University of Hertfordshire and DemandScience UK Limited KTP 24_25 R1

A marketing software company is teaching a large language model to handle sensitive business data without leaking it to the public internet. Business-to-business software companies sell complex products to other companies, not to individual consumers. Their marketing teams need to generate personalised emails, analyse sales conversations, and track which messages actually close deals. Off-the-shelf AI tools like ChatGPT cannot safely process this data because the information is proprietary—customer lists, pricing strategies, contract terms—and sending it to a public cloud server risks exposing trade secrets. Current marketing automation tools are not smart enough to handle these nuanced tasks. The project builds a secure wrapper around a large language model that keeps all data processing within the company’s own infrastructure. If it works, a sales team could ask the system to draft a follow-up email referencing a specific technical problem a prospect mentioned in a call, without that call transcript ever leaving the company’s servers. The expected benefits are more effective marketing campaigns, higher conversion rates, automated reporting, and reduced labour costs. For the companies that buy these tools, the change would be invisible but real: faster sales cycles and less time spent on repetitive writing tasks.