AI can help calculate cut and fill volumes, mainly by building and checking the spreadsheet rather than reading the ground. It writes end-area formulas well, applies swell and shrink factors you supply, and spots arithmetic slips. It misreads drawings and invents soil factors. A small firm should confirm every total with a simple hand check before pricing a job.
Most pages that rank for earthworks estimating are software lists, and few show the sum. Without it you cannot tell when a tool, human or AI, has quietly gone wrong. The commonest slip is treating cut minus fill as the export: on the 100 m platform below, that shortcut says 515 m³ to remove, when the soil factors put 560 m³ on the trucks.
The two numbers every earthworks sum depends on
Cut is the soil you dig out. Fill is the soil you place and compact. Two quirks make them harder than they sound. First, soil changes volume as you move it. Second, the volumes on a drawing are measured in the ground, but the volumes you haul and compact are not.
- Bank volume is the soil in its natural, undisturbed state. Drawings and section areas give bank volumes for cut.
- Loose volume is the soil after digging. It takes more space, so a truck carries fewer bank cubic metres than its stated capacity. The ratio is the swell factor.
- Compacted volume is the soil in a finished fill. The design shows fill in compacted volume, and you often need more than one bank cubic metre of cut to build one compacted cubic metre of fill, or sometimes less, depending on the soil and the compaction.
A tool that treats cut equal to fill as balanced, with no allowance for swell or shrink, produces wrong haul numbers every time. That single error is the most common reason a small earthworks job overruns.
A worked example: a 100 m platform with spoil to remove
Take an illustrative site where a small firm is asked to form a 100 m long platform for a shed and access track. The surveyor supplies cross-sections every 20 m. At each station, the section gives a cut area (square metres of ground below the existing surface to remove) and a fill area (square metres to build up). All figures are made up for the example.
| Station (m) | Cut area (m²) | Fill area (m²) |
|---|---|---|
| 0 | 12.0 | 0.0 |
| 20 | 18.5 | 0.0 |
| 40 | 22.0 | 1.5 |
| 60 | 9.5 | 6.0 |
| 80 | 0.0 | 14.0 |
| 100 | 0.0 | 17.5 |
Step 1: end-area volume between each pair of stations
The end-area method says the volume between two sections is the average of their areas, multiplied by the distance between them: V = (A1 + A2) / 2 x L. With 20 m spacing:
| Span (m) | Cut volume (m³) | Fill volume (m³) |
|---|---|---|
| 0 to 20 | (12.0 + 18.5) / 2 x 20 = 305 | 0 |
| 20 to 40 | (18.5 + 22.0) / 2 x 20 = 405 | (0 + 1.5) / 2 x 20 = 15 |
| 40 to 60 | (22.0 + 9.5) / 2 x 20 = 315 | (1.5 + 6.0) / 2 x 20 = 75 |
| 60 to 80 | (9.5 + 0.0) / 2 x 20 = 95 | (6.0 + 14.0) / 2 x 20 = 200 |
| 80 to 100 | 0 | (14.0 + 17.5) / 2 x 20 = 315 |
| Total | 1,120 bank m³ | 605 compacted m³ |
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 nowStep 2: apply the soil factors
Now suppose, for illustration, the soil report gives a swell factor of 25 percent (one bank cubic metre becomes 1.25 loose cubic metres) and says that one bank cubic metre compacts to 0.90 cubic metres in the fill. These are examples only. Yours come from the soil report or the supplier.
- Cut needed to build the fill: 605 compacted m³ divided by 0.90 = about 672 bank m³.
- Cut available: 1,120 bank m³.
- Surplus to remove from site: 1,120 minus 672 = 448 bank m³.
- Surplus as loose volume on the truck: 448 x 1.25 = 560 loose m³.
- Loads for a 10 m³ truck: 560 / 10 = 56 loads.
Apply an illustrative haul and tipping cost of $85 per load and the export costs $4,760. If the same job had been priced as "cut 1,120, fill 605, so export 515", it would ignore both swell and shrink and be priced at 52 loads (515 / 10, rounded up) instead of 56, which is $340 short on haulage at $85 a load. The difference seems small here. On a job ten times bigger the same shortcut misses about 45 loads, close to $3,800, and with a soil that bulks more than 25 percent the gap widens further.
