Cleaning is for the most part hands-on work at the location: fixed teams, changing sites, rosters that are adjusted daily to cope with illness, no-shows and seasonal peaks. The hours are not only in the cleaning itself, but also in planning it, in checking whether the work was carried out according to specification, in handling complaints and in reporting to the client. Whoever wins or loses a tender in this industry often wins or loses not on the price per hour, but on the question of whether the service delivery is demonstrably in order: no gaps in the roster, quick follow-up on complaints, verifiable quality. That is the environment in which AI is already taking over parts of the work today, and in which the comparison between providers is shifting as a result.
In part of the companies, software creates the basic roster, keeps track of staffing and absence, and draws up a replacement proposal. A planner assesses that proposal and approves or rejects it, with reasons. That is the second category: AI delivers the work, a human remains responsible for the decision. In other companies, planning is still done entirely by hand, with phone calls and a WhatsApp group. Both forms exist side by side, and the difference does not lie in the size of the company but in whether the underlying data — sites, hours, employees, availability — is already structured enough for a system to do something with it. Where that is not in order, planning remains human work, however much one might want to automate it.
The same pattern applies to quality control. Photos of a completed space, checklists that are automatically compared against the standard, deviations that are flagged before the client calls about them: that is work a system can prepare, after which a manager gives approval or intervenes. The physical cleaning itself remains in the third category: human work, and that will not change in the short term.
As long as planning and monitoring are largely manual work, a cleaning company wins on the dimensions related to staffing: how many planners there are, how experienced they are, how quickly they can close a gap. That was long the core of the proposition towards clients: we have the capacity to arrange this. The moment a competitor has that planning and monitoring largely run through a system with human oversight, the point on which the comparison lands shifts. Staffing of the back office is then no longer the distinguishing factor; speed of follow-up, consistency of quality control and the degree to which deviations are demonstrable become that instead.
This also affects how a tender is assessed. A client who asks about quality assurance gets from one company a description of procedures, and from another a dashboard with verified completion data. Both claims can be true; only the second is demonstrable at the moment the question is asked. Whoever claims in a tender to respond faster to complaints or to check more consistently can have that claim tested against underlying data, just as already happens elsewhere: see how that works with what you can read from a competitor's job postings about where they are focusing and with how you test whether distinctive capability really exists or is mainly claimed.
The shift does not happen at the same speed everywhere, and that has little to do with ambition. A company with many separate sites, short contracts and manual hour registration simply has less usable data to set up a system with than a company that has worked with fixed software for years. Comparable differences in pace can be seen in other sectors with a lot of planning work and physical execution, as can be read in what is changing in construction now that AI is taking over parts of the preparatory work and in how the installation industry deals with AI in planning and fault handling. The comparability does not lie in the sector, but in the structure of the work: a lot of execution on location, a lot of scheduling, a lot of dependence on human judgement in exceptions.
Importantly, this is not a staffing question being answered here. Whether and how a company adjusts its staffing as a result of what AI takes over falls under its own applicable legal requirements. What this page does say something about is which work lends itself to being taken over and which does not, and what that does to the comparison between providers.
The question of which work in your own company can actually be taken over by AI cannot be answered in general terms; that is precisely what the work scan from FTE TO AI answers per task. Separately from that, you can already examine which claims you are actually making in tenders and quotes about your own service delivery. The free dimension check lets you name where you think you win, and shows which of those claims can be defended with evidence. The full benchmark, with the scores of your peer group on the same dimensions, is under construction.