AI can measure floor areas from scans or photos, but accuracy depends on the tool, the device and the room, so check it yourself. Use it for marketing, rough budgets and internal records once it passes a laser-measure test on your own units, and hire a surveyor for any figure that sets money between people.
Every figure below is invented to show how to read a test, not a measurement of any product, and your own numbers will differ. The ten-unit test below is built to give you your own. It is the same test a contractor should run before buying takeoff software, because the result decides whether a tool saves time or creates quote errors.
Three kinds of tool, three kinds of error
Property measurement tools sort into three groups, and each fails differently.
- Depth-scan tools. A phone or tablet with a LiDAR sensor sweeps the room and builds a model. Apple's RoomPlan framework, for example, uses the camera and LiDAR Scanner on iPhones and iPads that have one to capture a room and identify walls, windows, openings and doors. Errors here come from glass, mirrors, black surfaces, moving people and furniture covering the wall base.
- Photo-only tools. These infer dimensions from ordinary photos or video, usually by estimating scale from a known object or from camera motion. The result depends on how well that scale estimate holds, and any error in it is multiplied across the room, which matters most in large or awkwardly shaped rooms.
- Floor-plan-image tools. These read an existing drawing or marketing floor plan and compute area, sometimes trusting a printed scale or a stated dimension. They inherit every error in the drawing, which may be rounded, outdated or drawn to a different wall boundary.
The label "AI" is attached to all three, but the evidence the tool starts from differs, so do not compare them on the brochure. Compare them against the same laser reading, in the same flats.
A ten-unit test against a laser measure
Here is the test, with invented results so you can see how to read them. Imagine a lettings portfolio of ten flats of different sizes, and a handheld laser measure; common models state a typical accuracy of a millimetre or two (one popular model lists plus or minus 1.5 mm), so read the spec for yours.
Set the rules before you measure
- Pick the boundary. Measure to the internal faces of the walls (net internal area style), and say so. Include or exclude bathrooms and hallways the same way every time. Decide how to treat alcoves, chimney breasts and door recesses, and write the rule down.
- Take each length twice with the laser. If two readings differ by more than 10 mm, retake.
- Compute the area in a spreadsheet. Break irregular rooms into rectangles and add them. That is your reference figure.
- Scan with each tool once, without looking at the reference. Record the area the tool reports and the time it took.
- Compare. For each unit calculate the error as tool minus laser, divided by laser, in per cent.
The Complete Excel Course with AI Integration
From formulas and pivot tables to Power Query, DAX, VBA and AI with Copilot, ChatGPT and Claude — hands-on from the first lesson.
By Sagnik Bhattacharya and Virginia Thorn, Primetime Emmy Award winner
Buy nowHow to read the results: an invented run
Nobody measured these flats. The numbers below are made up to show the arithmetic and the traps, not to rate LiDAR, photo or plan tools; in your own run the order and the gaps may come out quite differently. In this invented run the laser areas, in square metres, were 38.2, 45.6, 52.4, 61.0, 34.8, 70.3, 48.9, 55.5, 41.7 and 66.2, giving a portfolio total of 514.6 m². Three tools, one of each type, were scored against them.
| Invented tool (type) | Average absolute error | Worst unit | Units within 2% | Units within 5% | Portfolio total vs laser |
|---|---|---|---|---|---|
| Tool A (LiDAR scan) | 1.0% | 1.2% | 10 of 10 | 10 of 10 | +0.9 m² |
| Tool B (photo-only) | 7.4% | 8.9% | 0 of 10 | 0 of 10 | -6.2 m² |
| Tool C (floor-plan image) | 10.6% | 12.6% | 0 of 10 | 0 of 10 | +12.2 m² |
Two things in this table matter more than the ranking, which is invented anyway. First, the totals look better than the units. Tool B's total is only 6.2 m² out on 514.6, about 1.2 per cent, because over-measured and under-measured flats cancel. A manager who only compared totals would call it accurate. Averaged flat by flat, its error is 7.4 per cent, and on a 45.6 m² flat that is over 3 m². Always score unit by unit.
