Custom AI-powered system for GRUPA KRES — 100 orders a month in one pipeline

GRUPA KRES — a stoneworking company with a 20-person team — handles roughly 100 orders a month. Each order is several to a dozen steps: customer inquiry (phone, website form, email), scheduling an on-site measurement, taking the measurement, building the quote, customer approval, workshop order (cutting and stone processing), quality control, transport, on-site installation, final documentation. Each step historically lived in a different tool: landline phone, paper reception notebook, calculation Excel, Google Calendar, paper on the workshop floor, invoices in the accounting system.

The day-to-day cost showed. A customer would call asking about the status of their order and the reception had to find a paper card, check the calendar, sometimes phone the workshop. Time from first customer inquiry to ready quote averaged 5 days — inquiries reached different people, waited for a measurement (taken by 1 of 2 measurers, each with their own calendar), then waited for calculation (which depended on the availability of a specific person who knew the material prices).

An additional issue: pricing was inconsistent. The same projects — for example a 240×60 cm polished granite countertop — were calculated by different staff to different prices, because everyone used their own version of the material price list and their own margins. Some quotes were too low (the firm lost margin), others too high (the customer went to a competitor). Leadership wanted to standardize the quoting process without hiring a dedicated calculator.

Weeks 1–3: workshops with every role at GRUPA KRES (owner, reception, measurers, calculator, technologist, installers) plus discovery on a sample of 200 historical orders. Out came the full process map (nine stages, six roles) and the pricing knowledge base: 320 stone types with current supplier price lists, 18 edge-finishing variants, 12 surface-finish types, and calculation formulas accounting for installation specifics.

Build took 9 weeks. Backend (Node.js + PostgreSQL) with a full order model — from inquiry to final documents. AI layer (OpenAI GPT-4o) handles three things: parses incoming inquiries (phone transcription → text, web form, email, WhatsApp) and extracts structured project data (product type, dimensions, material, finishing); generates a preliminary quote from historical orders plus the current material price list; produces a ready quote PDF for the customer with scope and delivery timeline. Frontend (React) for the team with four views: daily reception pipeline, measurement calendar, workshop queue, installation schedule. A mobile app for installers handles on-site photo documentation at the customer.

The last 2 weeks were a staged rollout with hypercare. Week one — every new inquiry ran in parallel through the old process and the new system, with every AI-generated quote verified by the calculator. Week two — switch to the new system as primary, the old one as backup. Week three with no parallel — full trust in the system, with a 15-minute daily team stand-up to catch edge cases and fix them in the app the same day.

Time from first customer contact to ready quote dropped from an average of 5 days to one. A morning phone inquiry is parsed by AI into structured fields, the system automatically suggests a measurement slot based on the calendar, after the measurement the calculator receives a preliminary quote ready to accept — just verify and send to the customer. That directly improved conversion — customers who previously walked to competitors during the 5-day wait now buy at GRUPA KRES because they get the first quote.

Pricing became consistent. AI uses a single source of truth — the current material price list and the historical-orders base — so two different calculators reach the same quote for the same project. Quotes are no longer too low (the firm recovered margin it previously gave away) or too high (customers no longer leave for price). Margin per order rose by an average of 6 percentage points within six months of rollout.

Customers see their status in a portal accessible via a link in the confirmation email. „Your order is in the cutting stage at the workshop, planned installation: Thursday 18:00” — no need to call reception. Customer NPS rose by 28 points in the first quarter post-rollout, and reception fields far fewer „how's my order going?” calls, which freed them up for active work — contacting customers about repeat orders and post-sale service.