Creating Carin: Our ‘Client Voice’ AI Bot for MoneySuperMarket

Using AI to improve the regulatory adherence and brand relevance of press releases and client content, speeding up sign-off time and building stronger client relationships.

We developed Carin, an AI bot trained on MoneySuperMarket's brand guidelines and regulatory requirements. By analysing our draft content before client submission, Carin identifies potential tone, phrasing, and compliance issues, reducing review cycles and enabling us to get to the outreach phase of our PR projects faster, while ensuring brand consistency and regulatory adherence.

What we did

AI tool creation

Digital PR

The challenge

As an agency partnering with clients in a range of industries, we always strive to produce content that fits with existing tone of voice and brand guideline documentation, first time. For our clients who operate in highly regulated industries, there’s an additional need to ensure that our content is fully compliant and current. But, we’re humans and, much as we try, we don’t always get it right in the first instance. So, we wanted to find a way to reduce the delays in these instances and to delight our client first time, every time.

Our approach

We built Carin. Carin is an AI bot, trained up on MoneySuperMarket’s brand guidelines and tone of voice, but also all relevant guidance and regulations around the products we’re working with them on. Now, when we’ve got a first draft of a press release or some on-site campaign content, we run it through Carin. She flags any specifics where we can do better in terms of tone or phrasing, as well as any potential compliance issues. We’re then able to rectify the issues before sharing with the client; avoiding potential rounds of back-and-forth, and enabling us to get going with our PR outreach more quickly.

"The more we use Carin the better it becomes - we retrain it constantly with the new approved releases and on-site campaign content so that it learns and improves along with us."

Stephanie Finch, PR Director at Propellernet