---
title: 'New York Build 2027 AI sessions: test the preconstruction workflow behind the demo'
url: 'https://piper-ai.com/resources/new-york-build-2027-ai-preconstruction-evaluation'
description: 'Test an AI preconstruction demo at New York Build using a controlled package, revision and trade quote, with separate checks for evidence and human review.'
contentTypes: [guide]
publishedAt: '2026-11-17T09:38:29.101Z'
updatedAt: '2026-11-17T09:38:29.114Z'
author: 'Ido Gedanken, CEO'
readingTime: 3
---

# New York Build 2027 AI sessions: test the preconstruction workflow behind the demo

Bring one bounded scope decision to an AI demonstration and score whether a reviewer can verify it after a revision. A compelling answer on a frozen sample does not show what happens to an exclusion, its source and the GC's carry when the package changes.

This is an independent Piper planning guide. The organizer lists [New York Build](https://www.newyorkbuildexpo.com/exhibitors/new-york-build) at Javits on March 2–3, 2027. Its [AI practical-plan workshop](https://www.newyorkbuildexpo.com/new-york-build-conference-agenda-2027/workshop-ai-action-practical-plan-built-environment) and [preconstruction-tools panel](https://www.newyorkbuildexpo.com/new-york-build-conference-agenda-2027/panel-preconstruction-tools-changing-way-new-york-gets-built) are scheduled March 2. Check the organizer for registration and current times; this guide was checked November 15, 2026. The following test is Piper's, not a promised session exercise.

## Construct a test with a known answer and an unknown

Illustrative example: a mechanical duct penetrates a rated wall. The original specification assigns firestopping; a later addendum changes the relevant detail; the mechanical quote excludes firestopping. Before the demo, an estimator identifies the documents and the unresolved trade assignment. Do not assume the revised detail settles the assignment without reading it.

Use one company-approved, shareable package: original drawing and spec, the revision, and a trade quote. Label each version and preserve an untouched reference set. If permission to share is absent, use a synthetic package or defer the test. Give each provider the same inputs and task; ask them to identify the gap and show the evidence, without supplying the expected answer first.

| Test dimension | Pass evidence to seek | Record if missing |
| --- | --- | --- |
| Source citations | Correct document and location open to the reviewer | Untraceable or wrong-version finding |
| Revision handling | Changed finding and version are visible after the addendum | Old conclusion persists or change cannot be inspected |
| Unresolved items | Exclusion and assignment ambiguity remain visible | Unsupported certainty or omitted gap |
| Human review | Reviewer can accept, correct or escalate the finding | No clear accountable decision |
| Usable output | A scope or leveling item can be transferred with its evidence | Output loses context at handoff |

Score each dimension separately as demonstrated, partial or not demonstrated, with a note and source. Do not average away a missing citation with an attractive interface. Ask what was included, omitted or manually edited during the demonstration. An inability to demonstrate a capability in the available time is an open question, not proof of failure.

## Repeat after the addendum

First review the original result. Then introduce the revision and ask what changes, what stays open and how an estimator finds the prior conclusion. Compare the result to the reference set, including the quote exclusion. Have the estimator decide whether to solicit clarification, make a documented carry or escalate the interface. The [addenda management guide](https://piper-ai.com/resources/construction-addenda-management-for-gc-estimators) covers the wider document-control process.

A promising demonstration earns a bounded, authorized pilot with named reviewers and an error log. It does not settle security, procurement or performance across projects. Use the broader guide to [evaluating AI preconstruction software](https://piper-ai.com/resources/evaluating-ai-preconstruction-software), the event planning guide, and the distinct AGC pilot charter.

Before scoring, distinguish [AI scope-generation tool categories](https://piper-ai.com/resources/ai-scope-generation-tools-compared). A takeoff demonstration and a source-backed scope review answer different questions, so the same attractive output is not evidence that both jobs are covered.

## Keep the quote connected to the scope

[Piper Bid Leveling](https://piper-ai.com/use-cases/bid-leveling) compares subcontractor proposals with intended scope and links findings to sources. Apply the same review standard: the GC still needs to inspect the project evidence and decide how to treat the gap.
