By Bryan Dougherty, CTO, Arcesium
A new year brings new opportunities – and firms investing in private markets are entering 2026 with a focus on moving from AI experimentation to deployment. The global private credit market is estimated at roughly $2.28 trillion AUM in 2025, and is expected to nearly double to $4.5 trillion by 2030[i], reflecting continued momentum across corporate, asset-backed, real estate, and infrastructure lending. Amid this growth, firms are confronting both increased opportunity and complexity across the sector. Many are understandably looking to AI to drive efficiency, scale operations, and unlock insights faster.
But firms are discovering an uncomfortable truth – AI doesn’t solve foundational operational issues: it magnifies them.
This is the AI paradox playing out across the industry. Firms see enormous automation potential, yet in practice, they’re finding that AI can also amplify weaknesses in data quality, fragmented workflows, governance gaps, and undocumented processes that rely on institutional knowledge held by a small number of people within an organization. For institutional investment firms to get AI right in private markets, the most important question is not, “which model should we use?”, but instead, whether the data and processes that AI is running on are trustworthy.”
Why private markets are uniquely exposed
Private credit and private markets have always had a data challenge, but the last decade of growth has intensified it. Unlike public markets – where standardization and structured data are more prevalent – private markets operate on bespoke agreements, tailored loan terms, and custom documentation. That means the most important information, rather than being captured in standardized fields, often lives in unstructured formats – embedded in PDFs, emails, and other free-form documents, such as loan notices, amendments, credit agreements, multi-manager model agreements, LP statements, and investor communications. Extracting, standardizing, and validating that information is still largely manual, which may create inconsistencies and operational risk at scale.
This complexity is also increasing as firms expand across regions. In an industry survey, 72% of UK/EU lenders cited Europe as a top growth opportunity over the next 12 months[ii], highlighting the continued acceleration of activity across jurisdictions. The result isn’t just a volume problem, but also a coordination problem. As firms grow and strategies proliferate, they’re trying to scale operations, reporting, and client service without scaling headcount at the same rate. While generative AI is transformative when it comes to extracting meaning from text-heavy materials, AI does not eliminate the need for clean foundations. Instead, it makes them more urgent.
The AI magnifying glass
A useful way to think about AI is that it functions like a magnifying glass – while it accelerates insights, it also increases the visibility and impact of existing problems. In private markets, where critical information often lives across credit agreements, amendments, notices, and investor communications, that magnification effect can be especially pronounced. A firm that has strong processes and reliable data will enjoy amplified capability through AI, as it can help teams move faster, automate manual work, improve decision support, and personalize client communications at scale.
However, for firms with messy workflows, inconsistent data, unclear ownership, and weak controls, AI doesn’t magically turn a “B process” into an “A process.” It simply runs more “B-level” processes – faster, more broadly, and with higher downstream risk. That’s why firms that layer AI on top of fragmented data architectures and siloed manual workflows often get stuck. They may build impressive prototypes – for example, extracting terms from loan documents or drafting investor updates – but struggle to move into production without exposing risk across reporting, compliance, and operational workflows.
In regulated, high-stakes environments – where mistakes are costly and auditability matters – this isn’t a theoretical concern. In private credit, for example, if AI-driven covenant monitoring, borrowing base calculations, or investor reporting can’t be traced back to trusted sources, validated, and governed, it will remain a proof-of-concept rather than a strategic advantage.
AI’s benefit curve is lagging for good reason
The lagging benefit curve of AI is partly a symptom of expectations being misaligned with reality. Some believe AI should produce instant transformation, but technology shifts of this magnitude rarely work that way. A useful historical analogy is the early adoption of electricity in factories. Initially, factories used electricity to replicate existing workflows. The real productivity boom came later, once factories were redesigned around what electricity made possible.
AI is similar. Many early enterprise use cases resemble AI-enhanced search – a modern replacement for internal searches, where employees ask a chat interface instead of navigating folders or systems. In private markets, this often shows up first as teams searching across credit agreements, side letters, amendments, LP reports, and investor communications to answer questions that would otherwise require manual review. That remains a foundational use case as it delivers value quickly with relatively low risk.
