Game-changer or damp squib? AI in Private Credit

Reflections from the ACI COO Conference 

Kris Wilson, Chief Operating Technology Officer | Channel Capital 

I was on a panel at the ACI COO Conference recently alongside a group of senior operations and technology leaders. The question put to us was blunt and deliberately so: is AI a gamechanger for private credit, or is it another damp squib, a technology that generates noise without changing much that matters? 

The sub-questions were equally pointed. How much has AI meaningfully changed the way private credit firms operate? Which areas could benefit most? What operational challenges does it present? And will it genuinely reshape the industry, or join the graveyard of financial services buzzwords? 

I came away thinking those are exactly the right questions to be asking, and that the honest answers are more nuanced – and more interesting – than the hype on either side suggests. Here is how I see it. 

AI Agents Are Moving from Concept to Consequence 

The narrative around AI has shifted. We are well past the stage of debating whether AI belongs in financial services. The question now is how to deploy it with intent. 

Agentic AI, systems that plan, act, and complete multi-step tasks autonomously rather than simply responding to prompts, is where the most interesting and most demanding work is happening. For alternative asset managers, the use cases are compelling: deal screening, portfolio monitoring, investor reporting, regulatory document processing. These are not hypothetical. They are being built now. 

At Channel Capital, we think about this through the lens of what we call Digital Full-Time Equivalents, or DFTEs. These are AI agents designed to take on structured, repeatable workflows within a governed framework. The promise is significant operational leverage. The requirement is that the infrastructure around those agents, the data, the controls, the human oversight, is built to the same standard as the agents themselves. 

We are currently running small-scale experiments with multi-agent orchestration frameworks to understand what this looks like in practice. One tool we are evaluating is Paperclip, an open-source platform that structures AI agents into an organisational hierarchy with org charts, goal alignment, budget controls, and governance gates. What is interesting about this approach is the mental model it enforces. Rather than managing a collection of disconnected automations, you are operating a team, with reporting lines, defined roles, and a board-level override at every stage. For a regulated, governance-conscious environment like ours, that framing matters. 

The logical endpoint of this trajectory is what manufacturing has long called the dark factory; a facility so fully automated it can operate without the lights on, because there are no humans present who need them. Applied to financial services operations, the concept is worth taking seriously. Back-office functions running continuously and autonomously, with human oversight reserved for exceptions and strategic decisions rather than execution. We are not there yet, and for most firms the journey will be measured in years. But the firms that get there fastest will be the ones building the right foundations today. 

Governance Is Not a Brake on Innovation 

One of the strongest points to come out of the panel, and one I feel strongly about, is that governance frameworks for AI are not a constraint on ambition. They are what make ambitious deployment possible. 

Without clear ownership, audit trails, and policy, AI sits in pilots indefinitely. Too risky to scale, too valuable to abandon. With the right governance structure, you can move with confidence. 

The regulatory and standards landscape is also maturing quickly, and firms need to be across it. Three frameworks are particularly relevant for UK alternative asset managers right now. 

The EU AI Act is now in force and applies to UK-domiciled firms more than many realise. If you deploy AI systems that affect EU investors, counterparties, or data subjects, and most managers will, the Act’s risk-based requirements around transparency, human oversight, and high-risk system classification will apply to some part of your operations. This is not a distant compliance horizon. 

The FCA’s principles on AI make clear that the regulator expects firms to explain and justify AI-driven decisions, maintain appropriate human oversight, and ensure AI does not introduce new forms of bias or consumer harm. For asset managers, this has direct implications for how AI is used in investment decision support, client communications, and complaints handling. 

ISO/IEC 42001, the international standard for AI management systems, provides a structured framework for governing AI at an organisational level, covering risk assessment, accountability, transparency, and continual improvement. For firms already certified to ISO 27001, the two standards are designed to be complementary. The controls, governance structures, and audit disciplines you have already built translate directly, and 42001 can be layered in without rebuilding from scratch. 

At Channel Capital, we are actively assessing our position against all three. Our view is that firms which treat AI governance as a compliance obligation will always be behind the curve. Those that treat it as a genuine organisational competence will be better placed to deploy AI ambitiously and responsibly at the same time. 

The Economics of AI Are Not What They Appear 

There is a dimension to this that does not get enough airtime in conversations about AI strategy, and it is one that anyone with a finance background should be paying close attention to. 

The major frontier AI companies – the providers behind the models that most enterprise AI tools are built on – are, without exception, deeply loss-making. They are burning through venture capital at a scale that has few precedents in the technology industry. The current pricing that firms are accessing, whether through direct API usage or through SaaS products built on top of these models, reflects a market in which the underlying infrastructure is being heavily subsidised by investors. 

That will not last indefinitely. At some point, these companies will need to achieve sustainable economics. When they do, prices will rise; potentially significantly. The workflow you have built; the productivity gain you have measured; the business case you have signed off; all of it was modelled at a price point that may not hold. 

