Window cleaning robots do work, but within a narrow envelope: vertical or shallow-pitch glass, clean edges, and surfaces a tether can protect. On those, expect 70-90 % of a manual clean’s result in 30-50 % of the time. Outside that envelope, results fall off sharply.
The “do they work” question is the wrong shape. The useful version is: on which glass, at what pitch, with what recovery method, and who is responsible when it loses grip.
What does the evidence actually show?
Measured results converge on a few points. Robots clear surface dust, pollen and light rain spotting effectively, because those are loose soils that a damp pad lifts easily. They struggle with baked-on mineral film, bird residue and construction cement splatter, which need dwell time, chemistry or mechanical aggression a robot pad does not provide.
They also beat human consistency on large regular areas. A person cleaning a 40-pane curtain wall gets tired, skips corners and leaves streaks on the top row. A robot does not. That consistency is worth more than peak quality on a commercial job where the client inspects from below.
Where humans still win: anything above 55 degrees of pitch, frameless glass roofs with no edge reference, intricate facades, and any glass where the risk of a fall onto occupied space is unacceptable regardless of what the specification says.
What is the actual failure rate on site?
Field logs from facility teams suggest aborted runs cluster at 5-15 % of attempts on well-prepared glass, and rise to 25-35 % where the glass is dusty, the edges are damaged or the operator skips the pre-run wipe. Those numbers are not a product flaw. They are a preparation flaw that shows up as a product complaint.
The most common aborts, in order: suction loss near a damaged seal line, the robot stopping at a dark gutter it reads as an edge, battery sag mid-pass, and the pad clogging on a heavily soiled pane so it starts smearing instead of lifting.
Notice that three of the four are preventable before the machine leaves its case.
Why do reviews and reality diverge so much?
- Reviewers test on clean glass in mild weather. Sites are dusty and hot.
- Review units are new, with fresh seals and unworn pads. Fleet units are not.
- Vertical panels dominate review footage. Pitched glazing is rarer and harder to film.
- Before-and-after photos hide the pre-cleaning wipe the reviewer did by hand.
- Nobody publishes an abort log, which is the single most informative dataset.
If you want reviews that discuss failures openly, our skylight robot review collection lists what each machine does badly, and the WINBOT W2S PLUS review is worth reading before assuming the most popular model solves every case.
Does one robot handle both windows and skylights?
Rarely well. The physics differ: a vertical pane loads the seal evenly and gravity helps the pad down. A pitched skylight loads the seal asymmetrically and gravity pulls along the surface. A machine tuned for one is a compromise on the other.
Buyers who ignore that end up with a well-reviewed window robot that works poorly on a 40-degree roof, then conclude that robots do not work. The conclusion is wrong. The machine was wrong for the glass.
Who gains the most and who should not buy
Strongest gains: facility teams with large continuous glazing and repeat cleaning needs, homeowners with multiple tall windows, and operators who can standardise process. Weakest gains: single small skylight owners with awkward geometry, listed buildings where the glass must be treated gently, and anyone who will not do a two-minute pre-run check.
The pre-run check is not negotiable. Wipe the edge, look for sealant ridges, confirm the tether anchor, check remaining battery. That sequence is the difference between the 5 % abort rate and the 30 % one, and it takes less time than reading the instruction card it is printed on.
One last honest note. On glass above occupied space, the question is not whether the robot works. It is whether your tether, anchor and insurance document are in order. Get that part right first, and the cleaning result becomes the easy part.
Key Takeaways
- Robots deliver 70-90 % of a manual clean on vertical and shallow-pitch glass.
- Aborted runs sit at 5-15 % on prepared glass and 25-35 % on unprepared glass.
- Baked-on mineral film and cement splatter remain outside robot capability.
- One machine rarely covers both vertical windows and pitched skylights well.
- Above occupied space, tether and insurance documentation matter more than the machine.
Do robots help with the safety case?
Yes, materially. A machine on a tether keeps crews off ladders and ropes for routine passes, which removes the most common injury route in facade cleaning. That is often the real justification for purchase, ahead of cost or quality, and it is worth stating in the business case.
What should you measure afterwards?
Abort rate, minutes per square metre, and pad life. Three numbers, logged per job, over eight weeks. That dataset will tell you more about whether the robot works for your building than any review you can read, including ours.
Which number matters most in the business case?
Hours of crew time on ladders and ropes per year, not cost per square metre. Safety exposure reduction is the argument that survives a finance review, because it converts to insurance premium and incident likelihood, and those two numbers are easier to defend than a cleaning-quality comparison nobody will ever measure.
- Crew hours at height per year, before and after
- Abort rate per 100 runs
- Minutes per square metre
- Pad and seal spend per year
- Incidents and near-misses per year