Step 3: do not forget the items outside the sections
Topsoil strip, trench spoil, sub-base for the platform and any unsuitable material sit outside the cut and fill figures. Say the topsoil is stripped across a 1,200 m² area at 150 mm deep: that is 180 bank m³ (1,200 x 0.15) to stockpile and later respread, not to export. Add each as its own line so the quote is not hiding them in an average.
Building the sheet with an assistant
This is where AI is useful. Ask ChatGPT or Claude to write the spreadsheet logic, then check it with the example above. A prompt that works:
Build an Excel earthworks sheet using the end-area method. Inputs (blue): station chainage in column A, cut area in B, fill area in C, spacing is the difference between stations, swell factor, compaction factor (compacted m3 per bank m3), truck capacity in loose m3, cost per load.
Calculated: span length, cut volume per span = (B2+B3)/2*(A3-A2), fill volume per span likewise, totals, cut needed for fill = fill total / compaction factor, surplus or shortfall in bank m3, surplus in loose m3, number of loads (round up), export cost.
Give me exact formulas for the first span and the totals. Show a check cell that sums areas using the trapezoid rule so I can compare it with the span totals. Do not invent any soil factors; leave them as blue inputs with a note saying where to get them.A good output has formulas such as =(B2+B3)/2*(A3-A2) and a ROUNDUP on the number of loads. Read the cut-needed-for-fill line closely, because it is the easiest one to get backwards. If the draft multiplies the fill total by the compaction factor (605 x 0.90 = 544.5) instead of dividing by it (605 / 0.90 = 672), the surplus jumps to 575.5 bank m³, or 719 loose m³ and 72 loads. That is 16 loads, $1,360 at the illustrative rate, of haulage the job does not need, and a quote that loses to a competitor who did the sum the right way round. Paste the corrected sheet into your own file, enter the six stations above and confirm you get 1,120, 605 and 56 loads. If you do not, the sheet or the instructions are wrong, and you have found out before a customer did. See the ChatGPT Excel guide for more on getting workbook formulas right.
What AI gets wrong with earthworks
| Failure | How it shows up | Guard |
|---|---|---|
| Reading levels off a drawing | A plan image goes in, volumes come out, with no way to see how the levels were read | Take section areas from the surveyor or design software; do not extract them from a picture |
| Inventing factors | "Clay typically swells 30 percent" with no source | Soil factors are blue inputs with a named source cell |
| Mental arithmetic | A chat total is off by a few percent, with no workings | Calculate in the spreadsheet, not in chat |
| Mixed units | Metres with feet, or bank with loose, in one sum | Label every column with its unit and basis (bank, loose, compacted) |
| Balancing cut and fill without factors | "Cut 1,120 less fill 605 equals 515 export" | Always convert fill to bank equivalent first |
| Dropping the extras | Topsoil, trenches, unsuitable material and sub-base missing | A checklist of out-of-section items on the sheet |
| Sign confusion | Cut and fill added together as one volume | Separate columns and totals, never a combined one |
The common thread is that every failure produces a plausible number. That is why the hand check below matters more than the tool you use. For the underlying reason models stumble on multi-step sums, see why AI is bad at maths.
The hand check: four tests you can do in ten minutes
- Trapezoid cross-check. Add the cut areas with the first and last halved, then multiply by the spacing: (12.0/2 + 18.5 + 22.0 + 9.5 + 0.0 + 0.0/2) x 20 = 56 x 20 = 1,120. It matches the table. For fill: (0.0/2 + 0 + 1.5 + 6.0 + 14.0 + 17.5/2) x 20 = 30.25 x 20 = 605. If the sheet and the cross-check disagree, one of them has a typo.
- Average-section sense check. The average cut section over 100 m is 1,120 / 100 = 11.2 m². The sections run from 0 to 22 m², so 11.2 sounds right. If the average comes out above the largest section, something is wrong.
