How to experiment with the use of AI tools in your agency or communication team workflow.

The launch of artificial intelligence (AI) tools based around the GPT-3 dataset has polarised opinion in professional services.

The opportunity of AI is that it will enable us to work more effectively and efficiently. The threat are copyright and ethical issues that arise from the use of AI tools.

The accessibility of tools means that anyone with a web browser can try them out for themselves using command line prompts in a web interface.

Bain & Company has signed a services agreement with OpenAI. Its first customer is Coca-Cola which aims to use generative technology for marketing and consumer experiences. Meanwhile KPMG and JP Morgan have blocked access to AI tools.

A quarter of public relations leaders say they will never use AI tools according to a Confidence Tracker published by ICCO and the PRCA. 15% say they are using it frequently or fairly frequently.

The explosion of AI tools means that it’s daunting to know whether there is practical benefit to your organisation. The amount of information in the market is overwhelming and vendors typically oversell solutions.

A common approach to innovation in a mature organisation is breaking down existing workflow into discrete tasks to see how it might be disrupted by new technology.

Anthony Mayfield demonstrated this approach and used A toI generate a shopping list in his weekly newsletter, breaking the task down into a series of steps. He’s the CEO of marketing transformation agency Brilliant Noise.

“The point here is not over-engineering a simple task, but just noticing all the time where you might use AI to speed up or eliminate tasks that we might categorise as cognitive drudgery.”

“There are a lot of these tasks in the world because of our current incarnation of capitalism’s penchant for shifting labour to citizens and customers under the guise of automation.”

Tools based on the GPT-3 dataset may be getting all the attention but that’s only a fraction of the market. The FutureTools database created by Matt Wolfe in January has 1,050 entries categorised by function.

We looked at how AI tools could be used to help write a press release based on the recent CIPR report about the impact of AI tools on public relations practice researched and written by Andrew Bruce Smith and myself. Experimenting with an existing project is a good way of investigating the potential of the technology and how easy it could be integrated into your workflow.

Task 1 – Extracting the key points from the report

WordTune creates summaries of text based on a PDF, URL, or text. It is effective at pulling out the key messages and topics from a document. The output can be exported as a Word document or via Clipboard.

Task 2 – Developing the key points into a press release

The Wordtune text summary a good place to start to create a press release. You can also provide this dataset to OpenAI Playground and ask it to generate a press release. 

Task 3 – Generating a series of headlines for a press release

OpenAI Playground will also generate a series of headlines for your press release.

Task 4 – Generating a pitch

We also used OpenAI to generate a pitch for the press release for pitching journalists.

Task 5 – Creating a Q&A or talking points for interviews with journalists

Use OpenAI or Chatbot tool to generate a Q&A or series of talking points for a spokesperson for an interview. Andrew Bruce Smith used a no code tool to build one using the report as a dataset.

The experiment shows the potential of AI to support and replace human work in public relations practice. It is a relatively straightforward task based on manipulating text. The outputs are crude and need human refinement, but the potential is clear.

Using tools in real life examples also highlights issues. Wordtune works well to synthesise documents however we’ve found countless issues with OpenAI. There’s a limit to the amount of text that it can process and it works on a predictive basis and so will make stuff up.

I don’t think anyone in professional services can afford to ignore AI. The lesson from this experiment was that tools can be used to create excellent summaries of documents and potentially Q&As, but that generative applications aren’t ready for mainstream without human intervention.  

My suggestion would be to identify common tasks within your agency or communication team. These could be survey design, analysis of datasets, content production, or pitching. Next, build an innovation team or run team wide hacking sessions to experiment with how new AI tools could be applied to these tasks.

Let us know how we can help.