AI-Assisted Proposal Generation for an International Architecture Studio

Every new hotel project starts with the same question: how do we budget this accurately enough to write a competitive proposal? Producing a proposal meant estimating design hours across all phases, writing scope and deliverable descriptions, and building a fee table, all requiring individual judgment, manual comparison with past projects, and significant project manager time. The knowledge existed, but it lived in people’s heads and in scattered files. The process was slow, hard to delegate, and produced inconsistent results.

Before building anything, we made the decision logic explicit. We mapped the full proposal process into an AI Process Playbook, reviewed 15+ completed hotel projects, and worked with the team to surface the estimation and writing rules they were already using. This included careful alignment between how hours were tracked internally and how phases were labelled in proposals, and structuring a project database covering room count, square metres, scope, design type and level of complexity for each project.

We developed a Custom GPT agent trained on the studio’s own proposal history and logic. When a new client brief arrives, the team inputs the key parameters among others: hotel category, scope, design type, topographic complexity and size parameters. The agent produces three things: a draft hour breakdown across all five design phases, full proposal text including scope, deliverables, assumptions and exclusions per phase, and a draft fee table translating estimated hours into a structured pricing overview. The result is a complete proposal draft, ready for partner review.

Making estimation and writing logic explicit required articulating things the team had never needed to say out loud before. The first outputs were confidently wrong. It took multiple iterations and extensive testing to reach something reliable. We validated the model against past proposals, comparing AI-generated estimates and text to what the studio had actually budgeted, delivered and written, refining the logic until outputs were consistent and trustworthy.

What previously took 3 to 10 days now starts in under an hour. A complete proposal draft, covering hours, text and pricing, is generated the moment a brief comes in. The studio has a reliable starting point that captures their own logic rather than a generic formula, with a documented view of how estimation and scope description shifts across design type choices, useful for pricing, capacity planning and onboarding. And the best part: the system learns from past proposals and every new one coming in, becoming more accurate over time. The only step that remains manual is formatting: dropping the generated content into the Word template and applying the final layout.