How Electricians Use AI to Count Fittings From Drawings

Load the PDF drawings into a takeoff tool with symbol search, such as Bluebeam's VisualSearch, Togal's Image Search or Kreo's Auto Count, draw a box around one symbol from the legend, and let it find every match across the sheets. Then review the hits, especially lookalike symbols, before the counts go into your estimate. General chatbots are poor exact counters.

The caveat is that AI counting is pattern matching, so it's only as good as the drawings you feed it. Clean vector PDFs from the designer count far better than scans. Designers draw the same fitting differently, a legend symbol can get counted as a fitting, enlarged plans repeat areas already on the general arrangement, and revision B may not match revision C. The time you save is in the clicking; the thinking and checking stay with you.

Follow me on Instagram@sagnikteaches

What the software is doing when it "counts"

Takeoff tools count in two broad ways, and knowing which one you're using tells you where it will go wrong.

Connect on LinkedInSagnik Bhattacharya

• Visual search. You select one example of a symbol, and the tool finds everything that looks like it. Bluebeam's VisualSearch works this way, as does Togal's Image Search, which Togal describes as boxing one symbol on one sheet and running it across every sheet you uploaded, returning a count and location for each hit. It's fast and works with any symbol, but it's literal: a symbol drawn slightly differently on one sheet can be missed, and a lookalike can be caught.
• Trained detection. The tool has learnt what electrical symbols look like across many drawings and finds them without you picking an example. Drawer AI, for instance, is built for electrical takeoff: it detects devices and fixtures and links each one to its tags, such as panel name, circuit number and mounting height, then exports to Excel and marked-up PDFs.

Many drawings also carry text tags ("L1", "EM", "FA-SD"). Text search on those tags is often more reliable than visual matching, because two luminaires that look identical on the plan may have different type letters beside them.

Subscribe on YouTube@codingliquids

Tools that count electrical symbols, and what they cost

List prices below were checked on each vendor's pricing page in September 2026.

Most have free trials. Trial every shortlisted tool on the same real package you've already counted by hand, so you're comparing against a known answer rather than a demo drawing chosen to look good. For the wider estimating picture, AI estimating software for small electrical firms compares the tools that take counts through to a priced estimate.

Matching the tool to how often you count

The right choice depends mostly on volume. As a rough guide:

• One or two drawing packages a month: a Bluebeam Core licence gives you VisualSearch and count tools for $330 a year. Slower than the specialist tools, but you're paying for a year what they charge for a month or two.
• Commercial tenders every week: trial Togal and Kreo Pro side by side on a package you've already counted. At $175 to $299 a month per user, the test is whether they save an estimator several hours a week.
• Large packages where you also need devices tied to circuits and panels: an electrical-specific tool such as Drawer AI is worth a demo, because that linking is the part generic symbol search doesn't do.

A quick sum settles most of these decisions. Say an estimator's time costs the firm $45 an hour, and a package that takes a full day to count by clicking takes three hours with symbol search, as in the office example below. That's five hours, or $225, saved per package. At two packages a month, a $330-a-year Bluebeam Core licence pays for itself with the first package. Togal at $299 a month needs to save nearly seven estimator hours a month to break even at that rate. Against counting by hand, that's two packages a month. Against Bluebeam, it has to save extra hours on top of what symbol search already saves, which is why the specialist tools make sense for firms tendering every week rather than twice a month.

Whichever you pick, the workflow and checks below stay the same.

A counting workflow for one drawing package

1. Confirm what you've got. Check the drawing register against the files: sheet numbers, titles and revisions. A chat assistant can do this quickly if you paste the register and the file list, and it will flag "E-203 rev C on the register, rev B in the files".
2. Set the scale on every sheet. Counts don't need scale, but lengths do, and a wrong scale on one sheet will quietly ruin the containment figures later.
3. Build your symbol list from the legend. One row per symbol, with its description and any tag. This is a good job for a chat assistant (below).
4. Search one symbol type at a time, across the whole set. Exclude the legend, title blocks and any key plans from the search area so they aren't counted.
5. Review the hits. Most tools show each match highlighted on the sheet. Step through lower-confidence ones, and zoom in on anything near a lookalike.
6. Reconcile against schedules. If there's a luminaire schedule with quantities, compare. If there isn't, check a few rooms against what you'd expect.
7. Export with a per-sheet breakdown, not just totals, so a later revision can be updated sheet by sheet.

