We’ve become the first UK agency to create AI-powered client assistants – digital ‘twins’ that work with account teams on daily tasks to improve client service.

The assistants are powered by the latest version of OpenAI’s GPT-4 model and help our account teams get immediate answers on almost any aspect of a client’s business. Which will save our clients’ subject matter experts loads of valuable briefing time.

We can upload up to ten thousand files of client-specific information, and then interrogate them using AI to help with strategy and delivery. They’re only accessible to each client’s account team, and only built if the client gives permission.

 

Here’s a quick demo:

 

 

One of our clients TLC LIVE, a company that provides tutoring for schools, was involved in the testing phase.

TLC LIVE COO Josh Blackburn said:

“We were really happy to participate. Our AI assistant is essentially a new member of the Definition team, without any extra cost to us. The combination of AI and our TLC LIVE data is incredibly powerful and we look forward to it making our work together even quicker and slicker.”

Our AI Director Luke said:

“We’re always looking to improve client service with the latest technology. There’s a lot we want to do with AI but we’re taking a measured approach to ensure security and compliance.”

And Louise, our Chief Client Officer and Co-founder said:

“Harnessing the potential of AI – both in client service, but also in products that help brand-side marketing teams – is a key part of our strategic plan. Our client assistants are part of this drive. Imagine if you could hire an agency that has instant, in-depth knowledge of your business, combined with almost all knowledge ever digitised by human beings. These assistants are an incredibly powerful way to do make sure we deliver brand alignment in our work from day one.”

The assistants, built by our AI team, are powered by OpenAI’s Assistant technology and use a combination of GPT-4 Turbo and GPT-4o alongside OpenAI’s File Search and Code Interpreter (for data analysis) tools.

All answers they give include source links back to the files they’re from. This reduces hallucinations and makes sure our client teams can quickly double-check things.

Luke also said:

“OpenAI launched a new version of its file augmentation tool in April, which increased the number of files we could attach to its models from 50 to 10,000. It uses vector database technology that’s capable of both keyword and semantic search – this is a game changer from a usability perspective. Hats off to our dev team for spinning this up within our private AI environment in six weeks.”

Drop us a line for a chat about AI training, prompting or building.