This scenario plays out on construction sites every day. The average commercial project generates 40 to 60 change orders, and most of them start as verbal commitments during site walks, coordination meetings, or quick conversations in the field. By the time anyone formalizes them, the critical context — who said what, when, and under what conditions — is gone. The result is disputes, cost overruns, and relationships that deteriorate with every "he said, she said" argument that goes to arbitration.
The gap is not a tooling problem in the traditional sense. Project managers have plenty of software. What they do not have is a record of the conversation that produced the decision. And in construction, the conversation is the decision.
The Documentation Gap
Construction projects involve dozens of stakeholders: owners, architects, general contractors, subcontractors, engineers, inspectors, building officials. Every site meeting generates decisions that affect scope, schedule, and budget. Project managers are expected to capture all of it — verbal commitments, design clarifications, coordination decisions, RFI follow-ups — and turn that capture into a written record the entire team can rely on. Most fall back on handwritten notes, memory, or hastily typed summaries that miss the details that actually matter.
The numbers are brutal. The average construction project manager spends 15 to 20 hours per week in meetings and site walks. They are simultaneously expected to manage the conversation, watch the work in progress, and document every commitment. Taking detailed notes while running a coordination meeting is impossible, so they write down what they can remember later — and they miss the nuances that decide who pays when a dispute surfaces.
The stakes are high. Construction disputes cost the U.S. industry tens of billions of dollars annually, and most of those disputes start in the same place: a documentation gap. A verbal agreement that was never captured. A design change that was never formalized. A coordination decision that different parties remember differently. When a $2 million project ends up in arbitration over a $15,000 change order, the documentation gap stops being a paperwork problem and becomes a business-critical one.
Why Current Solutions Fall Short
Field notes are incomplete. Project managers cannot write fast enough to capture everything said in a site walk. They jot down key points, miss the surrounding context, and reconstruct the conversation later from memory. By the time they type up meeting minutes, the parts that matter — the conditional language, the qualifiers, the moment a sub agreed something was inside scope — are already smoothed over.
Meeting minutes arrive too late. Most PMs send meeting minutes 24 to 48 hours after the fact. By that point, participants' memories have diverged. The electrician remembers agreeing to move two boxes, not three. The architect remembers suggesting a solution, not approving one. The minutes that were supposed to be the source of truth become another point of dispute.
Voice recorders create transcription work. Some PMs record site meetings on a phone or a dedicated recorder. The recording captures everything — but raw audio is unsearchable and unusable. Transcribing a 45-minute coordination meeting takes 2 to 3 hours of focused work nobody has time for. Most field recordings sit on a device, never reviewed, until a dispute forces someone to scrub through hours of audio looking for one critical exchange.
Video meeting tools do not work in the field. Zoom and Teams recorders work great for office meetings. But construction happens on site — in noisy environments, with multiple speakers, often while walking through active work areas. Video meeting bots cannot capture these conversations, and most site meetings do not happen on Zoom in the first place.
What Effective Documentation Looks Like
Real-time transcription that works in the field
AmyNote runs on the project manager's phone and captures conversations as they happen — no bot, no video call required. Transcription runs through OpenAI's latest Speech API, which handles construction terminology accurately: RFI, submittal, punch list, AIA billing, substantial completion, liquidated damages, schedule of values, retainage. It works in noisy environments and does not require participants to speak into a microphone. The PM can put the phone in a shirt pocket and walk the site.
A project manager walks a site with three subcontractors. AmyNote captures the entire conversation, timestamps every statement, and identifies speakers. When the HVAC contractor says "we can move the ductwork if the electrician relocates those two junction boxes," that sentence is captured verbatim with a timestamp. No handwritten notes. No memory gaps. Just a searchable, timestamped record of who said what.
Speaker identification across multiple meetings
Construction projects involve the same people across dozens of meetings spread over months. AmyNote's speaker identification learns voices across sessions. Once it knows who the electrical sub is, it labels their statements correctly in every future meeting — even months later when the dispute surfaces.
This is the difference between "someone said we'd include that in the base scope" and "John from ABC Electric said on March 15th at 10:47 AM: 'We'll include the junction box relocation in our base bid.'" That level of specificity changes the conversation. Most disputes never escalate to arbitration once the verbatim transcript surfaces.
AI summaries that capture commitments
After each meeting, AmyNote generates a structured summary using Anthropic's Claude Opus. The summary identifies action items, decisions, and commitments — the things that matter for the project file. It captures who committed to what, by when, and under what conditions, and it formats the output so a PM can paste it into the daily report or the weekly progress meeting minutes without rewriting.
For a 45-minute coordination meeting, a typical summary highlights three scope changes discussed, seven action items with owners, two schedule impacts, and four follow-up RFIs. The PM reviews the summary, confirms accuracy, and sends it to participants within 15 minutes of the meeting ending — while everyone's memory is still fresh and the same as it was an hour ago.
Privacy architecture for multi-party projects
Construction projects involve sensitive information: bid pricing, schedule commitments, design changes, contractual obligations. AmyNote's architecture: transcription runs through OpenAI's Speech API, AI analysis is powered by Anthropic's Claude Opus, and both providers contractually guarantee that user data is never used for model training. Audio is encrypted in transit, processed, and not retained on provider servers. Transcripts and recordings are stored locally on the project manager's device with end-to-end encryption.
No project conversations sitting on a third-party server. No bid details feeding into model training pipelines. No data retention by AI providers after processing. When an owner's legal team asks where the project's audio archive lives, the answer is on the PM's encrypted device — and that is an answer most tools cannot give.
The Productivity Impact
| Before AmyNote | After AmyNote | |
|---|---|---|
| Time in meetings & site walks | 15-20 hrs/week | Same |
| Time typing minutes from memory | 5-8 hrs/week | ~30 min/week reviewing AI summaries |
| Disputes per project over verbal commitments | 3-5 | 0-1 |
| Time to find a past statement | 2-3 hrs scrubbing | 10 minutes via semantic search |
The time savings matter. But the bigger impact is dispute prevention. When every verbal commitment is captured with timestamps and speaker attribution, "he said, she said" arguments disappear before they escalate. Change orders get documented in real time. Scope creep gets caught the day it shows up, not the week it bills. Project managers spend less time in arbitration prep and more time building.
Choosing the Right Tool
When evaluating AI transcription tools for construction project management, prioritize these criteria:
- Zero-training guarantees. Construction conversations contain bid pricing, schedule commitments, and contractual details. Verify that your transcription provider contractually guarantees no model training on your data.
- Works without video calls. Most construction conversations happen on site, not on Zoom. The tool needs to capture in-person meetings, site walks, and field conversations.
- Construction terminology accuracy. Generic transcription tools mangle industry terms. Test the tool with real project conversations before committing.
- Speaker identification across sessions. Projects involve the same people across dozens of meetings. The tool should learn voices and label speakers consistently from week one through closeout.
- Searchable transcripts. When a dispute arises, you need to find specific statements quickly. Semantic search across all project meetings is essential — keyword search alone is not enough.
- Fast turnaround on summaries. Meeting minutes that arrive 48 hours later are useless. The tool should generate summaries within minutes so you can send them while memories are fresh.
Getting Started
AmyNote provides real-time transcription powered by OpenAI's Speech API and AI analysis by Anthropic's Claude Opus. Both providers contractually guarantee zero training on user data. All transcripts are stored locally on your device with end-to-end encryption.
3-day free trial. No credit card required. Works on iPhone and iPad. Use it on your next two coordination meetings and one site walk, and check what the transcript captured against what you would have written down. The gap will surprise you.
Originally published as an X Article: The Change Order Documentation Gap on X.
