Using AI to make emails personal - at scale

The Telegraph had seen stratospheric subscription growth over just a few years. But how could we ensure that these hundreds of thousands of new subscribers remained engaged?

The answer lay in a canny combination of machine learning, an LLM and a healthy dose of elbow grease and teamwork.

I kicked things off with our in-house tech team, who used machine learning to select articles that subscribers hadn’t read yet, but would more than likely want to, based on their previous behaviour.

Next, we developed prompts upon prompts upon prompts. We corrected, coached and edited until the LLM could summarise each article in a suitably engaging manner, without giving its content away.

We wrapped the whole thing up in a simple, engaging template, then worked with CRM to deploy our new solution.

We tested rigorously, checking for hallucinations and ensuring that subscribers would respond positively to our AI-driven solution.

And respond they did, with a +40% increase in engagement.