Baltimir now prepares KVVK estimate drafts with AI in about 30 minutes.
Book a 20-min working callProof at a glance
Savings compare Baltimir's previous manual estimate drafting routine with the AI workflow implemented after two working sessions. A human still reviews every draft before it is sent.
The Context
Baltimir Group is an Estonian construction company preparing detailed KVVK estimates for water supply, sewerage, heating, ventilation, underfloor heating and related building-services work. Each project arrives with PDFs, DWG drawings, specifications and an Excel workbook, so the sources and workbook rows have to stay traceable.
The Problem
Every estimate repeated the same expert logic.
The work followed a clear pattern, but the rules lived in the estimator's head and had to be reapplied project by project.
A generic AI chat was not reliable enough.
It could summarize documents, but it could also produce unsupported quantities, miss practical work rows or hide uncertainty inside a total.
The bottleneck was expert time.
Several hours went into preparing a draft before the real commercial review could start.
An AI workspace for KVVK estimates
In two working sessions, we turned Baltimir's estimating logic into a reusable AI workspace (built on OpenAI Codex): templates, rules, reference examples, source-to-row mapping, drawing checks and review gates.
What we built
Set up a repeatable project folder for every new estimate.
Documented KVVK-specific rules so the AI follows Baltimir's estimating logic.
Added source-to-workbook and drawing checks before Excel is edited.
Added a clear review status: ready for review, draft only or blocked.
The mechanics
Financial Impact
The AI prepares the estimate draft and separates the points that need human review.
The estimator starts from the prepared workspace, the AI prepares a marked draft, and the human review stays explicit.
The value comes from one focused setup that repeats on every future estimate.
*Simplified visualization of the actual solution delivered.