Step 1 is where a chat assistant saves the most time for the least risk. Paste the drawing register as text and the list of file names, and ask: "List every sheet on the register with no matching file, every file not on the register, and every revision mismatch. Match on sheet number, ignoring spaces, dashes and underscores." An illustrative reply for an 18-sheet package:

Revision mismatch: E-203, register rev C, file E-203_revB.pdf.
On the register, no file: E-105 Enlarged plan, kitchen core.
File not on the register: E105-RevC-Kitchen.pdf.

The first line is a real problem. The second and third are the same sheet: the file was named without the dash, and the model didn't match it despite the instruction. Read the "missing" list against the "extra" list before emailing the designer, or you'll ask for a drawing you already have. Two minutes of checking still beats ten minutes of comparing an 18-line register by eye.

What a chat assistant is good for on drawings

General assistants can read PDF drawings and images, and they're genuinely useful for the text-heavy parts of a takeoff. Reading a legend into a clean table is the best example:

This is the electrical legend from sheet E-001 (attached image).
Make a table with one row per symbol: symbol description, tag or type letter
if shown, mounting height if shown, and notes. Don't count anything.
If a symbol's description is unclear or cut off, write UNCLEAR.

Illustrative output:

| Description                          | Tag   | Mounting height | Notes                 |
| Twin switched socket outlet          | -     | 450mm AFFL      | -                     |
| Twin switched socket outlet, USB     | -     | 450mm AFFL      | Desk positions        |
| Recessed LED downlight               | L1    | Ceiling         | See luminaire schedule|
| Recessed LED downlight, emergency    | L1-EM | Ceiling         | 3hr maintained        |
| Smoke detector                       | FA-SD | Ceiling         | Fire alarm by others? |
| Data outlet, double                  | D2    | 450mm AFFL      | UNCLEAR: note cut off |

That table becomes your symbol list, and the "fire alarm by others?" question is exactly what you'd want surfaced before pricing. The same approach works for turning a luminaire schedule into a spreadsheet or comparing two revisions of a schedule line by line.

Exact counts are another matter. Ask a chat assistant to count downlights on a busy lighting plan and you'll typically get a confident number that's close but wrong, with no way to see which symbols it included. That's fine as a sanity check ("roughly 60?") against a takeoff tool's 58, and useless as the figure you price from. What ChatGPT can and can't read in PDFs and images explains why.

Counting a three-storey office fit-out

Here is how that plays out for a hypothetical six-person electrical contractor pricing the fit-out of a three-storey office. The package has 18 sheets: a power plan and a lighting plan per floor, enlarged plans of the toilet and kitchen cores, a legend sheet, schematics and schedules. The estimator identifies 14 symbol types to count.

Using symbol search in a takeoff tool, the first pass takes about an hour and a half: set up the legend, run 14 searches, glance at the totals. The review takes another hour and a half, and this is where it pays for itself:

• The enlarged core plans repeated 24 fittings already shown on the general arrangement plans. Excluding the enlarged plans from the count (or the core areas from the GA sheets) removed the double count.
• Emergency downlights were drawn as the standard downlight with a small solid dot. Visual search at the default similarity caught 41 of them as standard downlights. Searching the "EM" tag separately split them out: 196 standard, 41 emergency, and a price difference that matters.
• The legend appeared in the corner of each floor's power plan and had been counted once per sheet for three symbols, until the legend areas were excluded.
• One floor's power plan was revision B in a package otherwise at revision C, spotted when the register check flagged it. Counts for that sheet were redone when revision C arrived.

Counting the same package by clicking each symbol by hand might take a full day. The tool-assisted version took about three hours, with most of the value in the review, not the search. A firm that skipped the review would have tendered with 24 extra fittings and 41 emergency luminaires priced as standard ones.

Revision C of the second-floor power plan arrived four days before the tender deadline, and the per-sheet export made it a 20-minute job instead of a recount. The estimator reran the socket and data searches on E-202 only and compared the new sheet totals with the old: six more twin sockets and two more double data outlets, all in a meeting room that had been split into two. A chat assistant then compared the revision B and C luminaire schedules pasted side by side and reported one change, type L3 up from 12 to 16, which matched the extra downlights in the split room. Only the E-202 line and the L3 line in the estimate changed, and the tender notes record which revision each quantity came from.

Where AI counts go wrong on electrical drawings

• Lookalike symbols: single and twin sockets, switched and unswitched spurs, standard and emergency luminaires, smoke and heat detectors. Search each separately and compare totals.
• Symbol variants: the same fitting drawn at different sizes or rotations, or by a different designer on a later sheet. Check that rotated instances were found.
• Overlaps: symbols sitting on walls, text or other symbols get missed. Scan dense areas such as corridors and plant rooms by eye.
• Duplicated areas: enlarged plans, key plans and details that repeat what's on the general arrangement.
• Scans: blurred or skewed sheets reduce matches. Ask the designer for vector PDFs.
• Scope: fittings "by others", client-supplied luminaires, or items shown for coordination only. The count can be perfect and still price things you're not installing.

