A complex case lands, perhaps an odd early retirement from a closed section, or a member whose benefits were adjusted after a bulk transfer in the nineties, and nobody reaches for the procedures manual. They walk across the office and ask Sue.

To be fair, Sue is brilliant. She knows why that category of member is treated differently. She remembers the trustee decision behind it, the correspondence that never quite made it into the benefit specification, the quirk in the old system that meant the calculation had to be done a particular way. The case gets resolved, the member gets paid, everyone moves on.

Then Sue retires.

Pensions administration runs on accumulated memory far more than the industry likes to admit. The trust deed and rules tell you what the benefits are. They rarely tell you how the scheme has actually been run. Historic interpretations. Discretionary practice built up over decades. The reasoning behind manual adjustments. Data corrections made in 2006 and the assumptions sitting underneath them. Workarounds created because a legacy system could not cope with a particular benefit design.

Very little of this is captured neatly in one place. Much of it only becomes visible when a difficult case arises, or when the person who understood it is no longer at their desk.

Here is the uncomfortable part. A genuinely knowledgeable administrator can make an incomplete process look complete. They know which questions to ask, which records to check, where the exceptions sit. The service performs well, the reports look green, and everyone reasonably concludes that all is fine.

I think we need to be more honest about the difference between having experienced people and being dependent on particular people. The first is a strength, probably the biggest strength any administration service has. The second is a structural risk, and it rarely shows up in a stewardship report. Nothing about a well-handled case tells you whether the knowledge behind it exists anywhere other than in one person's head.

This weakness tends to stay hidden for years and then appear at exactly the wrong moment. Administration transitions. System migrations. Benefit specification exercises. Buy-in and buyout preparation, where an insurer wants evidence for every practice and every assumption. Calculation automation, where "we have always done it that way" is not something a developer can code.

I have sat in specification meetings where a single line of a calculation held up weeks of work because nobody could say why it existed. Not because anyone had done anything wrong. The reasoning was probably perfectly sound at the time. It just lived in someone's memory, and that someone left in 2011.

When an experienced administrator resigns or retires, the instinct is to book a few handover sessions and hope for the best. A couple of conversations and a spreadsheet of open cases will not capture twenty years of judgement. It never does.

What actually works is slower and less glamorous. Documented benefit specifications that get reviewed and approved, not filed and forgotten. Decision and interpretation logs, so the reasoning survives even when the person does not. Records of known data issues and the assumptions behind past corrections. Complex cases used deliberately as training material rather than quietly resolved by the same safe pair of hands. Peer review that forces knowledge out into the open. And someone clearly owning the job of keeping all of it current, because a specification that was accurate in 2019 is a liability dressed up as an asset.

The aim is not to move knowledge from one head to another. It is to embed it in the service itself.

Trustees do not need to manage their administrator's workforce. But they do need confidence that the service does not quietly depend on two or three individuals. So ask.

Which parts of our administration rely heavily on individual knowledge? Are our benefit practices formally documented, and have we approved them? What happens when key people are unavailable, or leave? Is succession risk something our administrator reports on, or something we have simply never discussed?

None of this needs to be confrontational. It is the same oversight trustees already apply to data quality and cyber resilience, extended to the knowledge that holds everything else together.

There is a temptation to assume AI will solve this. It will help, probably more than people expect. Modern tools can surface documents, spot inconsistencies between records and make guidance far easier to find.

But technology can only work with what has been captured. It can make knowledge easier to find. It cannot recover knowledge that was never recorded. No system, no matter how clever, can reconstruct an undocumented trustee decision from 1994.

We do need to be careful here, because none of this is an argument that experienced administrators matter less. If anything, the opposite. Some of the best-run schemes still lean heavily on one or two exceptional people. The difference is that they know it, and they are doing something about it.

Doing something about it also means investing in the next generation. Time for coaching. Real exposure to complex work rather than a diet of routine cases. Technical career paths that reward deep scheme knowledge instead of pushing everyone towards management. People need to understand why a process works the way it does, not just the steps involved. That is how judgement gets built, and judgement is the thing we are actually trying to pass on.

The strongest administration services are not the ones that always know who to ask. They are the ones where the answer does not depend on one person being available. Sue has earned her retirement. The scheme should barely notice she’s gone. That, oddly enough, is the greatest compliment an administrator can be paid.

Dan Taylor, Client Director – Trafalgar House