Cleaning Records: What They Prove — and What Machine Data Adds

The Leistungsverzeichnis — the service specification, LV for short — is the document everything else hangs on. It lists the areas grouped by how they are used, and assigns each position a task and a frequency, plus the time window, the responsibilities and the materials. It is part of the contract and it is the yardstick against which defects are later measured: without a precise LV there is nothing to measure quality against. A public example shows the level of detail this reaches. Karlsruhe Institute of Technology sorts the buildings on its north and south campus into 17 main room groups — offices, meeting and training rooms, break rooms and tea kitchens, circulation areas, two separate sanitary categories, technical rooms and workshops, laboratories, sports halls, IT and copy rooms, archives, stairs, the large kitchen, dining hall, lecture theatres, libraries and lifts — and gives each group its own performance figures and its own cleaning interval. That is what "specified" looks like; the same logic drives our offices industry page.

Inside that structure sit the two service depths that decide the invoice. KIT defines full cleaning as working through "a precisely defined service specification", with the type and scope depending on the room group. Spot cleaning it defines by condition: "the scope of cleaning depends on the degree of soiling. The minimum scope of a spot clean covers emptying and cleaning the waste bins and ashtrays; the maximum scope corresponds, depending on the degree of soiling, to the service volume of a full clean." The published interval table shows how the two interlock: offices get one full clean and four spot cleans a week, circulation areas two full and three spot, and for sanitary areas — cleaned in full five times a week — spot cleaning is explicitly not applicable. Neither term is defined in law. That is precisely why the client has to define them in writing.

Two more words complete the set. The Reinigungsplan, the cleaning schedule, translates the LV into everyday operation: which area, which weekday, which time window, which team, which method and which agent. The Reinigungsnachweis, the cleaning record, is the return channel — the confirmation that a scheduled service was performed. Everything together is what buyers mean when they say "Reinigungsdokumentation". In most buildings the last of these is still paper, and that is where the interesting problem starts.

The paper record has held on for good reasons, and they should be stated before the criticism. It costs nothing, everybody understands it without training, it is visible to users — which is half its effect in sanitary areas — and it ties a service to a named person at a named time. Nobody should remove it without a replacement.

The limits are just as plain. The sheet documents attendance and a point in time; the initials say nothing about the condition of the area afterwards. It is frequently filled in as a block at the end of a shift, which makes the timestamps a formality rather than a record. It cannot be checked retroactively: no one can reconstruct from a signed line whether a cubicle was actually cleaned. And it only ever documents what the form asks for — the initials, not the deviation, not the reason the third floor was skipped because a corridor was blocked. As evidence, it is a statement of intent that survived contact with the shift.

Behind this sits a distinction worth learning before anyone sells you a reporting feature. An execution record answers "was the service performed?" A quality inspection answers "is the result acceptable?" Those are two different questions with two different methods and two different documents. They are confused constantly — in tender documents, in software demos, and above all in robot sales pitches, where a route log gets presented as proof of cleanliness. It is not. It is proof of a route.

In practice, quality control is a walk-through. The site manager and the client inspect together, take samples per room group, record defects with a deadline, and check them again at the next appointment. Sampling is the only realistic method — nobody inspects a large building square metre by square metre — and that is exactly why the sample size, the assessment scale and the consequence of a defect have to be in the contract beforehand. A quality inspection without an agreed yardstick is a conversation about taste, and the party with the stronger negotiating position wins it.

There is a European standard for this: DIN EN 13549:2001-10, "Reinigungsdienstleistungen – Grundanforderungen und Empfehlungen für Qualitätsmesssysteme; Deutsche Fassung EN 13549:2001", 21 pages, ICS 03.080.30. According to the publisher's own description, the document formulates basic requirements and recommendations for quality measuring systems for cleaning services. Two things it is not: it is not a law, and it is not automatically part of your contract — a standard applies to you when you agree it. We quote its title and stated purpose only, because the text itself is behind a paywall and we do not assert contents we cannot show you publicly.

For one building type German standardisation goes considerably further. DIN 77400:2015-09, "Reinigungsdienstleistungen – Schulgebäude – Anforderungen an die Reinigung", specifies according to DIN minimum requirements for regular cleaning and contains a comprehensive overview of which surfaces in schools and their associated sports facilities are to be cleaned how and how often, together with a service description with minimum frequencies for cleaning contractors and a checklist of the factors that drive cleaning effort. It was drawn up by working committee NA 159-01-20 AA "Reinigungsdienstleistungen" and replaces the 2003 edition. For a school authority, that is the most concrete anchor German standardisation offers for an LV — which is why documentation questions come up earlier in school buildings than almost anywhere else. Whether and how a standard binds you is decided by your contract and your procurement office, not by this article.

Every connected machine logs its run, and the manufacturers are refreshingly precise about what that log contains. Nexaro describes the cleaning report for the NR 1700 as providing, based on time, customer and deployment area, "information about the cleaning time, the cleaning area, and the start and end times". Pudu markets the CC1 with the "digitalisation and visualisation of cleaning results": a cleaning report is generated automatically, and the operator receives real-time notifications including the time and the area cleaned. Read the wording carefully — "results" here means time and area. That is an excellent execution record: machine-generated instead of signed afterwards, timestamped, tied to a map, and impossible to fill in as a block at the end of a shift. It still answers the question whether the machine drove.