The implication is straightforward. You need a rigorous handle on the ROI of every AI tool and workflow you are running, and you need to be constantly reassessing it. Not just “is this useful?” but “is this useful enough to remain justifiable if the cost doubles?” That is a different question, and one that most organisations are not yet asking. 

This is not a reason to avoid AI investment. It is a reason to be disciplined about it, to know exactly what value you are getting, where it is coming from, and how sensitive that value is to changes in cost. Firms that cannot answer those questions clearly are carrying more commercial risk than they realise. 

Practical Adoption Is the Hardest Part 

The most candid moments of the panel were around adoption:; specifically, the gap between what AI can do in a demo and what it does when deployed across a real organisation with real people. 

A few things came through clearly, and they map closely to what we have experienced ourselves. 

Deployment and adoption are not the same thing. This is the most underestimated challenge in enterprise AI rollouts. Putting a tool in front of people is straightforward. Getting them to use it well, consistently, and with confidence is an entirely different undertaking. At Channel Capital, we have found that small, focused user groups – ; people who are genuinely curious, sharing what works, and building practical knowledge together –, are far more effective at driving real adoption than any top-down rollout. Peer credibility travels further than policy. 

People need to trust it before they will use it. Staff who are uncertain about AI; whether it will make mistakes, whether it affects their role, whether it is appropriate to use; will find reasons not to engage. Building that trust through clear communication, practical training, and real examples of AI working well in context is not optional. It is the adoption strategy. 

The individual user is responsible for the output. This is a point we are deliberate about at Channel Capital. AI is a tool, not a colleague. The person who uses it owns the result. That accountability cannot be delegated to the model. 

Hallucinations are a feature, not a bug, and AI can be very confidently wrong. Large language models are designed to produce fluent, coherent output. They are not designed to know when they do not know something. Errors arrive without any signal that something is wrong. A hallucinated figure, a misattributed source, a subtly incorrect legal summary, all can read exactly like a correct one. This is not a flaw that will be patched away. It is a characteristic of how these systems work, and it demands a corresponding discipline from every user. 

The burden shifts rather than disappears. AI reduces the mechanical effort involved in many tasks; drafting; summarising; structuring. But it does not eliminate the cognitive load; it relocates it. The work moves to the front and the back: upfront design of the prompt, the context, and the constraints; and post-output review to verify accuracy and fitness for purpose. Teams that treat AI as a shortcut around that discipline will eventually be caught out. 

AI is a powerful abstraction, and all abstractions leak. When web development went mainstream, an entire generation of developers built applications without needing to understand how a CPU scheduler or network I/O worked at the operating system level. That was fine, until it was not. When an application hit a CPU-bound bottleneck or an I/O contention issue, the abstraction leaked, and the developers who could not get under the hood were suddenly exposed. 

AI represents a new abstraction layer of the same order of magnitude, and the same principle applies. A developer who relies entirely on AI-generated code without being able to read, reason about, and debug it is one production incident away from a serious problem. An analyst who accepts an AI-generated financial model without understanding the underlying methodology is carrying a risk they may not be able to see. The abstraction will leak at some point, in some context, and when it does, the question is whether your team has the foundational capability to respond. The answer is not to avoid AI. It is to be deliberate about not letting core domain expertise atrophy in its wake. 

Start narrow, prove value, then expand. The firms seeing the best results are not trying to transform everything at once. They are identifying one or two high-value workflows, deploying carefully, measuring outcomes, and building internal confidence before scaling. Speed comes from discipline. 

Quality of output depends on quality of input. In alternative assets, where data can be fragmented across deal documents, fund structures, and counterparty systems, getting your data estate in order is a prerequisite, not a follow-on. 

Game-Changer or Damp Squib? 

So, what is the verdict? My answer is that the question itself contains a false choice. 

AI will be a damp squib for firms that treat it as a product to deploy rather than a capability to build. Drop a tool into an unprepared organisation; without the data foundations, the governance framework, the user trust, or the domain expertise to interrogate its outputs; and the results will be underwhelming. At some point, even damaging. 

For firms that approach it deliberately, by; building the right infrastructure, investing in adoption, maintaining clear accountability, and staying close enough to the underlying work to know when the abstraction is failing them,; AI is a genuine gamechanger. Not because it eliminates work, but because it changes what skilled people spend their time on. That is a meaningful shift in how a private credit firm can operate, and one with real competitive implications over a five-to-ten-year horizon. 

Most firms, including us, are still experimenting more than transforming. But the gap between those building the right foundations now and those waiting for the technology to mature will be difficult to close once it opens. 

At Channel Capital, we are committed to being on the right side of that gap. We will continue to share our thinking as the work develops. 

If you were at the conference and would like to continue the conversation, or if you are working through similar questions at your own firm, I would be glad to connect.