PrivAIcy matters: why we built our own AI*

When genAI made its grand entrance, we had to rethink our approach to technology. We want to empower our people to experiment with the fascinating abilities of genAI, but without breaching our commitments to data privacy and confidentiality.

Heather Baker, our CEO, said, “Many organisations share the same AI challenges. Unfortunately, some are learning the hard way that data entered into genAI can be used to train the model. And then it can be churned out elsewhere, at any time. Others are aware of the data privacy issues and are dealing with them by creating complicated rules for employees to follow….and employers to police.

“We wanted to do things differently. We’ve given everyone the freedom to use genAI, but without the risk of compromising confidential data.”

Tailor-made AI

Definition’s AI Director, Luke Budka, worked with our development team to build Definition AI, a bespoke genAI ecosystem that only our employees can access.

Importantly, all the data we enter into Definition AI is private. It’s owned by us and it’s stored on our own servers, so it doesn’t get fed back into the large language models.

We’ve also taken the opportunity to combine all the best genAI in our model, with both ChatGPT (GPT-3.5 Turbo and GPT-4) and Dall-E built-in. And we’re not limited to OpenAI – we’ve got the option to add new tools as we go.

As Luke Budka said, “Definition AI has been designed to evolve. Its architecture is built so that we can incorporate secure access to any genAI tool with an API. We’re working on a few already that are in high demand in our team.”

We launched Definition AI in September 2023 and it’s already popular with our people. Here are some of the ways we’re using it:

  • Invite it to a brainstorm – want the perspective of a certain demographic? Then write their persona and get them to help you during a brainstorm.
  • Sum up a transcript – download the transcription from a call, copy and paste it into the AI and ask for a summary. Very useful if you, or people in your team, couldn’t join a meeting.
  • Check your grammar – get a second pair of eyes on your emails, newsletters, media releases, case studies…

The results (so far)

  • 40% of our people are already using Definition AI regularly.
  • During the last two weeks of October, we saw 10x increase in usage across the business.

We’ve delivered genAI training to our whole team. And we’re already talking to businesses about how we can build a bespoke private environment for them.

Let’s talk privAIcy

*Our private AI came up with the PrivAIcy pun