The office fit-out shows how the scope trap works. The legend output flagged "fire alarm by others?", and the count found 64 smoke detectors across the three floors. The estimator emailed the main contractor, who confirmed that a specialist fire alarm firm was supplying and installing the whole system, and the electrical contractor's work was limited to a fused spur for the panel. Without the question, 64 detectors and their cabling would have gone into the price, and a tender that looked expensive for no visible reason could have lost the job. The counts stay in the file, because they're useful for coordination, but they don't go into the estimate.

Write the basis of the count into the tender itself, so everyone knows what the quantities cover. A filled-in qualification for this package:

Quantities are taken from drawings E-001 to E-303 at the revisions listed in the attached register (E-202 at revision C). Fire alarm devices and cabling are excluded as by others; a fused spur for the fire alarm panel is included. Luminaires are priced to the types in the luminaire schedule dated 12 September. Enlarged core plans were used for detail only and not counted separately.

If the drawings change after the tender, that paragraph is what lets you price the difference as a variation rather than argue about what was included.

Checking the count before it goes in the tender

A short, fixed routine catches most of the errors above:

1. Pick three rooms of different types and count them by hand. They should match exactly.
2. Compare lighting totals with the luminaire schedule, where one exists.
3. Check that each lookalike pair was counted separately and adds up to the combined visual count.
4. Confirm legend, title block and enlarged-plan areas were excluded.
5. Check every sheet's revision against the register.
6. Have a second person look over high-value items: fire alarm devices, emergency lighting, specialist luminaires.

Check 3 is worth doing with numbers. In the office package, a search for the twin socket symbol at the default similarity returned 312 hits. The legend showed a USB version drawn as the same symbol with a small "U" inside, and a zoomed look at a desk area found both types highlighted. Raising the similarity setting and searching again gave 294 standard twins; a separate search on the USB symbol gave 18. Since 294 plus 18 equals the original 312, nothing was lost between the searches, and the USB sockets, which cost noticeably more, are now priced correctly. If the two totals don't add up, a stricter setting has started missing rotated or slightly different instances, so step back down and review by eye instead.

Keep the marked-up PDFs with the tender. If the job is won and a variation dispute comes later, a sheet showing every counted symbol highlighted is far better evidence than a number in a spreadsheet. For the rest of an electrician's week, everyday AI uses for electricians beyond estimating picks up the other jobs worth handing over.

Takeoff questions from electricians

Can AI count fittings from a scanned or photographed drawing?

Some tools can, but accuracy drops. Vector PDFs exported from the designer's CAD software give clean, identical symbols to match. Scans add blur, skew and speckle, so lookalike symbols get confused and small ones get missed. Togal says it handles scanned and photographed plans. Whatever the tool, ask the designer for the native PDF first and spend more time reviewing when you can't get one.

Will the same tools measure cable and containment runs?

Most takeoff tools also measure lengths once the sheet's scale is set, and some detect linear runs automatically. Cable lengths from a plan are still an estimate: drops, routing around structure and allowances for terminations aren't on the drawing. Use measured lengths as the base, add your usual allowances, and check a sample run against the site or sections.

Is AI takeoff worth it for a small electrical firm doing mostly domestic work?

Usually not as a dedicated subscription. Domestic jobs rarely come with large drawing sets, and a house can be counted by hand in minutes. The case builds when you tender for commercial fit-outs with dozens of sheets every month. In between, a Bluebeam plan with symbol search, at a few hundred dollars a year, is a cheaper first step than a specialist AI tool.

Further reads

• Can AI Price a Job Accurately? What Tradespeople Should Check — Counting is step one; pricing it is where errors cost.
• How to Set Up an Approval Step for AI-Written Quotes — A second check before a tender goes out.
• How to Compare Supplier Quotes Side by Side With AI — Price the counted fittings across suppliers.
• How to Build a Job Costing Sheet in Excel With AI Help — Track whether the tendered job made money.
• Seven AI Mistakes That Cost Tradespeople Jobs — Other AI slips that lose trades work.
• AI Estimating Software for Small Builders: Six Options Compared — How builders' estimating tools handle takeoff.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Bluebeam pricing page and VisualSearch feature descriptions; Togal.AI electrical takeoff page and pricing; Kreo pricing page; Drawer AI product and pricing pages.