economics and customer experience – fewer manual reviews, faster approvals, and more personalized financing terms. As a result, Affirm can offer many 0% APR promotions (subsidised by merchants to drive sales) while still expanding its user base (active customers grew 23% yoy last quarter) with disciplined risk management.

Lemonade – the insurance platform built for AI from day one

Few consumer experiences are as universally frustrating as filing an insurance claim, a fact reflected in the industry’s dismal net-promoter scores. Lemonade sets out to reverse that reputation by offering first-rate service at a fair price and, in the process, completely re-imagining what insurance can be. Born in the cloud, the company treats insurance as a digital, AI-native consumer product: tightly integrated data and machine-learning models replace brokers and paperwork, while chatbots do the heavy lifting. Its unified software platform spans renters, homeowners, pet, life and now motor policies – avoiding the siloed systems that hobble legacy insurers – and every touchpoint, from sign-up to claim, runs through proprietary bots (“AI Maya” for sales, “AI Jim” for claims). Lemonade thus shows how application-layer disruptors can out-manoeuvre infrastructure-heavy incumbents by delivering a faster, friendlier and markedly cheaper customer journey.

The AI-driven productivity gains are striking: c.98% of claims now begin with AI Jim handling the first notice of loss, and c.40% of claims are resolved end-to-end with no human intervention at all. Lemonade’s AI swiftly verifies coverage, analyses the claim via computer vision and natural language processing (e.g. scanning videos of damage), and instantly approves and pays out legitimate claims – once even in three seconds, setting a world record. This level of automation slashes overhead costs and delightfully speeds up service – the company’s use of AI has cut claims processing time by a whopping 90%.

Deep data integration is key: companies that haven’t architected themselves for AI, with data entwined across functions, will be hard to glean the type of deep insights Lemonade has built its business upon. In Lemonade’s case, every new customer (now nearly 1.8 million, skewing young and digital-native) provides behavioral data that continuously trains its AI models. This growing data trove creates a flywheel: more users and claims improve the algorithms, which in turn enable better underwriting and user experience. As the pool of data increases, Lemonade’s risk pricing is becoming more accurate and profitable, evidenced by a 30% increase in premium per customer as the company expands into higher- value policies.

Upstart – AI lending platform building the foundation model for credit

First introduced in 1989, the FICO score gave American lenders a standardised yardstick for judging creditworthiness. It was a breakthrough in its day, yet the lending industry has since fallen behind the data and AI revolution, leaving millions of would-be borrowers underserved and unable to obtain affordable credit. Upstart set out 15 years ago to solve that problem. Its AI-powered lending platform provides granular credit insight to institutions ranging from banks to motor dealerships, enabling them to write better loans at sharper prices.

Rather than build its own branch network, Upstart sells a lending-as-a-service model to more than 500 banks and credit unions – a distribution strategy that has delivered nationwide reach. Through these partners and a direct-to-consumer funnel, Upstart ingests a torrent of proprietary data: application details, repayment histories and macro-economic signals far richer than any single FICO score. The impact is striking. Upstart’s personal-loan model evaluates over 2,500 variables per applicant and is trained on 86 million repayment events, updating itself daily as new data arrive. Independent studies show the system can approve 43% more borrowers at the same loss rate, or cut APRs by a third while holding approvals steady – a step-change in both efficiency and financial inclusion, achieved through superior application-layer intelligence.

Upstart’s unified platform automates the entire lending workflow – marketing, application, identity check, risk assessment, and even compliance – demonstrating huge AI-driven productivity gains. Over 92% of Upstart-powered loans are fully automated, with no human intervention in the decision. Customers get an instant, all-digital experience (often receiving funds the next day), while bank partners benefit from lower operating costs and faster loan growth. Upstart’s AI also adapts in real time to macroeconomic changes via a proprietary index and adjusts lending criteria accordingly, something static credit models can’t do. By moving the brains of lending into the cloud, Upstart has shifted where value accrues: instead of banks gaining advantage from owning branch networks or securitization pipelines, they now seek out Upstart’s application-layer solution to stay competitive.

Doximity – The “LinkedIn for Doctors” making practitioners more efficient

Often described as “LinkedIn for doctors,” Doximity already reaches more than 80 % of all US physicians – a distribution beach-head that is almost impossible to replicate. On top of that trusted network, the company has stitched together a single, HIPAA[i]-compliant workflow suite (newsfeed, tele-health video, secure fax, scheduling) so every click feeds one unified data lake. The result is a platform-level moat: users stay because their colleagues, patients and day-to-day tools are all in the same place, while life-science marketers and hospital systems keep spending because that’s where clinicians actually engage.

Over the past year, Doximity has pushed value even further up the stack by embedding generative AI directly into those workflows. Its ChatGPT-powered Doximity GPT drafts referral letters, insurance appeals and patient education in seconds, and is offered free to every verified clinician – driving rapid adoption. Usage numbers tell the story: in Q1 2025, over 620 000 unique prescribers used the company’s workflow (AI, tele-health, fax, scheduling), up five-fold year-on-year.

That deeper engagement is translating into economics: trailing-12-month net revenue retention sits at 114 % (rising to 121 % for the top 20 pharma customers), showing that the more AI features clinicians tap, the more advertisers and hospitals expand their spend. By combining unrivalled reach, proprietary real-time data and tightly integrated generative-AI tooling, Doximity is shifting healthcare value from legacy IT rails to the intelligent application layer – exactly the pattern we expect to define the winners as AI matures.

_{[i] HIPAA =  Health Insurance Portability and Accountability Act}