Most marketing teams believe their programmatic advertising is "data-driven". It is not. It is data-assisted, at best. Teams pull levers inside the walled gardens of Google's DV360 or The Trade Desk, guided by last-click attribution models that are fundamentally flawed. They manually test a handful of creative assets, celebrate a 5% uplift, and call it optimisation. This is not the future. It’s a digital assembly line, and it’s about to be dismantled.
We followed one UK direct-to-consumer (DTC) furniture brand, "Urban Timber," as they swapped this archaic process for a lean, agentic AI stack. The results were not incremental; they were transformative. This is a real-world account of moving from manual grunt work to autonomous, intelligent execution in programmatic marketing.
The Stagnation Problem: High CPA, Low ROAS
Urban Timber is a familiar story. A successful Shopify-based brand with a turnover of around £15m, they had hit a plateau. Their in-house team, consisting of two skilled programmatic managers and a designer, was wrestling with diminishing returns. Their Cost Per Acquisition (CPA) was stuck at an uncomfortable £85, and the Return on Ad Spend (ROAS) had flatlined at 2.5x.
The team was trapped in a cycle of manual labour. An estimated 60% of their collective time was consumed by the Sisyphean task of campaign setup, budget management, and A/B testing a meagre 5-10 ad variations per month. Creative fatigue was rampant, and the insights gained were minimal. They were data-rich but knowledge-poor, reacting to performance dips rather than proactively engineering success. The campaign workflow was a one-way street: build, run, report, repeat. There was no intelligent, automated feedback loop.
The Agentic Stack: An Orchestrated System for Under £3k
The solution was not another monolithic SaaS platform promising a silver bullet. Instead, the brand constructed a bespoke agentic stack — a collection of specialised, API-first tools orchestrated to function as a single, intelligent system. This is the crucial distinction: it is not just "AI," but an agentic workflow where different components communicate and trigger actions autonomously.
Here is the breakdown of the stack, which cost a remarkably lean £2,550 per month:
### 1. Data Foundation: Segment (£1,000/month)
First, all data sources were unified. The brand used Segment to consolidate customer data from Shopify, their email service provider (Klaviyo), and website behaviour tracked via Google Analytics 4. This created a single, coherent view of the customer, forming the bedrock of the entire system. Without clean, unified data, any AI is simply guessing.
### 2. Predictive Engine: Custom Model on Vertex AI (£500/month)
This is the brain. Instead of relying on the black-box algorithms of ad platforms, the company deployed a custom-trained propensity model on Google Cloud's Vertex AI. This model analysed the unified data from Segment to identify and predict high-value audience segments that standard ad platform targeting could never isolate. For example, it could create dynamic audiences based on "users who have viewed oak dining tables three times in the last week, have a predicted lifetime value over £2,000, and have not opened a marketing email."