The next wave, with agents executing tasks across systems, is more ambitious, but also more difficult. Workflow transformation is inherently hard, especially for critical private market operations – where firms cannot tolerate disruption in areas like NAV calculations, cash reconciliations, or investor reporting. Most firms won’t – and shouldn’t – change core workflows overnight. Real AI value will be achieved iteratively through controlled rollouts, data governance layers, and gradual expansion across workflows and asset classes.
Agentic AI raises the stakes
As the industry moves toward agentic AI, the stakes increase. The most consequential change in 2026 is not simply better chat interfaces, but systems that can plan, coordinate, and act across workflows. For that promise to become real, AI must function more like a hive mind: coordinating across tools, data platforms, and domains. Regardless of the technical approach, agents are only as reliable as the data and controls behind them.
In private markets, the implications are tangible. If AI is responsible for reconciling cash, validating covenant requirements, or drafting investor reporting, firms must be confident that it is pulling from the right sources, that the data is accurate and current, and that outputs can be reviewed, approved, and audited. This is where many AI strategies fall short. It isn’t enough to deploy a powerful model. AI must be taught where the truth lives – across systems and documents – while also aligning on what “truth” means for portfolio data, fund reporting, and client communication.
Unstructured data isn’t the only challenge
One of the more interesting contradictions in the AI conversation is that the technology is strongest where finance is least standardized: language. AI is highly capable with text, but the private markets space also depends heavily on structured numerical data – positions, valuations, exposures, and cash flows – where accuracy and reconciliation are critical. Many general-purpose language models still struggle with complex numerical reasoning unless carefully designed into workflows.
This creates a key insight for institutional investment firms. The differentiator isn’t who has access to the best base model, but in how well a firm builds the data foundation, retrieval strategy, and domain-specific controls around private markets workflows. AI success isn’t a model competition – it’s an implementation competition.
How firms can accelerate time to value
Accelerating AI time-to-value starts with focusing on workflows rather than demos and choosing high-friction operational areas where success can be measured. In private markets, those areas often include covenant monitoring, onboarding and deal documentation, cash and fee reconciliation, and investor communication and reporting. The goal is to remove conditions that the AI paradox exposes – fragmented workflows, inconsistent data, and processes that rely on institutional knowledge only select team members possess.
Firms should map workflows end-to-end, because without understanding a process, it can’t be transformed. That mapping should include all pathways, including the exceptions and judgement calls often handled by those few individuals on a team, such as resolving discrepancies between notices and internal systems, or applying bespoke terms embedded in agreements. If workflows aren’t clearly defined and repeatable, AI will simply automate inconsistency – scaling the very friction firms are trying to remove.
Data curation should be treated as a central AI activity. Organizing data, consolidating it across systems, and ensuring it is interoperable and governed is what makes AI reliable, scalable, and safe. Just as importantly, it reduces reliance on institutional knowledge by making key terms, data definitions, sources of truth, and other key decisions explicit. Finally, governance should be designed as an enabler rather than a blocker. Strong access controls, lineage, review mechanisms, and auditability don’t slow innovation – they make it sustainable. When agents are introduced, they should be deployed in controlled ways, with clear oversight and measured expansion.
The AI winners will be those that build foundations first
AI is transformational, but it isn’t magic. Like any technical revolution, it can be both incredibly powerful and frustrating. The institutional investment firms best suited to reap the benefits of private markets will be the ones that treat data strategy, workflow discipline, and governance as strategic infrastructure.
AI doesn’t make organizations smarter. It gives them leverage. And leverage amplifies what already exists. If an organization’s foundations are strong, AI will amplify its capabilities. If they aren’t, AI will amplify risk.
—-
Bryan Dougherty is the Chief Technology Officer at Arcesium, overseeing platform development, infrastructure and security. Prior to the formation of Arcesium, Bryan was Head of Middle and Back Office Technology at the D. E. Shaw group for nine years.
—-
[i] Preqin, Private Markets in 2030 report on October 16, 2025. https://preqin.com/insights/research/reports/private-markets-2030?original_referrer=https%3A%2F%2Fwww.google.com%2F
[ii] Proskauer, Trends in Private Credit Risk report on February 3, 2025. https://prfirmpwwwcdn0001.azureedge.net/azstgacctpwwwct0001/uploads/0d10d4aad36f5f2587182ffd02bba395.pdf