Second, even an average error of 1.0 per cent, Tool A's invented figure, has a size. On a 55 m² flat that is about 0.55 m², roughly the size of a large doormat. Whether that matters depends on what the number is for, which is the next question.
What accuracy is good enough for each use
Set a tolerance by purpose, not by tool. The table below is a rule of thumb I would start from, to be adjusted with your adviser.
| Use of the area figure | Tolerance to aim for | Reference needed |
|---|---|---|
| Marketing text and a first look at a listing | Around 5% | Spot check against a laser on a sample |
| Internal records and rough budgets (paint, flooring estimates) | Around 2 to 3%, plus waste allowance | Laser check on the first few units |
| Rent per unit area, service-charge apportionment, lease plans | A surveyed figure to a stated standard | A surveyor's measured survey |
| Rent review, sale, dispute | Surveyed and signed | A surveyor, whose report you can rely on |
So in the invented example Tool A would earn its place in the first two rows, and Tool B in the first row only. Neither is a replacement for a measured survey where a figure sets money between two parties. "Close enough" is a business choice for you; "correct to a standard" is something a signed survey provides.
Why a tool can be accurate and still disagree with a surveyor
A frequent complaint after these tests is "the scan says 52.4 and the plan says 49.8". The gap is often not an error but a different definition. Measuring to internal wall faces gives a smaller figure than measuring to the outer face, and a "gross" figure includes wall thicknesses and sometimes shared areas. Fitted wardrobes, kitchen units, stairwells, balconies and areas with low ceilings can be in or out depending on the standard.
Before comparing figures from different sources, ask each one what it measured. If the AI tool's output does not say (many do not), assume it measured to visible internal surfaces. When the number matters, tell the surveyor which standard you need the figure in and which areas to include, and ask for them to be stated on the report.
Conditions that break scans: the awkward-room test
Add two or three deliberately hard rooms to your test, because that is where tools fail. Say the portfolio has a loft conversion with sloped ceilings, a lounge with floor-to-ceiling glass doors, and a dark-painted hallway with a mirror. Scan each once with the best tool and compare to the laser. Examples of what often goes wrong, and what the numbers tell you:
- Glass and mirrors. Depth sensors can read a mirror as an extra room or a window as an open space. A 16 m² lounge might appear as 18 m² when the scan includes a reflected corridor. Check by looking at the model, not only the number.
- Sloped ceilings. Floor area under a slope is a matter of definition, and measurement standards differ on whether low-headroom areas count. A tool will not know which standard applies unless the app lets you set it, so ask your surveyor which rule your figures should follow.
- Stored clutter. A wall hidden behind boxes may be placed on the box face. Scan after the flat is cleared, and re-scan if a tenant is still in residence.
- Walking speed. Moving too fast leaves gaps that tools fill with guesses. Slow down, overlap passes and finish each room before the next.
Record the error for each awkward room separately from the ten ordinary units. If the tool is within 2 per cent on ordinary units and 6 per cent on awkward ones, your rule becomes: use for ordinary rooms, laser-check every awkward one.
Cost and time: what the test is really measuring
Accuracy is only half the case. Time and cost decide whether the tool is worth using. Record, for each tool, how long it took per flat from walking in to a usable figure. In the same invented run: laser and spreadsheet, 22 minutes per flat including twice-measured lengths; LiDAR scan, 9 minutes including a tidy-up of the model; photo-only, 6 minutes; floor-plan image, 2 minutes. Over ten flats, that is 220, 90, 60 and 20 minutes. If your own depth-scan tool saved 130 minutes over the laser for a 1 per cent error, it earns its keep for the uses in the first two rows of the tolerance table. If your portfolio needs a figure to a stated standard on every unit, that saving will not reach the final number, because the surveyor's measurement is still needed.
Also note subscription and device costs from the vendor's pricing page on the day you test, and whether the report you need (area table, plan drawing, export) is on the plan you are trialling. Price and plan names change, so a figure from a year ago is not a basis for a decision.
