AI Takeoff for Electrical: Where It Actually Helps

Electrical estimators keep asking whether AI can just do the takeoff for them. This post comes from my step-by-step electrical quantity takeoff video, where I measure light fittings and cable tray on a real drawing set in ZZ Takeoff and build the assemblies that turn those counts into a full bill of quantities. It’s written for electrical estimators who want the manual method done properly, plus an honest read on where AI fits around it.

Key takeaways

• Electrical takeoffs split into primary quantities you measure (light fittings, outlets, cable tray, switchboards) and secondary quantities that are derived, never measured (brackets, splice plates, sub-circuit cable, labour hours).
• The measuring and counting itself stays human. AI’s honest fit sits around it: proposing the item list, drafting assembly ratios, and cross-checking the finished bill of quantities.
• On a real drawing set, 16 counted light fittings automatically derived 16.8 support brackets, 88 m of cable, and 32 man hours, once the assembly was set up correctly.
• Building and maintaining the assembly library, not the counting, is the slow and valuable part of an electrical takeoff.
• Counting a secondary item like a cable clip or splice plate straight off the drawing is the most common electrical takeoff mistake.

What is an AI electrical takeoff?

An AI electrical takeoff uses AI around a manual electrical quantity takeoff, not instead of it: proposing what to count on the drawings, drafting the assembly ratios that turn those counts into a bill of quantities, and cross-checking the finished result. The measuring and counting itself, light fittings, cable tray, conduit runs, stays with the estimator in a dedicated takeoff tool.

That’s the same line that holds for construction takeoffs generally, and electrical is one of the cleanest trades to see why. Almost everything on the drawing is either a primary quantity you count or measure, or a secondary quantity an assembly derives from it. AI has a defensible job on both sides of that split. It just never touches the drawing itself.

What do you actually measure on electrical drawings?

You measure anything that’s visible on the drawings and not already provided to you elsewhere, and for electrical that typically means cable containment, light fittings, switchboards, power outlets, fire detection, data outlets, and earthing. That’s the rule from my video, and it’s a good default for any trade, not just electrical. If the client hands you a cable schedule, you use their numbers instead of re-deriving cable lengths from the layout. If they don’t, cable becomes one more thing you measure, or more often, one more assembly you derive from something else you’ve already counted.

For electrical, the primary quantities you’ll typically measure are:

• Cable containment: tray, conduit, trunking (linear metres)
• Light fittings, usually split by fitting type (counted)
• Power outlets and accessories (counted)
• Switchboards and distribution boards (counted)
• Fire detectors and alarms (counted)
• Data and communications outlets (counted)
• Earthing and lightning protection (counted or measured, depending on the item)

Cable itself is the one that trips people up. Most clients supply a cable schedule, so you use it. When they don’t, you’re either measuring conductor run lengths off the layout or deriving them from an assembly tied to something else you’ve counted, like fittings or outlets.

How do assemblies turn a count into a full bill of quantities?

An assembly is a fixed ratio that converts one primary quantity into everything that comes with it, material, fixings, cable, labour, and plant, calculated automatically once you’ve set the ratio, so a single count or measurement drives the rest of the bill of quantities. You never measure the secondary items directly. You measure the primary once and the assembly does the rest.

Take a hanging light fitting. The primary quantity is the fitting count. Everything else, the suspension chain and hook set, the mounting bracket, the junction box, roughly 2 m of cable per fitting, and the fixings, comes from the assembly. On the drawing set in the video, 16 counted light fittings produced:

Item	Assembly ratio	Quantity for 16 fittings
Light fitting (primary)	1:1	16
Support bracket	1:1, plus 5% wastage	16.8
Sub-circuit cable (2.5 mm²)	5 m per fitting, plus 10% wastage	88 m
Electrician labour	2 hours per fitting	32 hours

Cable tray works the same way, just as a linear measurement instead of a count. 24 measured metres of tray, rounded up to the nearest 3 m length it’s actually bought in, came out to 27 m, 10.26 labour hours at a 0.4-hour-per-metre installation rate, 12.2 support brackets at one every 2 m, and 9 splice plates at one every 3 m. None of that second group of numbers gets measured off the drawing. It falls out of the assembly the moment the 24 m primary quantity is in.

