AI at a General Contractor: the honest map

What to buy, what to build, and why the difference is Dawson's own data

Prepared for Dawson Construction
Lunch & Learn · July 2026

Buy: the commodity layer

If a vendor sells it to every GC in the country, buy it, don't build it. Takeoff-from-PDF, photo search, transcription: these are solved, cheap, and getting cheaper. Building a worse clone is wasted money.

Build: the moat layer

If the value comes from Dawson's own data (your unit costs, production rates, estimate-vs-actual history, standards), no vendor can sell it to you, and no competitor can copy it. That's where custom pays.

Preconstructionwin the work

Quantity takeoff from drawingsBUY

Togal.AI, Kreo, STACK: ~$35–300/mo, near-instant counts & areas. Commodity. Use their output as the input to what's below.

Dawson-calibrated estimatingBUILD

Generic quantities × Dawson's historical costs, standards & cost codes → a bid in Dawson's Excel format. The engine sketch you'll see today. Ryan's 3-FTE line item.

Sub bid levelingBUILD

Parse incoming sub quotes (PDF/email) into one normalized comparison of scope gaps, exclusions, and plugs, in minutes on bid day.

Spec review & reconciliationASSIST

LLM reads the spec book against the estimate: flags scope in the specs that never made the numbers.

Proposals, prequal, SOQsASSIST

First drafts from past submissions + project facts. Hours → minutes; humans keep the voice.

Executionprotect the margin

Weekly exec brief per projectBUILD

Daily logs, RFIs, submittals, cost events → an auto-written Monday brief: what changed, what needs a decision, what's aging. Zero new data entry.

Issue & RFI aging radarBUILD

Surfaces the silent schedule-killers, like an RFI sitting 12 days with the architect, before they consume float. Days saved = real money at $10M+ scale.

Submittal review vs. specASSIST

First-pass check of shop drawings/product data against spec sections; human approves. Emerging vendor space; watch before building.

Daily-log pattern detectionBUILD

"Third water-intrusion mention in two weeks": patterns across field notes no one has time to re-read.

Photo documentation searchBUY

Procore and others already do AI photo search. Use what you own.

Meeting minutes → actionsASSIST

OAC meetings transcribed, decisions and owners extracted, pushed to the project record automatically.

Closeout & the loopcompound the data

Estimated-vs-actual captureBUILD

Every closed division feeds a historical cost database of real Dawson unit costs and production rates, by building type. This is the moat everything else drinks from.

Closeout package assemblyASSIST

O&M manuals, warranties, as-builts collected and indexed automatically from the project record.

Bid post-mortemsASSIST

Won/lost analysis across bids: where Dawson prices high, low, and why, from data already in the estimate files.

Templated spreadsheets → toolsBUILD

Any Excel file the team fills out the same way every project (buyout logs, SOV trackers, proformas) becomes real software with automations and alerts.

↻

The flywheel: estimate → build → capture actuals → estimate sharper next time. Vendors sell tools; none of them can sell Dawson this loop, because it runs on Dawson's data. The method: consolidate scattered data into one truth, then put the right surface in front of each role: exec, PM, estimator, superintendent.