Marketing directors are not paid to run experiments. They are paid to deliver growth, profitably. The relentless pressure to stretch budgets and prove ROI has turned many marketing departments into glorified accounting functions, endlessly optimising for marginal gains.
But what if you could change the fundamental equation? What if, instead of incremental improvements, you could achieve a step-change in output and effectiveness without a corresponding explosion in headcount or agency fees? This isn’t a hypothetical. It’s the reality of what one UK retailer achieved by deploying a bespoke, agentic AI stack for its Paid Social marketing.
This case study details how a household-name, but budget-conscious, UK retail brand went from producing around 20 unique ad creatives a month to over 200, translating directly into a 47% increase in Return on Ad Spend (ROAS). All for a software and automation outlay of under £3,000 per month.
Before the intervention, the brand’s Paid Social process was painfully familiar. A lean in-house team of three, supplemented by a respected London media agency, was perpetually bottlenecked. The creative process was a linear, human-powered assembly line.
Concepting meetings led to briefs, which were passed to a designer, who produced a handful of static and video assets. These were then tweaked, approved, and handed to the media agency for trafficking on Meta and TikTok. The entire cycle, from idea to impression, took the better part of two weeks. Testing was limited, fatigue set in quickly, and the sheer operational drag meant that opportunities were consistently missed.
Their agency, like many, excelled at optimising campaigns within the platforms. Yet, their model disincentivised a high-volume creative approach. Their fees were structured around media spend, not creative output, creating a silent antagonist in the quest for performance.
An agentic workflow is not about replacing humans with a single, magical AI. It's about orchestrating a series of specialised AI tools to perform complex, multi-step tasks with minimal human intervention. The goal here was not to fire the marketing team, but to transform them from factory workers into fleet commanders.
The stack was designed to automate the most time-consuming, repetitive parts of the creative workflow. Here’s the breakdown:
1. Intelligence & Briefing: The "Director" Agent
Tools Used: A custom-built agent using OpenAI's GPT-4 API, connected to the brand's live product and sales data via an Airtable database.
Function: This agent acts as the strategic brain. It analyses real-time sales data from their Shopify store, identifies the top 20 performing products, and cross-references them with Google Trends data for the UK. Based on this, it generates 10 distinct creative concepts per product, each tailored to a specific audience persona (e.g., "The Bargain Hunter," "The Style-Conscious Millennial"). The output is a highly-detailed creative brief, written in markdown, and automatically dropped into a shared Slack channel. This replaced the weekly, two-hour concepting meeting entirely.
2. Creative Production: The "Art-Worker" Agents