Where AI has actually been useful in construction

The principle behind every win

Do not ask AI to be your calculator or your price book. Make it read, cross-check, ask for what's missing, and draft — and have software do the arithmetic while your own database holds the prices. Every tool that followed that split got used. Every tool that didn't got quietly abandoned.

Estimating and budgets

Catching math and ratio errors before they go out

Recalculates an estimate from the ground up — line extensions, section totals, markup, contingency, grand total — with the arithmetic done in code, not by the AI. Every discrepancy is reported with its source cell. Built after a real ROM review surfaced a cost-per-square-foot figure on a competitor's budget that came from a stale area number — caught by hand, late, by an experienced estimator. That gap is the whole product.

Asking for the missing information first

A senior estimator's first move on a new request is never to estimate — it's to ask for the five things he needs. We encoded the firm's own estimating conventions so the assistant identifies exactly what's missing before producing a number. It makes seniority transferable to junior staff.

A company cost book with provenance

The firm's own unit prices and buy-out rates, each with a required source and full history. The AI can read from it and propose updates — but never supplies prices itself, and every update requires a source and human approval. Pricing knowledge stops living in two people's heads.

Comparable projects and escalation context

Closed budgets kept as comparables with stale-basis warnings; published producer price indexes tracked as escalation context — and labeled that way, never as bid-ready pricing. "What did we see on the last one like this?" finally has a fast answer.

Starting from the real template, merging without overwriting

New budgets start from the firm's actual template structure. A peer's budget merges into yours by adding the lines you're missing — flagged and annotated — while your numbers stay where the two disagree. Nothing is silently overwritten.

Drawings and scope

Comparing drawing revisions

Two revisions of the same sheet, compared automatically, with findings marked directly on the drawing and saved for recall. Revision churn between bid set and permit set is where scope quietly moves — and where it otherwise surfaces later as a change order.

Reconciling quantities against the estimate and the budget

Tagged drawing counts, estimate line items, and the project budget — three sources of truth nobody has time to cross-check — reconciled deterministically, with AI used only to interpret how a particular estimator labeled things.

Checking a proposal for what it does not say

Reads a subcontractor proposal against the firm's own scope definitions and marks each item Present, Missing, or Ambiguous. The risk in a sub proposal is rarely a wrong number — it's an unstated exclusion that reviewers can't see because it isn't on the page.

Communication and operations

Meeting minutes as real records

Transcripts from the meeting recorder the company already uses become structured minutes — agenda topics, attendance matched to people records, action items as checklists — written into the project management system after explicit human approval. The single most-requested tool from the team.

Drafting owner correspondence

Client-facing emails explaining a number, drafted with the framing an experienced estimator would use — the difference between "our number is higher" and "here is the scope reconciliation." The tool drafts and stops. It never sends.

Job-site content for marketing

Social post suggestions drawn from the project's real activity timeline and site photos, with specifics cited so a person can verify before publishing. The raw material was already being captured every day and never used.

For electrical, mechanical, and specialty contractors

Most of this translates directly to trade contractors:

• Deterministic checking of your own bid math before it goes out
• Proposal and scope-gap review — especially what a spec does not say about your scope
• Drawing revision comparison, where most trade change-order exposure originates
• A cost book with provenance for your assemblies and buy-out rates
• Meeting minutes and owner/GC correspondence drafting

Quantity reconciliation adapts too — for a trade contractor the count is fixture and device schedules against your takeoff rather than tagged drawing counts against a division-based estimate.

Questions this work already answers

How should AI handle construction estimating math?

Software does the arithmetic from the estimate itself: line extensions, section totals, markup, contingency, and the grand total. AI reads, cross-checks, and reports discrepancies with a source cell. It does not invent prices.

What is a company cost book with provenance?

The firm's own unit prices and buy-out rates, each with a required source and a full history. AI can read from it and propose updates. Every update needs a source and human approval. The AI never supplies prices.

How does drawing revision comparison work?

Two revisions of the same sheet are compared automatically. Findings are marked on the drawing and saved. That is where scope moves between bid set and permit set.

How do you check a subcontractor proposal for missing scope?

The proposal is read against the firm's own scope definitions. Each item is marked Present, Missing, or Ambiguous so unstated exclusions are visible.

How an engagement works

We embed with your team — your Procore, your standups, your job sites — through a paid discovery phase that maps where hours and margin actually leak, then a build phase that ships working software from the first weeks. The engineer who scopes the system is the engineer who builds it. More on how we work.