---
title: "Politikos"
url: "https://www.omnicliq.com/en/case-studies/politikos-shop-attribution/"
description: "How two probability models showed Politikos Shop which channels bring real sales and moved the budget there, every day."
category: "ecommerce"
client: "Politikos"
datePublished: "2026-04-18"
author: "Dimitris Bachtsevanis"
locale: "en"
---

# Politikos

> How two probability models showed Politikos Shop which channels bring real sales and moved the budget there, every day.

## Project

**Client:** Politikos
**Industry:** Fashion
**Duration:** 2019–2020
**Services:** Analytics, ML & AI, Google Ads
## Key results

- **+231%** Revenue
- **+225%** Orders
- **2** AI models
- **52** Refreshes
## Last-click points to the wrong channels as the source of sales

Once the Politikos Shop account was automated and growing, the question was where the next euro should go. Platforms give the whole sale to the last click.
A fashion buyer usually sees a product on Instagram, finds it again in a Shopping ad and buys days later after a brand search. Last-click sees only the final step.
So campaigns that bring new buyers look ineffective and get cut, while channels that would have closed the sale anyway take the budget.
The account narrows without anyone noticing: fewer new buyers today means fewer brand searches tomorrow. We needed measurement that sees the whole path.

## We measure each channel's contribution and split the budget accordingly

We brought buyer paths from GA4, Google Ads and Meta into one data warehouse, without double counting across devices.
A Markov Chain model reads every path before a purchase and measures how much sales would drop if each channel were missing. That is its real contribution.
A Multi-Armed Bandit algorithm splits the budget every day where the expected profit is highest, while testing lesser-known campaigns to learn how they perform.
Every budget change comes with an explanation for the team, and attribution is refreshed weekly. Decisions follow the latest data and stay under human review.

## Results

- **Revenue** — More than three times the revenue, from campaigns that last-click would have judged ineffective and cut.
- **Orders** — More purchases from audiences that were undervalued before.
- **AI models** — Markov Chain (attribution) and Multi-Armed Bandit (budget) run continuously. They are our own models, not licensed software.
- **Refreshes per year** — Attribution refreshes every week, 52 times a year, so the budget always follows the latest data.