# GuidedTours: studied, then rebuilt as one system

> Tourism agency · pilot. A tourism agency studied first, then rebuilt around one operating system: enquiries, bookings, calendar, customer portal and a site built from data.

- Page: https://plexus.al/work/guidedtours/
- Language: English
- Albanian version: https://plexus.al/sq/work/guidedtours.md

- Services: [Business systems](https://plexus.al/services/business-systems.md), [AI & automation](https://plexus.al/services/ai-automation.md)
- **Problem:** Enquiries in inboxes and chats, ads pointing to the homepage, no clear link from an ad to an enquiry.
- **Built:** Staff and customer portals, one inbox, a pipeline, and a new site generated from data.
- **What changed:** Over 60 enquiries in three months handled in one place, each with its source.

## Problem

Enquiries arrived by email, forms and WhatsApp in several languages and lived in inboxes and chats; one was lost when the mail watcher stopped. 26 of 28 ads sent visitors to the homepage. The website was slow, and it was hard to tell which ad brought which enquiry.

**What had to change:** Everything was studied before building: how enquiries arrive, which pages convert (specific pages converted at 16.2 % against 6.5 % for the homepage), which keywords bring traffic, and what makes a booking worthwhile. Rules decided before code: no price without its basis, a source for every numeric fact, the site before the ads.

## System

A staff portal with one inbox, leads and pipeline, bookings, a calendar used as the daily work planner, statistics and a website editor; a customer portal opened with a booking number; a mobile app; and a new website generated from structured data.

**Connected:** The business mailbox (IMAP/SMTP), Google Ads click IDs, GA4, Cloudflare Pages and Functions, DataForSEO, and a WordPress plugin on the current site.

**Automated:** Lead intake with attribution and 30-day de-duplication; a mailbox sweep every 5 minutes; AI-written content checked against sources before publishing (113 factual errors caught); AI-search visibility measured.

## Implementation

Every enquiry is captured with its source and ad click ID and de-duplicated; the mailbox is watched around the clock; the pipeline moves on replies and payments; the site is built from a data model of tours, destinations and places, with modern image formats, structured data and redirects that keep click IDs; a gate stops any deploy that fails a check.

## Result

Over 60 enquiries in about three months handled in one place, most of them carrying the Google Ads click that brought them, and 4 from ChatGPT. The new site scores 92–97 for performance and 100 for accessibility on Lighthouse mobile, in preview; it is not live yet.

## Technical detail

**Architecture in detail:** A domain model of 20 concepts and 22 relations and a register of decisions; a staff portal and a customer portal on the business’s own domain; payment amounts computed on the server; production kept out of the portal’s reach by design.

**Layers:** Business, Workflow, Software, Data, AI, Integration

**Stack:** NestJS, React, PostgreSQL, Static site generator, Cloudflare Pages, Cloudflare Functions, IMAP / SMTP
