ExpertsFMDavid Vélez - Building Nubank and the Future of Global Banking - [Invest Like the Best, EP.495]
Invest Like the Best with Patrick O'Shaughnessy · October 13, 2026
5 Selected Lessons:
heuristic · 2 reasoning_pattern · 1 framework · 1 priority · 1
Bear case vs bull case framed around internationalization and AI adoption
Source
Invest Like the Best with Patrick O'Shaughnessy — David Vélez - Building Nubank and the Future of Global Banking - [Invest Like the Best, EP.495] (October 13, 2026) · at 1:06:05 ▸
When this applies
when articulating the explicit bear and bull scenarios for a scaling business to clarify what strategic bets determine the outcome
The takeaway
When forecasting a company's range of outcomes, identify the one or two strategic levers that simultaneously explain both the worst-case and best-case scenarios, and treat those as the top organizational priority.
Why it works
The bear case (market saturation in existing geographies, failure to internationalize, missed AI opportunity) and bull case (successful global expansion riding digitization tailwinds) share the same root variable — whether the company executes international expansion and AI adoption — making that the single highest-priority strategic lever.
One level up
Bear and bull cases that hinge on the same lever reveal the company's true highest-priority risk/opportunity, cutting through noise about secondary factors.
Where it stops holding
This stops working when bear and bull cases are driven by genuinely independent factors, in which case no single lever dominates the outcome.
Detail
hierarchy:
- Successful internationalization beyond Brazil/Mexico/Colombia (top priority)
- Full exploitation of AI-driven efficiency and new capabilities
- Continued domestic growth in saturating home markets (lowest priority/limited upside)
rationale: Vélez frames both failure and success scenarios as hinging on the same variable: whether Nubank internationalizes and captures AI-driven tailwinds, since domestic markets are nearing saturation (65% of Brazilian adults already customers).
In their words
We've saturated them completely. Growth declines. We haven't really been able to find any other growth alternatives from a customer perspective
Founder vs CEO as distinct, sometimes conflicting operating modes
Source
Invest Like the Best with Patrick O'Shaughnessy — David Vélez - Building Nubank and the Future of Global Banking - [Invest Like the Best, EP.495] (October 13, 2026) · at 54:58 ▸ · at 58:16 ▸
When this applies
when a founder-leader must decide how much to behave like a patient, systems-building manager versus an impatient, visionary disruptor as the company scales
The takeaway
When scaling a company, consciously separate founder energy (urgency, vision, risk tolerance) from CEO energy (patience, systems, management), and deliberately choose to lean toward founder mode especially during periods of technological disruption.
Why it works
Founders bring urgency, impatience, and willingness to attempt the 'impossible'; CEOs bring systems-thinking and patience needed to scale management — these traits can genuinely conflict, and in technology-shift environments the founder traits create more advantage.
One level up
Durable outperformance from founder-led companies, per this view, comes from preserving founder-mode behaviors even after formally becoming CEO, rather than fully transitioning to conventional management style.
Where it stops holding
This stops working when the organization is in a stable, non-disruptive environment where systematic management, not visionary urgency, is what prevents operational breakdown.
Detail
procedure:
- Recognize founder traits (impatience, big vision, risk tolerance) and CEO traits (patience, systems thinking) as distinct
- Resist fully converting to conventional CEO management style
- Deliberately preserve founder-mode behaviors especially during tech disruption
- Supplement with hires who bring complementary systems/management skills rather than replacing founder energy
rationale: Vélez found that hiring experienced professional managers to replace his founder instincts was the wrong move; what the organization needed was more of his original big-vision founder energy, not more conventional management.
In their words
I think the key, though, for me has been to realize that the CEO title itself and the founder title, you can redesign in different ways too
Dictator's dilemma — ground truth decays through management layers
Source
Invest Like the Best with Patrick O'Shaughnessy — David Vélez - Building Nubank and the Future of Global Banking - [Invest Like the Best, EP.495] (October 13, 2026) · at 47:36 ▸
When this applies
when a leader of any growing organization needs accurate real-time information but is separated from frontline reality by hierarchy
The takeaway
Reduce organizational layers and build direct information access (whether via culture of going directly to whoever has the answer, or via AI systems that surface ground-truth data) to counteract the natural distortion that hierarchy introduces.
Why it works
Every layer of management a message passes through introduces distortion, so the more layers between leadership and frontline work, the less accurate the leader's picture of reality becomes.