- Plan area times depth. Estimate the cut zone at roughly 70 m by 16 m wide with an average depth of about one metre: 70 x 16 x 1 = 1,120. It is crude, but it will catch an answer that is out by a factor of two or ten.
- Load count against the haulier. Ask your driver or haulier how many loads of loose soil they would expect for 560 loose m³ at their truck size. If they say 80 and you say 56, find out why before you quote.
Run all four on every job. Any one is quick. Together they catch most real errors, including ones the sheet shares with the tool that built it.
More examples across the process
A before-and-after: the quote line
Before: "Earthworks to form platform, $9,500." After: "Strip topsoil 150 mm over 1,200 m² (180 m³, stockpile on site). Excavate 1,120 m³ bank to formation. Place and compact 605 m³ fill in 200 mm layers. Export 448 m³ bank surplus (560 m³ loose, about 56 loads). Allowances stated on the volumes sheet dated [date]." The second line lets the customer see what they are paying for, and it protects you if the ground turns out different. See the groundworks quote template tutorial for the full wording.
A realistic mistake: the soil that would not compact
Here is how this tends to go wrong, as an illustration. A firm prices a platform using a compaction factor of 1.0, assuming one bank cubic metre equals one compacted cubic metre. The soil, a wet clay, needs more material than expected, and spoil the firm has already exported has to be bought back as imported fill. The fix is to list soil type and moisture in the sheet header and to get the factor from a source. Add a contingency line or a rate for variations if the ground turns out different from the report.
A prompt for sanity-checking someone else's volumes
When a subcontractor or a customer gives you a volume, ask an assistant to test it without trusting it: "Here are the stations and areas. Recalculate cut and fill with the end-area method, show each span, and tell me if my total of X is within 3 percent." Providing the numbers, rather than a drawing, keeps the assistant on arithmetic, which it can show step by step.
A small trench job
For a 60 m service trench 0.6 m wide and 1.2 m deep, the bank volume is 60 x 0.6 x 1.2 = 43.2 m³. Pipe and bedding take up space, so backfill needs less than the dug volume, and the rest goes to spoil. An assistant can set this up quickly, but the pipe diameter, bedding depth and the share of spoil that can be reused are decisions for the person who knows the specification.
Choosing more sections where the ground changes
Where the ground is nearly flat, sections every 20 m are plenty. Where it changes sharply, a hump or a hollow, use closer spacing. In our example the span from 60 to 80 m moves from 9.5 m² of cut to none, which is the kind of transition where the average-area method tends to overestimate. If that span mattered to the price, ask for an extra section at 70 m and recompute. The sheet does not care how many rows you have; the cost is only the surveyor's time.
A basement dig, checked two ways
Consider an illustrative basement excavation 13 m by 9 m at the base and 2.5 m deep. With vertical, shored sides the volume is 13 x 9 x 2.5 = 292.5 bank m³. At a 25 percent swell that is 365.6 loose m³, or 37 loads of 10 m³ (36.6 rounded up). If the ground allows 1:1 side slopes instead, the top of the hole is 18 m by 14 m. Averaging the two end areas gives (117 + 252) / 2 x 2.5 = 461.3 m³. The more accurate prismoidal formula uses the mid-height area (15.5 x 11.5 = 178.25 m²): 2.5 / 6 x (117 + 4 x 178.25 + 252) = 450.8 m³. The end-area shortcut overstates by about 2.3 percent here. That is acceptable for a quote, and it is a good example of an error you should know the size of. Ask an assistant to run both formulas for you, then compare the answers with your own calculator.
Asking the assistant to show its working
If your plan can run code or a calculator tool, ask the assistant to compute with it and print each intermediate result. A good instruction is: "Do not give me a total until you have shown the volume for every span. Then recompute the total a second way and tell me if the two differ." Two routes to the same total catch most slips. If the assistant cannot show workings, treat the result as a guess and recompute it in the spreadsheet yourself.