Checking the result is a separate feature, and today it is the exception rather than the rule. Pudu describes the CC1 Pro as carrying the world's first rear-facing AI camera for monitoring cleaning performance in real time; after every task it generates heatmaps, in two types — one highlighting the most heavily soiled areas, one marking stubborn stains that remain after several cleaning cycles — alongside an operational display of task execution and completion rates and maintenance frequency. That is the step from execution record to result record, and it is the one genuinely new thing robots bring to this topic. The full comparison of what that costs and what it does not buy is in our CC1 versus CC1 Pro article.

Two boundaries belong next to this, and we state them in every sales conversation. First, a camera that checks cleaning results is a camera in your building: works council, data protection and the question of where the data is stored belong before the commitment in a tender, not after it — our article on data protection and cleaning robots walks that conversation. Second, a machine report is supporting documentation for the cleaning concept you wrote; it is never a substitute for it, and no authority certifies a machine as compliant with anything. We work through that argument in full on the food-hygiene example in retail cleaning. What we do add operationally is continuity: in our full-service model the machines are monitored remotely and serviced on site at least once a week, so a run that silently stopped three weeks ago is not something you discover during the audit.

Public buyers have a legal frame for this, and it points in a useful direction. § 121 (1) GWB requires the subject of the contract to be described "as unambiguously as possible, so that the description is understood in the same sense by all undertakings". § 31 (2) VgV requires the characteristics to be given "in the form of performance or functional requirements or a description of the task to be solved, to be formulated as precisely as possible" so that bids are comparable. And § 31 (6) VgV prohibits referring to a particular production, origin, process, trademark or type where this would favour or exclude particular undertakings — such references need the addition "or equivalent". The practical consequence is clean: "we want a Pudu CC1" does not belong in a public specification. "For area X we require a machine-performed run with report Y, and here is who handles the exceptions" does. Private clients are freer, but the logic holds for them too, because a specification written around a make cannot be compared across bids.

The questions worth making bidders answer in writing are unglamorous and decide the project. Which areas exactly are to be cleaned autonomously, in square metres, by room group and by floor covering — and which are explicitly not? Is the machine class released by its manufacturer for that specific floor covering, or merely described as "suitable"? What is the time window, and what happens when the area is blocked by pallets or stacked chairs? What exactly does the delivered report contain — time, area, map, and does it check the result or only the run — in what format, at what interval, to whom, and kept for how long? Who responds when the machine stops at two in the morning: which response time, which escalation, whose staff? Who fills, empties, cleans and replaces wear parts, and is that in the price? What stays manual — edges, sanitary areas, glass, under furniture — and is that hand work costed in the same LV, so you neither pay for it twice nor discover afterwards that nobody owns it? And what happens on failure: repair time, or replacement machine?

One warning from the trade itself belongs here. In its 2025 industry report the Bundesinnungsverband des Gebäudereiniger-Handwerks writes about school tenders that many cities and municipalities tender by the "cheapest principle", with price as the sole award criterion — and that things only get cleaner when the public sector is willing to increase cleaning scopes, invest in daytime cleaning and terminate contracts with poor providers. Evidence quality follows the same rule: reporting costs money, so if it is not weighted as an award criterion, the bid without it wins. Before you specify anything, it helps to know which machine class is even in question for your floors — our guide on which cleaning robot fits which operation sorts that — what actually drives the cost, which we break down in what cleaning robots cost, and how a rollout runs from survey to unattended operation, described in introducing cleaning robots.

The honest limit first: a coverage map showing 100 % shows 100 % of the planned route. It does not show 100 % of the room. Edges and skirtings a disc brush never reaches, the strips under fixed furniture, stairs, sanitary areas, glass, door handles and light switches — none of that is in the machine data, because none of it is in the machine's job. That is not a defect in the technology; it is a definition, and it belongs in the LV so that both sides read the same report the same way.

The second limit is a category, not a percentage: these machines clean, they do not disinfect. Disinfection is a separate, deliberately controlled step with its own agents, contact times and documentation, and no floor robot performs it. Anything a report says about hygiene therefore says something about floor cleaning — a foundation, not a hygiene concept.

For the scale of what is left, the trade's own number is more useful than any vendor estimate. In its 2025 industry report the Bundesinnungsverband cites the Federal Employment Agency's Job-Futuromat: only 3 of 10 activities in building cleaning are automatable, against, for example, 7 of 8 core activities in retail sales — and the report concludes that the sector will remain "a people's business" for the foreseeable future. For the evidence question that translates directly: machine data covers the automated share, and the rest stays human quality control by people who are increasingly hard to find, which is the subject of our article on the staffing shortage in building cleaning. A cleaning record built only on robot reports documents a minority of the work — precisely, verifiably, and incompletely.