The short answer
- AI estimating from past jobs means finding completed jobs similar to the new one, comparing their estimates with their actual costs, and drafting a new estimate from what those jobs really took.
- It only works if past jobs were costed to consistent cost codes and units. Clean the data first.
- AI drafts and flags. A person checks scope, quantities, current prices, exclusions and margin, and approves before anything goes to a client.
- Start with one job type you sell often, run AI drafts alongside your normal estimate, and compare before relying on it.
What 'estimating from past jobs' actually means
Most trades companies already estimate from past jobs informally: an experienced estimator remembers what the last few kitchens or patios took. The problem is that memory is selective, it leaves when the estimator does, and it rarely includes the jobs that went over. A structured approach does the same thing on purpose, using the records instead of memory.
| Approach | Starts from | Strength | Main risk |
|---|---|---|---|
| Gut feel or old spreadsheet | Memory and the last similar quote | Fast for experienced estimators | Repeats past misses; hard to hand off or check |
| Generic price book or unit costs | Published or vendor pricing | Consistent and quick | Ignores how your crews, subs and sites actually perform |
| Your own past jobs (with or without AI) | Estimate vs. actual on comparable completed jobs | Reflects your real production and overruns | Only as good as your cost coding; can average unlike jobs |
Data checklist: what you need before AI can help
Check these before choosing any tool. If most are missing, the first project is data cleanup, not AI.
- A consistent cost code or category list used on both estimates and actual costs.
- Completed jobs with final actual costs by cost code: labor, materials, subs, equipment and disposal.
- The original estimate for each of those jobs, so you can compare estimate with actual.
- Labor hours, not just labor dollars, so rates can be updated without rewriting history.
- Job type and size in consistent units, such as square feet, linear feet, tons, or system capacity.
- Dates for the estimate and the work, so old prices are recognized as old.
- Notes on what made a job unusual: access, concealed conditions, weather, client changes, rework.
- Approved change orders kept separate from the base scope, so a job with many extras doesn't look like an overrun.
Step by step: drafting an estimate from past jobs
- Describe the new job in the same terms as your records: job type, size in your standard units, scope list and known site conditions.
- Find comparable completed jobs. Filter by job type and size range first, then review the matches. AI can help search notes and documents for similar scope.
- Compare estimate with actual on each comparable job, by cost code. Note which codes consistently ran over or under.
- Adjust for differences in scope and conditions between the comparables and the new job. List each adjustment in writing.
- Refresh prices. Replace old material and labor rates with current approved rates, and label the date of each.
- Draft the estimate with quantities, rates, assumptions, allowances and exclusions shown separately.
- Estimator review. A person checks takeoff quantities, production rates, scope, exclusions, overhead and margin, then approves or rejects the draft.
- Close the loop. When the new job finishes, add its actuals back to the history so the next estimate learns from it.
Examples by trade
The method is the same across trades; the units and risk factors change. These are illustrative examples of what to compare, not prices or benchmarks.
| Trade | Comparable jobs by | Watch for | Common miss to check |
|---|---|---|---|
| Remodel (kitchen, bath, addition) | Room type, square feet, finish level, occupied or vacant home | Concealed conditions, allowances vs. actual selections, protection and access | Demolition, finish carpentry and project management time |
| Landscape design-build (patio, walls, planting) | Feature type, square or linear feet, material | Site access, excavation and soil, material handling, disposal | Equipment time and haul-off |
| HVAC replacement | System type and capacity, home size, single or multi-zone | Ductwork condition, electrical upgrades, permits, access to attic or crawlspace | Extra labor hours on difficult installs |
Trade-specific pages: remodelers and general contractors, landscape and design-build, and plumbing, HVAC and electrical. For service shops, see also AI for an HVAC business.
Where AI helps, and where it doesn't
| AI can help | A person must own |
|---|---|
| Searching past jobs, notes and documents for similar scope | Deciding whether a past job is truly comparable |
| Summarizing estimate vs. actual by cost code across comparables | Measuring the site and confirming quantities |
| Flagging missing information and scope differences | Judging concealed conditions and site risk |
| Drafting line items, assumptions and exclusions for review | Setting current prices, overhead and margin |
| Showing which cost codes ran over on similar jobs | Approving the estimate and any client commitment |
Estimating from past jobs: symptom, cause, fix
| Symptom | Likely cause | Fix |
|---|---|---|
| Drafts are consistently too low | Comparables were estimated low and actuals were never compared | Use actual costs, not past estimates, as the baseline |
| Drafts swing wildly between similar jobs | Unlike jobs are being averaged together | Tighten the comparable filter by job type, size and conditions |
| Labor looks wrong | Only labor dollars were recorded, at old rates | Record hours by cost code and apply current rates |
| Jobs with lots of extras distort the numbers | Change orders mixed into base scope | Keep approved change orders separate from base scope |
| Material prices are out of date | Old purchase prices copied forward | Refresh from current approved sources and date every rate |
| Nobody trusts the draft | No source jobs or assumptions shown | Show the comparable jobs, adjustments and assumptions with every draft |
Tools: start with what you already run
Check your current software first. Construction and field service platforms increasingly include estimating templates, cost catalogs and AI features. JobTread, for example, offers built-in AI and an official AI Connector that lets external tools such as Claude or ChatGPT work with JobTread data. Our JobTread AI automations guide covers how to pilot that safely, and the JobTread implementation page covers budget templates built from real job actuals.
On Aspire? See AI and automation on top of Aspire. On Jobber? See the Jobber guide. Before sharing job records with any AI provider, confirm what it keeps and who can see it.
A 30-day pilot plan
- Pick one job type you sell often and name one estimator as the owner.
- Clean the cost codes and units on a set of completed jobs of that type.
- For each new estimate of that type, produce an AI-assisted draft alongside your normal estimate.
- Compare the two: what the draft caught, what it missed, and how long review took.
- Keep, adjust or stop the pilot based on those comparisons, then add the next job type.
How Trades Agent helps
We help trades companies get their job data into shape and build estimating assistants on top of the system they already use. Start with the Jobwalk, a two to four week fixed-fee audit that includes whether your job history is ready for this. See AI implementation for what we build, or browse more guides.