My houseplants were looking a little bit ragged this weekend. I’ve never had a green thumb, but I watered them and they had been doing well for a few years. Turns out my house cleaner has been doing the real work to make sure they thrive and she’s been out of town for a month.
I saw something working and never questioned whether there was something happening that I didn’t see.
Organisations have two versions of how they operate. There is the one on the org chart: the process manual, the workflow diagram, the answer leadership gives when a board member asks how work gets done. And there is the one that actually runs the business day to day: the workaround a coordinator built because the official system doesn’t do what the job needs, the judgement call one person makes from memory because nobody ever wrote down the rule, the informal relationship between two team leads that quietly resolves a handoff the process was never designed to handle.
Leadership teams can usually describe the first mode. Most have never seen the second and some don’t even realise it exists.
That second model is not a documentation gap you can close. It is infrastructure. It carries load. Remove the person who holds it, or the workaround that compensates for it, and the work does not slow down. It stops.
Why it stays hidden
Leaders tend to assume that if something matters, it shows up in reporting. But that’s not always true. Reporting usually tells you what happened after the fact. It is not built to show the detail behind those results. A dashboard can show that a project delivered on time without ever revealing that it delivered on time because one analyst spent three days a week manually reconciling data that two disconnected systems were supposed to share.
Ask a leadership team how confident they are in their operating model, and most will describe the version on the chart. Ask the people doing the work how it actually gets done, and a different picture appears. There’s judgement nobody formalised, coordination nobody assigned, ownership of the seams between functions that exists only because someone decided, unasked, to bridge the gap.
This is not a story about broken organisations. Well-run businesses depend on this layer constantly. The workaround is often a sign of competence, not failure. Someone saw a gap and closed it without waiting for permission. The problem is that leadership is making decisions, and now increasingly building AI investment cases, on the assumption that these informal layers don’t exist.
Where it trips you up
On its own, invisible infrastructure can be a scaling risk. It works while the business is small enough that everyone still knows where the load-bearing points are. It stops working as the organisation grows, because growth adds people who were never told where the informal system lives, adds handoffs that route around structures nobody documented, and adds distance between the leaders making decisions and the workarounds those decisions depend on. Eventually someone who held a piece of that infrastructure leaves, and a part of the business that leadership believed was systemised turns out to have been running on one person’s memory.
AI raises the stakes, because AI is the first technology that depends on this layer being visible.
AI systems need data and context to function. If the real way work happens is undocumented, fragmented across disconnected tools, or held informally in someone’s head, there is no reliable foundation for AI to learn from or act through. This produces one of two outcomes, and both are expensive.
The first is that the AI investment surfaces the gap only after it has been made. The programme reaches the point where it needs connected, trustworthy data, and discovers that the data either doesn’t exist in usable form or lives in a workaround nobody accounted for in the business case. At that point the organisation is rebuilding infrastructure mid-programme, on an extended timeline, at a cost nobody budgeted for.
The second is worse. The organisation deploys AI anyway, on top of the fragmented reality. The workaround gets automated along with everything else. A judgement call that one experienced person quietly adjusted case by case gets encoded as a fixed rule, at scale, without anyone deciding that was the intention. Invisible work does not become visible when AI touches it. It becomes invisible faster, and harder to unwind, because a system is now enforcing it. Decisions are made that no one can explain and might have serious business impact.
Automating invisible infrastructure shouldn’t be the goal in every case anyway. Sometimes the right answer is for work to deliberately stay in the hands of people, with leadership having full visibility and awareness that this is how things work.
What to hold onto here
AI does not create this risk. As with many other things, it reveals a condition that was already there and makes the cost of ignoring it a lot higher. Every organisation has been running on some version of this second, undocumented model for as long as it has existed. What changes is the tolerance for not knowing.
Before an AI programme gets designed, it is worth asking a more foundational question than most business cases ask: if you had to show, in detail, how work actually happens in your organisation today, could you?