The measuring itself is the easy part:

“You measure anything that is visible on the drawings and that isn’t already provided in other information.”

Tim Fairley, in the source video

Where the real time goes is building an assembly you can actually trust: getting the wastage percentages right, the bracket spacing right, the productivity rate right, so the numbers it spits out hold up when someone checks them.

Can AI do an electrical takeoff for you?

No, not the counting or the measuring. AI is unreliable at precise visual tasks, and the reliable pattern is the same one that holds across AI for construction estimating and takeoff generally: AI proposes what to measure, drafts the assemblies, and cross-checks your finished bill of quantities, while you do the counting and measuring in your takeoff software.

The evidence for that is stark. On ClockBench, a benchmark that tests whether AI models can read an analog clock face, a task easier than reading a floor plan, humans average roughly 91 percent accuracy. The best current model manages about 50 percent, and several score under 10 percent. A model that can’t reliably read a clock has no business counting your light fittings or scaling your cable tray runs. That’s why ContractorOS’s own ai-fit map puts measuring and counting quantities in the no-go column, with a high blast radius if it’s wrong and nobody catches it. It’s the same reason it stays manual in ZZ Takeoff, PlanSwift, or whichever tool you use.

Three honest jobs AI can do around an electrical takeoff, none of which involve reading a drawing at scale:

1. Building the item list. Read the drawing set and the legend, propose what needs counting, whether that’s the fitting types shown, or a fire alarm system you might otherwise miss, before you open the takeoff software.
2. Drafting assembly ratios. Read the specification and propose a starting assembly (bracket spacing, cable length per fitting, wastage percentages), which you then check against your own numbers before you trust it.
3. Cross-checking the finished bill of quantities. After you’ve measured, a second pass over the BOQ can catch a fitting type on the legend that never got a count, or a quantity that looks off against what’s on the drawing.

There’s one narrow exception worth naming. ContractorOS’s own field testing on a live takeoff project found counting text-tagged items to be high-confidence, while scaling a dimension off a drawing stayed low-confidence throughout. Even there, every count still gets checked line by line before it goes anywhere near a price. Measuring stays fully manual either way.

Where AI actually fits in an electrical takeoff

Electrical takeoff step	AI fit	Why
Building the item list from the drawing set	Good	Reads the legend and layout, proposes what to count before you open the takeoff software
Counting light fittings, outlets, switchboards	No	Stays human, in the takeoff software. Vision isn’t reliable enough for a count you’re pricing off
Measuring cable tray, conduit, or trunking length	No	Scaling a dimension off a drawing is a precision visual task, and AI is weak at it
Drafting assembly ratios from the spec	Good	Proposes a starting ratio you check against your own numbers, not a final answer
Deriving secondary quantities once primaries are set	Good	Deterministic maths the takeoff software already does without AI once the assembly is set
Cross-checking the finished bill of quantities	Good	A second pass after you’ve measured, to catch a missed fitting type or an off quantity

Common mistakes in electrical takeoffs

• Counting a secondary item straight off the drawing. Cable clips, splice plates, and support brackets rarely get their own symbol on the plans. If you’re counting them individually, that’s the textbook mistake. They belong in the assembly ratio, not the count.
• Ignoring access and productivity differences. A light fitting installed from a ladder and one that needs a boom lift use the same material assembly but very different labour. If the takeoff is feeding an estimate rather than a procurement order, split them out.
• Reusing a generic assembly library without checking it against the spec. Wastage percentages, bracket spacing, and cable length per fitting change project to project. A library built for one job is a starting point, not an answer.
• Skipping the client’s cable schedule when it’s provided. Re-deriving cable lengths that are already given wastes time and creates a second, competing source of truth.
• Trusting AI to do the counting because the first few outputs looked plausible. A plausible count and a correct count aren’t the same thing, and there’s no way to tell them apart without checking the drawing yourself.

The same primary-and-assembly split runs through AI takeoff for HVAC and AI takeoff for plumbing, just with a different item list each time. The assembly templates I use across all three trades, and the workflows built around them, are shared inside the ContractorOS community.

Watch the full video above to see both assemblies built in ZZ Takeoff, from the first light fitting through to the finished bill of quantities.