Goldsmith Growth
Case Study Construction · Automation · Procore · Bluebeam

Cutting document processing time by 70%. Validated against every submittal and change order on a $50M multifamily project

The build that can give PMs back 5+ hours a week

70%
Time reduction
measured in testing
500+
PM hours recoverable
per project
3,500+
Real documents
processed
01 / The Discovery

“How in the hell can I eliminate some keyboard strokes?”

VP of Operations, construction firm

That question, asked unprompted by an ops leader at another company entirely, is the whole problem in one sentence. This pain isn’t one developer’s quirk; it’s industry-wide.

When I sat down with the lead PM on a completed $50M multifamily project, he put it another way: “I spend half my week doing paperwork, not managing the project.”

Then he showed me. Every one of the project’s 2,500+ submittals ran through the same mechanical pipeline: download from Procore, hunt through a 900-page spec book for the relevant section, build a cover sheet by hand, set up a Bluebeam review session, invite reviewers. The same information typed three times: into Procore, into the cover sheet, into Bluebeam: because none of the tools talk to each other.

So I built the automation layer, working directly with that PM: and then tested it against the project’s complete historical document set to see exactly how much time it gives back.

02 / Scope

What I am (and am not)

×
Not VC-backed.

I answer to my clients, not a board.

×
Not “Procore 2.0 with AI.”

Procore stays Procore. Bluebeam stays Bluebeam.

×
Not another platform.

No new logins, no new dashboards, no retraining.

×
Not an agency that disrupts your workflow.

I learn how your people actually work, then build underneath it.

03 / The Problem

Not a crisis. Just Tuesday.

On the $50M project, the lead PM was losing hours of skilled time every week to administrative work that required zero judgment. Not a crisis. Just Tuesday. But across a project lifecycle, it compounds into months of skilled people doing data entry.

Submittals: the biggest time drain

~45 minutes each, times 2,500+. Download, classify against a 900+ page spec book, create the cover sheet, set up the Bluebeam session, invite reviewers. Pure pipeline work.

Change orders: the biggest margin risk

~10 minutes each of manual re-entry and mental-math rate checking, times 1,000+. Every uncaught discrepancy is money that quietly leaves the job.

At $50,000/day to keep a project site running, a PM losing hours a week to paperwork isn’t a productivity issue: it’s a margin emergency.

04 / The Solution

Two workflows, one silent automation layer.

I lead with change orders: easiest to demonstrate live, and the one that directly protects margin. Submittals are the larger time-savings engine.

Change order processing

PDF arrives Extract line items Validate against rate cards Flag anomalies PM approval Procore API

A change order PDF lands from a subcontractor. The system extracts every line item, validates each rate against your rate cards and historical patterns, flags discrepancies, and delivers a ready-to-approve summary. The PM reviews and approves; the system logs to Procore with a full audit trail.

Submittal processing

Submittal received Download + classify Cover sheet generated Bluebeam session created Reviewers invited Notifications sent
↑ Spec book (context) ↑ Reviewer matrix

The system watches for new submittals, classifies each against the project spec book, generates cover sheets with spec cross-references, spins up the Bluebeam review session, and invites the right reviewers: automatically. The PM guides the actual review. The machine handles everything before it.

The principle across both: the PM still approves everything. Human judgment stays where it belongs. The busywork underneath disappears.

05 / The Results

Timed, not estimated.

I ran the automation against the project’s complete historical dataset: every submittal, every change order, every spec sheet: processing them exactly as if the project were live. These are timed results from thousands of real documents, pressure-tested across repeated runs.

Metric Manual Automated Reduction
Submittal processing ~45 min ~13 min ~70%
Change order processing ~10 min ~3 min ~70%
PM hours recoverable 5+ hrs/week

32 minutes saved per submittal across 2,500+ submittals = over 1,300 hours. Add ~7 minutes per change order across 1,000+ change orders and the total clears 500+ recoverable hours per project: returnable to managing the job instead of managing documents.

Why test on a completed project?

Because it’s the only way to measure honestly. A finished job gives you the complete document load: every submittal, every change order, start to finish: plus the ground truth of how each was actually handled, so every automated output could be checked against the real human decisions. A live project can’t give you that rigor. The one thing this isn’t is a production deployment over a full project lifecycle: and that final proof is precisely what the pilot delivers: on your project, with your documents, at production stakes.

06 / How It Was Built

Four stages. No shortcuts.

01

Extract the tribal knowledge

I interviewed the PM closest to the pain, mapped every step of his actual workflow, and found the highest-volume friction: not what leadership assumed was broken.

02

Codify it into a repeatable system

I documented the ideal workflow, mapped every decision point, and separated where human judgment is essential from where busywork had crept in.

03

Automate underneath the existing tools

API-driven, no new software, working inside Procore and Bluebeam. Built and validated in weeks, not months.

04

Layer on visibility (roadmap)

Dashboards and signals so ops leaders can see how the team is actually performing: delivered after the core engine is proven, not before.

Most AI projects in construction fail because nobody did the unglamorous integration work. The tool is usually fine. What’s missing is the boring part: deciding exactly which task it replaces, showing the crew how it fits, and sticking around until the new habit beats the old one. That boring part is my entire job.

07 / Technical Foundations
API-driven, no new software.

Works inside Procore and Bluebeam. No new logins, no training required.

Stateful, not stateless.

The system remembers prior decisions: rates, reviewer assignments, spec classifications. It doesn’t re-ask questions it already knows the answer to.

Human-in-the-loop on every approval.

The automation extracts, classifies, and assembles. A human approves every output.

Client-owned.

Code, documentation, training: you own everything I build.

No data stored outside client systems.

Full audit trail on every automated action.

08 / What This Means

What this means for an ops leader

01

Change order risk gets caught early. Automated rate validation flags discrepancies before they become margin erosion.

02

Submittal turnaround drops from 45 minutes to 13. PMs spend their time on the actual review, not the pipeline in front of it.

03

More projects per PM. When those hours come back, one PM carries more work without burning out.

04

Junior staff actually develop. Project engineers learn project management instead of typing the same information three times.

Experienced people are retiring out of this industry faster than they’re being replaced. When you can’t hire your way out, the only lever left is making the people you have dramatically more effective. That’s not AI hype. It’s operational math.

See this on your project

A single-workflow pilot on one of your projects. I scope the workflow, define success metrics, and build. Results guaranteed: you only pay for what delivers.

Start a pilot

Live within 7–10 business days of receiving
Procore sandbox access and sample documents

Related services
Change order automation → Submittal automation → More case studies →