Named Case Study
EVA (Enhanced Virtual Assistant), productised as FORGE. A multi-agent AI system built for a UK premium kitchenware retailer managing 6,000 SKUs and 6 to 7,000 new product additions every year.
94%
Time reduction per listing
£68K+
Annual savings
100%
Voice consistency
In Their Words
At Harts of Stur, we manage a significant volume of new SKUs entering our system each year (6 to 7,000), and we were looking for ways to make this process more efficient and cost-effective. We also wanted to enhance and streamline our ecommerce listings to improve both consistency and the overall quality of our product content.
After meeting Chris and Tim, the Ignite AI Solutions team at a local networking event, we reached out to explore how their expertise could support our goals.
Together, we developed unique tailored tones of voice for each of our product categories and established a structured specification rule set to guide the AI model's writing style. This had to work with an extremely varied set of inputs, from a single product image all the way through to multi-page technical specification documents.
With Ignite AI Solutions, we built a custom platform which we nicknamed 'E.V.A', which allows us to enter classic product information from multiple data sources such as PDF, CSV, URL, image, text and SKU/Barcode/EAN to generate fast, accurate, and consistent listings.
The entire process was smooth, collaborative, and genuinely transformative for our content workflow. Ignite AI helped us achieve clarity, consistency, and personality across our product listings, and the results have had a significant impact on how we present our products online. Our productivity for the marketing team has been extremely significant, I would say in the high 90% compared to the old way we drafted product listings.
We look forward to finding out other ways that Ignite AI Solutions can help us throughout our business in the future.
Clare Smith
Digital Marketing Manager, Harts of Stur
Station Road, Sturminster Newton, Dorset DT10 1BD
Manual product listing at 45 to 60 minutes per item across a 6,000+ SKU catalogue, with 6 to 7,000 new SKUs entering the system every year. The bottleneck was not the data: supplier PDFs, images, and tech specs were available. The bottleneck was turning that raw material into on-brand, accurate, search-optimised product copy.
Adding new product lines meant either adding editorial headcount or letting catalogue freshness slip. Both options were unacceptable.
EVA (Enhanced Virtual Assistant) is a multi-agent AI system that ingests supplier PDFs, product URLs, product imagery, CSV data, and SKU/Barcode/EAN inputs, then produces on-brand product descriptions at the Harts of Stur editorial standard.
Architecture built in Python, orchestrating multiple AI agents: content extraction, voice matching, factual verification, and quality gate. Editorial team signs off per launch run. Voice profile trained against a curated Harts of Stur style corpus, not a generic e-commerce style guide.
EVA is now productised as FORGE, IgniteAI's pattern for scalable content automation. The system continues to operate at Harts of Stur and is available to other clients with structured-input, structured-output content needs at scale.
The FORGE pattern applies to any business with structured input, an editorial standard, and a high-volume content backlog. Product descriptions. Marketing copy at SKU level. Regulatory filings. Internal knowledge base articles. Technical documentation.
If raw inputs are not being turned into finished content fast enough, FORGE is built for that.
Multi-agent AI splits the task across specialist agents that collaborate. EVA at Harts of Stur uses agents for content extraction (parsing supplier PDFs, URLs, images), voice matching (writing in the editorial standard), factual verification (cross-checking product specs), and quality gates. The editorial team signs off per launch run.
Initial deployment ran across several weeks. Voice profile training took the longest single step, curated against Harts of Stur's existing best content rather than a generic e-commerce style guide.
Yes. The FORGE pattern applies to any structured-input, high-volume content task: marketing copy at SKU level, regulatory filings, internal knowledge articles, technical documentation. The agent architecture stays the same; the voice profile and source-material types change.
ChatGPT generates one piece of content per prompt without persistent voice training, factual verification, or human-in-the-loop quality gates. FORGE's multi-agent architecture handles those at scale, with your editorial standard trained in. The output ships ready for editorial sign-off, not raw drafts that need rewriting.
Yes. Scoping starts with a SPARK Assessment to confirm fit. The FORGE engagement is then scoped against your specific content types, source materials, and editorial standard. No fixed entry price; pricing reflects the actual scale of your content backlog and the editorial complexity required.
Every IgniteAI engagement starts with a SPARK Assessment. From £5,000, 4 to 6 weeks. You walk away with a Findings Report, a scored evidence summary, and a 90-minute Findings Session with Chris.