What the result means for choosing contractor takeoff tools
The same ten-unit method is the best filter when a contractor, an estimator or a facilities team is choosing measuring software. Vendors of takeoff tools show impressive demonstrations on clean drawings. Your drawings are not clean. Run your test on five of your own jobs, compare to hand-measured quantities, and keep the vendor that is within tolerance on the awkward ones. A tool that is within 2 per cent on plain rooms but 10 per cent on awkward ones will quote you perfectly until the job that loses you money. For tool types by trade, see five tool types compared by job and plan or photo takeoff for a small crew.
A worked sum: one awkward lounge, step by step
Here is the calculation for one L-shaped lounge so you can see where an error in a scan comes from. The laser gives the main part as 5.40 m by 3.80 m, which is 20.52 m². The side bay is 2.10 m by 1.60 m, which is 3.36 m². The reference area is 23.88 m². Now say a LiDAR scan reports 24.35 m². The error is 24.35 minus 23.88, which is 0.47 m², divided by 23.88, so about 2.0 per cent. That sits on the edge of the tolerance in the table above. Opening the scan model, you find the main rectangle was drawn 5.44 m long instead of 5.40 m, because a deep skirting board was read as part of the wall line. 5.44 by 3.80 is 20.67 m², which with the 3.36 m² bay gives 24.03 m², a difference of 0.15 m² from the reference, so most of the 0.47 m² gap comes from somewhere else, perhaps a door recess counted as floor. You would then check the recess against the laser. The lesson is that a good test does not stop at the percentage: you open the model and find which wall moved, so you learn whether the tool has a habit (skirting boards, recesses) or made a one-off slip. Two or three such checks will tell you which rooms to laser-check every time.
The same method on a floor plan someone else drew
Floor-plan-image tools deserve their own small check because the plan may be wrong before the tool starts. Say a portfolio manager takes the marketing plan for a 48.9 m² flat and runs it through a tool that reports 54.3 m². A gap of 11 per cent looks like a tool failure. A closer look shows that the plan was drawn with a printed total of 54.3 m² that includes a communal stair landing and a store cupboard outside the front door. The tool read the plan correctly. The plan is simply a gross figure, and the lettings agreement quotes a net one. Add a column to your test sheet called "source figure type" and fill it in: laser internal, plan gross, surveyor internal, and so on. Many apparent tool errors turn out to be mismatched definitions.
Privacy and access when you scan a tenant's home
A scan or photo set is a record of someone's home. It can capture post, medication, photographs and valuables that have nothing to do with floor area. Before scanning an occupied unit, give the tenant notice in writing, say what will be captured and how long you will keep it, and offer to scan only empty rooms or to keep belongings out of shot. Store the models in your normal property file, not on a personal phone, and delete the raw captures once you have the area table and the plan you need. If an app uploads the data to the vendor's cloud for processing, read its current privacy terms and note it in your own records. If tenants object, the laser is the fall-back.
A filled-in test sheet
Unit 4 (2-bed, occupied) | Laser internal 61.0 m2 | LiDAR 60.4 (-1.0%) | Photo 57.2 (-6.2%) | Plan image 66.9 (+9.7%)
Minutes: laser 24, LiDAR 10, photo 6, plan 2 | Conditions: sofa against party wall, glazed balcony door
Notes: LiDAR stopped at sofa, wall line inferred; photo tool scale from door width; plan is gross figure (includes landing)
Boundary rule used: internal wall faces; alcove under 0.5 m deep excluded | Measured by: [initials] | Date: [date]That single line is enough to explain, months later, why Unit 4 was treated as an exception.
A routine for scanning a whole portfolio
Once a tool passes, the next risk is sloppy use. A routine of five habits keeps the accuracy you saw in the test. Scan in the same direction in every flat, usually clockwise from the front door, so the models are comparable. Switch on every light and open internal doors before you start. Pause after each room and check the live preview for gaps. Note the device, app version and date on every file, because an update can change results. And laser-check one unit in every ten, picked at random, and log the error: if it drifts above your tolerance, stop and find out why before carrying on. Say a manager scans 40 flats over a fortnight and checks four with the laser, finding errors of 0.9, 1.4, 1.1 and 4.8 per cent. The 4.8 is the unit to investigate, and the first question is what changed: a new tenant's furniture, a different phone, a rushed scan. A sample check is cheap insurance, around 25 minutes a unit, and it makes the rest of the portfolio figures believable when an owner or auditor asks how they were produced.
