The next platform won't fix what platforms can't fix.

A wave of peer-reviewed research is making a buying case for AI-personalised learning. Studies from 2024 and 2025, including a 2025 survey of 268 university instructors, report measurable improvements in learner performance and engagement when machine learning systems adapt content, pacing, and difficulty to the individual. The pattern holds across cultural contexts and platform generations. Most CHROs will see a vendor deck citing some version of it before the end of the quarter.

A longer arc of evidence sits underneath that wave, and it has been saying something different for a decade. A systematic review of LMS use in higher education, alongside studies applying the DeLone and McLean information systems success model and the LPOMR (leadership, planning, organisation, management, resources) framework, all converge on the same shape: less than a third of whether a learning platform lands is the platform itself. The rest is sponsor cover, who owns the boundary between the tool and the work, change management that survives a reorganisation, and what gets rewarded in practice when no one is reporting on it. That finding has been stable across institutions, regions, and three generations of platform.

The two streams of research read like the same conversation. They are doing different jobs. The personalisation studies report what happens when an individual learner uses a tool that is already configured, sponsored, and integrated into a workflow of real value. The implementation studies report whether any of that orchestration ever exists. Most buying decisions read the first answer to a question only the second answer can settle.

The implementation literature names the same elements every time. A sponsor with stake in the outcome, not sign-off on the budget. Named owners for the boundary between the platform and the workflow it is meant to change, with explicit escalation when the two meet and one of them gives. Change management framed as design, not as communications. Governance written down for what knowledge the model can reach, who can revoke that access, and what evidence would justify revoking it. Capability signals the executive team will accept that do not come from the vendor's own dashboard.

Three asks make the difference between repeating the pattern and breaking it. A diagnosis, in writing, of why the last platform under-delivered: a specific account of which conditions were missing and which leaders did not move when the system asked them to, rather than a story about the vendor. A small set of non-platform capability signals you can stand behind, of the kind a workforce data strategy tends to land on. A sponsor pair (CHRO and CTO/CIO) committed in writing to the changes the platform will surface, not only to the platform itself.

This is the work a capability diagnostic does in six weeks. Less about the tool and more about the ecosystem the tool is dropped into: what the organisation already teaches by default through incentives, calendar pressure, and who gets believed in a meeting; where the proxies are lying; where the previous procurement lost altitude and why nobody quite said so at the time. The output is plain-language, because the next conversation it has to survive is in a leadership forum with L&D out of the room.

For senior L&D leaders the new personalisation evidence sharpens the strategic mandate. AI-personalised learning is a real tool with real lift, in the right ecosystem. The market will reward vendors who can show strong individual-outcome data on adaptive pathways. It will not, on its own, reward L&D functions willing to argue that no platform produces capability in conditions a platform never landed in. That argument is where the function earns its place in the executive room. If that is your remit and you want a direct conversation about what those conditions look like in your organisation, submit a brief and we will tell you in writing whether we are the right fit.