By Duggan Matthews, Chief Investment Officer, Marriott Investment Managers
Part 2 in our series on how Marriott's investment process is evolving to enhance outcome predictability in a more complex, faster-moving global environment.
In Part 1, we described how Marriott rebuilt its research architecture from first principles. We designed an investment process that doesn't just bolt-on AI, but is built for AI, and is engineered to improve as AI improves.
Our thesis was simple: in an environment where analytical capability is rapidly becoming commoditised, the critical distinction is not the tools you use but rather in the architecture through which they are applied – this is where the edge lies.
Rapid advances in AI capability over the past two weeks have offered a clear illustration of why this distinction matters for asset managers aiming to improve the quality and consistency of analysis without the disruption of continual reintegration. The developments discussed below show how this plays out in practice.
On 16 April, Anthropic released Claude Opus 4.7 – its most capable generally available model to date. For investment professionals, several of the advances are directly relevant:
For a firm that has bolted AI onto an existing workflow, each of these improvements is both an integration challenge and competitive threat. The old model is breaking. Process improvement no longer scales with headcount. The traditional sources of competitive advantage, team size, coverage breadth, manual monitoring depth, are being systematically eroded. And the erosion accelerates with every model release.
For Marriott however, the experience is different. Because we broke down our process into clearly defined analytical components, the improvements in Opus 4.7 flow directly into our existing architecture. Financial statement analysis becomes more precise. Monitoring becomes more reliable. The quality of structured assessments improves. No redesign is required. The process simply gets better.
This is what compounding looks like in practice. There’s no new project, no new implementation cycle. We simply get a measurable improvement because the architecture was built to receive it.
Less than two weeks before the release of Opus 4.7, Anthropic announced Claude Mythos Preview, a model it described as the most capable it has ever built. Mythos was not released to the general public. Instead, it was made available to a small group of technology firms through an initiative called Project Glasswing, focused initially on cybersecurity. The reason for the restricted release was telling: the model's reasoning and analytical capabilities were considered too advanced for unrestricted deployment without additional safeguards.
For the investment industry, Mythos is not yet a working tool. But it is a signal that deserves serious attention. If Opus 4.7 represents a meaningful step forward in analytical depth and reliability, Mythos suggests that the next step will be significantly larger. When models of that capability become broadly available, the gap between firms with architecture designed to absorb them and firms still adapting to the last generation will widen considerably.
This is the dynamic we anticipated when we designed our process. We were not thinking of any single model, but of AI's trajectory. The rate of improvement is not slowing. It is accelerating. An investment process built around a fixed set of tools will require continuous reinvention to keep pace. A process built around system architecture will compound with each advance automatically.
Until now, the investment industry has tended to evaluate technology in discrete cycles: adopt, implement, stabilise, repeat. That made sense when tools changed slowly. It does not hold when the underlying technology improves as rapidly as AI is doing, with each generation arriving materially stronger than the last.
Firms that treat each new AI model as a separate adoption decision will find themselves in a permanent state of catching up. The promised efficiency gains are consumed by the cost of continuous adaptation.
The alternative is to build once, correctly – to design an architecture that benefits from every advance without being disrupted by it.
This is what Marriott has built. And every new release, whether it is Opus 4.7 this week or Mythos in the months ahead, is confirmation that the approach is sound.
The pace of change is not a risk to be managed. For firms with the right architecture, it is the advantage itself. What does this mean for our investors? It means the quality of portfolio management they receive improves at the same pace as AI itself.
Note: Whilst this article was being finalised, OpenAI released their latest and most capable model.
Local
Offshore