Professor Gottschalg advocates for machine learning to redefine private capital decision-making

Professor Oliver Gottschalg argues that machine learning can materially strengthen decision-making across private capital markets, particularly in fund selection and secondary transactions, by supplying consistent, empirically tested allocation signals that human committees struggle to deliver at scale.

Gottschalg, a long‑standing scholar of private equity who teaches at HEC Paris and edits the quantitative volume Private Equity Mathematics, has combined academic rigour with practitioner tooling through Gottschalg Analytics. According to his analysis, modest, model‑driven reweighting of existing limited partner portfolios , rather than exotic strategies or privileged deal flow , would have produced substantial incremental value in conservative back‑tests using US public pension commitments. The gains, he contends, arise because algorithmic systems can reliably synthesise dozens of interacting variables over time in ways that humans, prone to storytelling and inconsistent calibration, do not. According to the report by Private Equity International, his work offers practical methods for benchmarking manager skill and portfolio construction that appeal to both GPs and LPs.

The structural features of private markets make them a better fit for systematic prediction than public markets, Gottschalg says. Transactions are executed infrequently, valuations are opaque and trade signals are slow to surface; these frictions limit the speed with which predictive edges are arbitraged away. That dynamic is especially pertinent in secondaries, where empirical evidence and market commentators point to persistent pricing anomalies. Forbes reports rapid growth in the secondaries market and stresses that the absence of standardised valuation protocols contributes to pricing inefficiency, while industry commentary has flagged instances where stakes bought at steep discounts were later marked up, underscoring opaque pricing dynamics.

Gottschalg’s approach reframes secondaries valuation away from granular, bottom‑up NAV adjustments and towards a simultaneous assessment of policy and capacity factors: how conservative a general partner’s valuation practice is, the realistic scope for future value creation, and whether strategy drift or team focus might compromise outcomes. Industry analysts note that the window for large discount capture is narrowing; MLC Asset Management documents a compression in average secondary discounts from around 10% in 2022 to approximately 5% by 2024, and PEI150’s reporting shows that GP‑led continuation vehicles accounted for a large share of 2024 GP‑led volume and frequently priced close to NAV. Those market trends mean models must be adaptive: opportunities remain, but margins for error have diminished and competition has shifted the landscape.

A central pillar of Gottschalg’s case is pragmatic treatment of imperfect data. Private equity datasets are fragmentary and delayed, yet his tests suggest that consistent modelling, conservative validation and repeated out‑of‑sample checks can produce robust, decision‑useful signals even when inputs are incomplete. That conclusion aligns with independent commentary questioning the sustainability of certain secondary pricing practices; AccreditedInsight highlights examples where aggressive discounting followed by material mark‑ups suggests room for more disciplined, data‑driven pricing frameworks.

The growing use of complex predictive models brings an interpretability trade‑off. Gottschalg acknowledges that highly expressive algorithms can be difficult to explain causally; trust, he proposes, should be earned through disciplined training, conservative assumptions and evidence of stability across cycles rather than narrative clarity. For allocators accustomed to qualitative judgements and reputation‑based choices, this represents a cultural shift: model outputs should guide positioning, while governance and calibration prevent systematic errors.

Human judgement, in his framing, remains indispensable , but in a constraining, not promotional, role. Practitioners should be empowered to override models to reduce expected returns where exogenous, credible information is missing or newly emergent, such as sudden governance breakdowns or confirmed key‑person departures, but not to ratchet allocations upward on the basis of relationships or gut conviction. He uses the analogy of advanced driver assistance: machines manage routine complexity while humans act to prevent catastrophic mistakes.

If adopted at scale, these practices would reshape incentives for both LPs and GPs. For limited partners, disciplined, algorithm‑assisted allocation challenges the notion that superior individual judgement is a durable source of outperformance; governance, calibration and process design may become the core skills. For general partners and secondary managers, increased systematisation raises questions about fee models and the defensibility of formerly opaque valuation and sourcing advantages. Market‑level effects could include lower liquidity costs and more standardised products, though commentators warn that rising capital flows into secondaries and tighter discounts are already compressing traditional arbitrage margins.

Gottschalg’s thesis is not a technological panacea but a testable proposition: when models are conservatively specified, back‑tested across regimes and integrated into robust governance, algorithmic decision support can outperform many human allocations in private markets. The balance of evidence from market reports and practitioner analysis suggests the opportunity is real but evolving , constrained by increased competition, narrowing discounts and the ongoing challenge of translating model signals into institutional practice.