AI-Powered Estimation System: From 20-Minute Spreadsheets to 2-Minute Quotes
A concrete cutting company does over 1,300 estimates a year across six service types. All of it ran through a Google Sheets calculator that somebody spent two months building. Every estimate took 15 to 20 minutes. About 3 to 4% of estimates went out wrong.
Challenge
The Problem
This concrete cutting company does over 1,300 estimates a year across six service types. All of it ran through a Google Sheets calculator that somebody spent two months building. Every estimate took 15 to 20 minutes. About 3 to 4% of estimates went out wrong, roughly 40 bad quotes per year.
The spreadsheet was fragile. Formulas broke when columns shifted. Different team members produced different results for the same job. There was no version control and no way to validate a quote before it went out. Management had no visibility into what was being estimated or how.
At 1,300+ estimates a year, even a small error rate adds up fast. The real cost wasn't just the bad quotes. It was the time spent checking work that should have been trustworthy from the start.
Solution
What We Built
A four-phase system that replaced the spreadsheet entirely, giving the team two ways to generate estimates depending on the situation.
AI Data Extraction
The original spreadsheet's pricing logic was extracted and formalized. Every formula, lookup table, and service-type rule was audited and converted into structured configuration data that could actually be tested and maintained.
React Web Application
A structured input form for all six service types replaced the spreadsheet. Fields validate in real time. Pricing calculates instantly. Output is a formatted quote ready to send, not a cell reference chain that might be broken somewhere upstream.
Automated Test Suite
Coverage across all six service types means price updates don't introduce silent regressions. When the office manager edits the config table, tests run automatically and confirm the numbers are still correct.
AI Chat Agent via MCP Server
A second interface for when someone is on the phone. Type a natural language job description and get a formatted estimate back in seconds. The chat agent uses an MCP Server to connect to the same pricing engine as the web app, so both interfaces always produce consistent results.
Results
The Impact
2 to 3 minutes per estimate instead of 15 to 20. Zero calculation errors since launch. 325+ hours freed up annually. The office manager can update prices by editing a config table in minutes, with tests confirming nothing broke.
The two-interface approach matters. When someone calls in a job, pulling up the web form while on the phone adds friction. The chat agent removes that friction: describe the job in plain language, get the number. Same pricing engine underneath, different interaction model on top. For context on how to calculate the business case for automation, the methodology used here applies directly. See also another AI system handling real-time business interactions.