The term “AI-native” has entered the fund management vocabulary, but most uses of it describe something less radical than it sounds. Understanding the structural difference matters for allocators evaluating the next generation of systematic managers.
AI-Assisted Versus AI-Native
An AI-assisted investment manager uses machine learning tools to do parts of the job better. Portfolio managers screen ideas faster. Analysts process earnings call transcripts at scale. Risk teams run more scenarios. The humans still make decisions — they are just better-equipped to make them.
An AI-native investment manager is built differently. The architecture starts from a different premise: that human judgment in the investment decision itself is not the bottleneck to be improved, but the failure mode to be removed.
The distinction is not one of degree. It is structural.
At AvB Capital, there is no portfolio manager who can override the signal. There is no investment committee that signs off on a trade. The research engine discovers, tests, and validates signals. The execution system sizes and deploys them. The governance layer — position limits, drawdown floors, kill conditions — is encoded as constraints the system cannot route around, not policy documents that humans can interpret flexibly.
Why Remove the Human Layer?
The case rests on empirical data, not ideology.
89.5% of active equity managers underperform their benchmark over 15 years, net of fees. Across all 22 equity categories in the SPIVA Year-End 2024 report, zero beat their benchmark consistently. This is not a run of bad luck — it is a structural pattern stable across decades and geographies.
The mechanism was documented precisely in Akbas et al. (Journal of Finance, 2023): studying 783 institutional portfolios, buying decisions showed skill, but selling decisions lost 50 basis points per month relative to random selection. There was no learning effect. The failure was consistent across all experience levels.
The structural disadvantage of human judgment in investment management is not that humans are unintelligent. It is that human cognition under pressure, at speed, with incomplete and contradictory information, is systematically outclassed by the volume and speed of modern markets. The solution is not a smarter human. The solution is a different architecture.
What the Architecture Achieves
An AI-native manager runs a continuous research loop, evaluating far more hypotheses than any human team can assess and rejecting the vast majority before they touch capital. What survives that process is not what a senior analyst is most convicted on — it is what the evidence, rigorously tested out-of-sample, demonstrates to be real.
The discipline of rejection is as important as the quality of research. Anyone can generate signals from data. The edge is in the fraction that pass honest, out-of-sample validation.
Once live, the system does not hold positions because a portfolio manager has publicly defended them. It does not delay exits because the market narrative feels uncertain. It executes according to rules that were defined before the trade, not revised during it.
The Compounding Advantage
The reason AI-native architecture matters is not just that it removes human error. It is that the learning is structural.
A human portfolio manager improves slowly — constrained by cognitive bandwidth, organisational politics, and career risk that makes underperformance relative to benchmark professionally dangerous. A system that calibrates from every outcome compounds its research advantage continuously.
This is the core argument for why AI-native management is not a better version of active management but a different category: the rate of improvement scales with data throughput, not headcount.
AvB Capital is an AI-native systematic manager based in London. Track record and strategy documentation are available to qualified investors on request.
Frequently Asked Questions
What makes an investment manager AI-native versus AI-assisted?
An AI-assisted manager uses AI tools — screening, analytics, research support — but retains human portfolio managers who make the final investment decision. An AI-native manager eliminates that discretionary layer entirely. The signal generation, portfolio construction, and execution all run through deterministic systems with no human override between research output and trade.
Is an AI-native investment manager the same as a quantitative hedge fund?
There is overlap, but the distinction matters. Traditional quant funds built brilliant human organisations that use quantitative tools. The humans still oversee, approve, and occasionally override. An AI-native manager starts from the premise that the human oversight layer is itself the source of underperformance, and designs around its absence.
How does an AI-native manager handle market regimes it hasn't seen before?
Through systematic validation and hard risk controls. Every signal is tested across multiple market regimes before deployment. Position limits and drawdown floors are encoded as constraints the system cannot override — not policy documents that humans can set aside.
Can an AI-native manager outperform in all market conditions?
No strategy outperforms in all conditions, and any manager claiming otherwise should be avoided. The structural advantage of removing human judgment is that the system does not chase recent performance, does not hold positions because of sunk cost or career risk, and does not make decisions differently on a bad day versus a good one.
What is the minimum track record needed to evaluate an AI-native manager?
Track record is necessary but not sufficient. The key evaluation is the research process itself: how many hypotheses were tested, what was the rejection rate, is the validation framework documented, are out-of-sample results separated from in-sample? A short live track record with a rigorous documented process is more valuable than a long track record with no reproducible methodology.