When to hire a surveyor, in plain terms
Hire one when the figure sets money or rights between people. Examples are a rent review that depends on area, an apportionment of service charges, a sale or valuation, a dispute over a deposit or dilapidations, and a lease plan that will be attached to a legal document. Hire one also when the building has features the test showed to be tricky: sloped ceilings, glass, irregular shapes or shared areas. Ask the surveyor to state the standard used and the date, and keep the report in the file. In an ordinary lettings workflow, the surveyor's measured figure can then be the reference for your own AI scans in later tests.
A surveyor's fee is not the thing to save on in those cases. A single mismeasured unit that is carried across a whole building's apportionment can cost more than the whole survey.
How to keep the results usable
Store the test in one spreadsheet with a row per unit and these columns: unit, laser area, tool area, error per cent, minutes taken, notes on conditions. Add a second tab with your boundary rules, the date, the tool and app version, the device, and who measured. A year later, when the vendor updates the app or you change phone, rerun a three-unit mini test and compare. A line like "re-tested on three units 4 Nov, average error 1.1%" is evidence that your routine still works.
For the wider picture of which free or paid tools a small portfolio can run, see what a small portfolio can do with free AI tools, and for software that links tenancy tasks, AI software for landlords compared.
Questions property managers ask about measuring with AI
Which type of tool is usually most accurate?
There is no independent ranking you can rely on. Depth-scan tools start from measured distances, photo tools estimate scale and floor-plan tools inherit the drawing's errors, so it is reasonable to expect depth scans to start ahead, but each app, device and room differs. Run the ten-unit test below and use your own numbers.
Do I need a LiDAR phone?
LiDAR scanning needs a device with the sensor. In Apple's range that means iPhone and iPad models that have a LiDAR Scanner, which its RoomPlan framework uses alongside the camera; check the model's specifications. Other phones can use photo-based tools, which need their own accuracy check. Look at each app's supported-device list.
Is a laser measure the right reference?
For a first test, yes, as long as you measure the same way every time: the same wall faces, the same rule for alcoves and door recesses, and each length taken twice. Use a surveyor's measured figure as the reference for anything with legal or financial consequences.
How many units should the test cover?
Ten is enough to see a pattern, and twelve to fifteen if you have different layouts. Include at least two awkward rooms, such as a sloped ceiling or an L-shaped lounge, because that is where tools fail first.
Further reads
- AI Landlord Updates and Owner Statements in Minutes — Owner statements and landlord updates drafted in minutes.
- How Much of a Property Manager's Week Can AI Take Over? — How much of a property manager's week AI can take over.
- How Property Managers Use AI to Triage Maintenance Requests — Use AI to triage maintenance requests.
- AI Estimating Software for HVAC: Replacement Quotes and Duct Takeoffs — What estimating software measures and misses on duct takeoffs.
- How Electricians Use AI to Count Fittings From Drawings — Counting fittings from drawings with AI.
- AI Home Inspection Software: What It Writes, Who Signs and What It Costs — What inspection software writes and who signs.
- AI Estimating Software for Flooring: Test Accuracy on Your Own Plans — What flooring takeoff tools measure from PDFs and photos, how waste and pattern allowances work, and a five-plan test to run before you trust the numbers.
- Window Measurement App: Measure and Quote on the Doorstep With AI Help — Windowmaker Measure, PrefMeasure, DoorWin, Aufmaß App and others compared, with a phone-against-laser test and a survey sign-off for installers.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Apple RoomPlan developer documentation and overview page (camera and LiDAR Scanner room capture); Bosch Professional GLM 50-27 C product page (stated typical accuracy). All test figures in this tutorial are invented to show the method, not measurements of any product or tool type.