Haul distance and cycle time are separate questions
Volumes tell you how much soil moves. They do not tell you how long it takes. A site with a tip 5 km away and a 40-minute round trip needs more trucks, or more days, than one with a tip next door, even for the same 56 loads. Keep the volume sheet and the programme sheet separate, and use the volumes as the input to the haul plan rather than mixing the two. A cut and fill number that includes a guess about haul time is hard to check.
Telling the customer what the number does and does not include
A volumes sheet is only useful to a customer if they can see its assumptions. Ask the assistant to turn your sheet into a short plain-English note: the survey date, the section spacing, the soil factors and where they came from, and the items excluded, such as rock, groundwater, contaminated material and services found during the dig. A sentence such as "Volumes are based on sections at 20 m spacing and the soil report dated [date]; if ground conditions differ, quantities will be re-measured and priced at the rates shown" protects both sides. Check the note against the sheet before it goes out, and make sure every exclusion in it is one you would really stand behind.
When a spreadsheet is enough, and when it is not
A spreadsheet with a hand check is a good fit for pads, access tracks, small roads, trenches and basements. Move to dedicated earthworks or terrain-modelling software when you routinely receive digital terrain models, price jobs in the tens of thousands of cubic metres, or need haul-distance and mass-haul analysis. Dedicated tools also read the ground surface for you, which no spreadsheet can. For more on choosing tools for takeoffs by trade, read six AI takeoff tools matched to trade and plan type, and for turning volumes into a price, pricing excavation work by the hour, load or fixed sum.
Keeping the sheet reusable
Save the workbook as a template with the formulas locked and the inputs in blue. Add a notes tab recording where each soil factor came from, the date of the survey and the version of the sections. Next job, you replace the stations and areas, adjust the factors and run the four hand checks. Over time you collect a set of real jobs where you can compare the estimated volume with what the machines actually moved, which is the only feedback that truly improves a cut and fill estimate. If you would like help setting up a template and an AI checking routine around your own job types, that is something an AI implementation consultation can cover. The method above works on its own.
Cut and fill questions from small earthworks firms
Can I upload a site plan and get cut and fill from ChatGPT?
Not reliably. A general assistant can misread contours, levels and scale on a drawing and will still give a confident number. Use it for the spreadsheet formulas and for checking your sums, and take the section areas from survey data or design software.
What swell and shrink factors should I use?
Use the figures from the geotechnical report or your earthworks supplier for the actual soil. Typical ranges differ widely between sand, clay and rock, so any number in a general tutorial is only an illustration. Never let an AI choose a factor without a source.
Is the end-area method accurate enough for a quote?
For a small firm quoting standard platforms, roads and trenches it is a recognised, practical method, as long as sections are spaced closely enough where the ground changes. It slightly overestimates on tapering sections, so check those spans and ask for more sections if the margin is thin.
When is dedicated earthworks software worth buying?
When you routinely work from digital terrain models, price jobs with thousands of cubic metres, or need haul-distance and mass-haul analysis. For simple pads, trenches and small roads, a carefully built spreadsheet plus a hand check is often enough.
Further reads
- How Much Does Estimating Software Cost? Price Bands With and Without AI — What dedicated estimating software costs before you buy.
- Best AI for Quantity Surveyors: Five Tool Types Compared by Job — Tool types for measured quantities, compared by job.
- AI Tools for Construction Estimating: Bid Checks After the Takeoff — Bid checks to run after the takeoff is done.
- Asphalt Paving Estimating Software: What AI Adds and What It Costs — Quantities for the surfacing that follows the earthworks.
- AI for Civil Construction: Plant Pre-Start Checks and Service Schedules — Keep the machines behind the volumes safe and serviced.
- Groundworks Tenders: Build a Tender Pack and Method Statements With AI — Package volumes into a tender with method statements.
- Concrete Estimate Template: Volume, Pump and Finishing Lines With AI — A filled-in concrete estimate sheet: volume with waste, pump and finishing lines, and the AI audit that catches the items you forgot.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: standard average end-area volume method and bulking and compaction conventions used in earthworks estimating; ChatGPT and Claude plan pages. All figures are illustrative; take soil factors from your geotechnical report.