I used to check my phone at dinner because I was scared of missing something. I do not do that anymore.
Summary
A Europe-based construction business owner was managing six multilingual Gmail accounts, PDF invoices, and a large financial tracker in Google Sheets entirely by hand. Mornings disappeared into triage before real work could begin, and payment deadlines slipped on a regular basis.
WiserBrand built a private AI assistant that reads every inbox around the clock, flags what actually needs attention, and answers financial questions on demand. It runs on the client’s own Mac Mini and is accessed through Telegram, by text or voice, in any working language. One-time fee, no subscription, full ownership.
Cooperation Period
One-time build, 3 weeks
Location
Industry
Technology Stack
Business Challenge
- Email ran the morning before work did
Six Gmail accounts in German, English, Italian, and Polish, covering suppliers, subcontractors, clients, and authorities. One to two hours of manual triage every morning. Messages buried in the fifth or sixth inbox could sit a full workday before anyone saw them.
- Payment deadlines lived inside the PDFs
Due dates arrived buried in attachments, with nothing pulling them into a single view. Two to four deadlines slipped every month, costing late fees and straining supplier relationships.
- Financial answers took minutes, not seconds
“How much have we paid Berger this quarter?” meant opening the Sheet, filtering, scrolling through hundreds of rows, and summing manually. Several times a day.
What We Did
The assistant runs a simple four-step loop around the clock: read, classify, flag, notify. Below is how the system was built and delivered.
1
Mapped the real workflow
Before building anything, we audited dozens of real invoices, contracts, and fine notices. Tuning the system to actual vendor names, local VAT formats, and site authority terminology was the difference between a demo that works and a tool the owner would use every day.
2
Built the three-agent assistant
An orchestrator handles all user interaction. An email agent scans the inboxes every four hours, classifies urgency, and extracts payment details from attachments. A sheets agent parses financial queries. Two cloud models (GPT and Gemini) cross-check critical decisions like payment flagging to reduce hallucination risk.
3
Deployed to the client’s hardware
The system runs on the client’s existing Mac Mini (8 GB RAM). Credentials stay on the device. Nothing flows through WiserBrand servers. All integrations are read-only by default: the assistant never sends email or modifies data without explicit confirmation.
4
Tested and tuned on the live workload
Before handover, the system ran against real email volume for two weeks. Classification thresholds were adjusted based on what the owner flagged as important. Delivery included a 30-day post-launch support window.
Project Results
Three months in, the recovered time goes to site visits and client calls, not inbox triage.
Mornings reclaimed
Email stopped being the first task of the day
Zero missed
Payment deadlines no longer slip
Languages handled
Multilingual mail stopped being a bottleneck
Full ownership
No subscription, no lock-in
Client Review
The owner did not set out to build a tech-forward business. He just wanted his mornings back.