One level up
Power and scale naturally separate decision-makers from reality; deliberately engineered flat structures and information systems are required to counteract this, not optional nice-to-haves.
Where it stops holding
This stops working when flattening layers removes necessary coordination/mentorship structures that less experienced staff need, or when the leader lacks the bandwidth to triage direct information flow.
Detail
approach: Flatten the organization, build an AI-driven 'company brain' that proactively surfaces ground-truth data, and personally bypass chain-of-command to talk directly to whoever has the best information.
signals:
- organization naturally adds management layers as it scales to 'put out fires'
- information gets distorted as it passes through 10 layers
- CEO increasingly relies on filtered reports instead of direct data
rationale: Removing layers and building direct information access systems closes the gap between leadership perception and operational reality.
In their words
I like this idea of the dictator's dilemma that as someone gets more powerful at the top of an organization, it's harder for them to know what is real and what is true
AI-era moat combination: speed vs. scale/data/regulation
Source
Invest Like the Best with Patrick O'Shaughnessy — David Vélez - Building Nubank and the Future of Global Banking - [Invest Like the Best, EP.495] (October 13, 2026) · at 12:45 ▸
When this applies
when evaluating whether a nimble AI-native startup can displace a regulated, data-rich incumbent in a licensed industry
The takeaway
In regulated, trust-dependent, data-intensive industries, incumbents sit in a 'sweet spot' where they can adopt AI for speed while retaining licensing, data, and brand moats a new entrant can't replicate quickly — unlike pure software, where AI can let anyone rebuild a whole product fast.
Why it works
AI-native startups gain speed because they're unregulated and don't carry legacy systems, but regulated industries require licenses, trust, and large proprietary datasets that AI alone cannot manufacture quickly.
One level up
The size of AI's disruptive advantage for new entrants is inversely proportional to how much of the business's moat comes from regulation, trust, and proprietary historical data rather than pure software capability.
Where it stops holding
This stops working when the industry has no regulatory licensing barrier and data isn't a decisive advantage, allowing a well-funded, fast-moving entrant to fully leapfrog.
Detail
rule: Weigh an AI startup's speed advantage against the incumbent's regulatory license, data volume (e.g. 20-30k data points per customer over 13 years), brand trust, and proprietary foundation models.
threshold: Once a business requires banking licenses and multi-year regulatory approval, pure speed advantage caps out — you 'still cannot get ChatGPT to get you a banking license.'
rationale: Regulated, data-intensive businesses benefit most from AI because they already have the moat components a new entrant lacks, while gaining speed from AI themselves.
In their words
You can ask ChatGPT or Codex to rebuild an entire software strategy, but you still cannot really get ChatGPT to get you a banking license.
ROE differential as proof of a true cost-structure moat
Source
Invest Like the Best with Patrick O'Shaughnessy — David Vélez - Building Nubank and the Future of Global Banking - [Invest Like the Best, EP.495] (October 13, 2026) · at 36:22 ▸
When this applies
when deciding which financial metric best proves a company has built a genuine structural advantage versus just bought growth
The takeaway
To test whether growth is sustainable versus bought, examine return on equity relative to competitors alongside customer growth, NPS, and CAC; a superior ROE despite lower prices is the clearest signal of a structural moat.
Why it works
Charging fewer fees (lower revenue) while still achieving higher return on equity than fee-charging incumbents proves the cost base — not pricing power — is the source of advantage.
One level up
The most convincing proof of a durable competitive advantage is winning on both price AND profitability simultaneously against the incumbent.
Where it stops holding
This stops working when ROE is inflated by excessive leverage or accounting choices rather than true operating efficiency, or when growth is subsidized unsustainably by capital raises.
Detail
rule: Look at customer growth + NPS (found a real need), LTV/CAC (growing sustainably), and ROE relative to competitors (proof of true cost advantage).
threshold: Nubank's ROE is 2-3x that of incumbent banks despite charging no fees — this combination signals a genuine structural moat, not a subsidized land-grab.
rationale: Low fees plus high ROE together rule out the alternative explanation that growth is just bought via overspending on marketing or underpricing.
In their words
We don't charge a lot of fees. We literally charge no fee, so we have less revenue, but then we end up with a return on equity that is two to three X their return on equity. That basically means that you found a cost structure advantage.