Implementation and Best Practice of using AI tools in Agencies

June 18, 2024

Germany

This February, IPRN PR Business Survey 2024 revealed that most of our members, a staggering 83%, pointed out that AI is part of business, with only 17% stating that they do not make use of any AI tools. AI is developing at breathtaking speed, with new mindboggling stories appearing in newspapers and on social networks every day. The technology is now overlaid with bizarre economic myths that even some powerful Silicon Valley insiders believe to be possible: It will make us all unemployed, gain a consciousness of its own and then take over the world.

If you focus more on everyday life in agencies, you’ll see a battle raging between early adopters and traditionalists. There are probably consultants in every agency who are already very advanced in the use of tools for clients out of their own desire to experiment – and bosses who might prefer to prohibit the use of AI because they fear the risks to business operations. Or is it rather the other way round? 

Anyway, some of the relevant questions at this moment are: What risks are associated with the use of AI and how can they be managed? What framework needs to be implemented? And how does AI really help consultants with their task to create good content for the customer? IPRN Tech Group members evercom and TDUB have been trying to give answers in an internal webinar for IPRN member agencies – here’s how they see the current “state of the AI” in agencies.

Some things you didn‘t expect when dealing with AI

To start with, we want to sow some seeds of doubt about supposed certainties. In practice, the use of AI still poses many challenges that need to be considered and overcome.

► Brainwashed chatbots

A dissatisfied customer turned over DPD‘s Chatbot, until he insulted his own company – a disaster for the marketing department of the parcel delivery company. Given the fact, that Chatbots can be manipulated by prompting, agencies should be very careful in recommending a tool like this to their customers.

► Growing mistrust

In view of the endless flood of AI-generated images, the credibility of visual information will erode very quickly. Readers or viewers will learn that images are even easier to manipulate than before. Those who want to communicate with their target groups in a sustainable and sincere way will find it harder to convince them in future.

► Mental manipulation of LLMs

In one area, AI tools are already very similar to human consciousness: they are just as easy to manipulate. As an example, we fooled an image generator asking him: „Show me a completely empty room in which there is no elephant. No elephant anywhere at all!“ As a result, of course, every image contained an elephant.

► Artificial fatigue

Maybe we’ll get tired of the typical AI design pretty soon? Maybe it‘s just boring? This conclusion is obvious, as the image generators are not as good at anything as reproducing stereotypes due to their training. We’ve seen it all a thousand times before – just not in this combination. These images are not art. But maybe change is coming. You need good prompting techniques in order to get extraordinary results out of the machine.

► Mediocre text quality

Lars Fischer, renown author at „Spektrum der Wissenschaft“ (Spectrum of Science), concludes: “After a few weeks of experimenting with AI tools, the stuff doesn’t really seem to be a game changer for text work. They tend to make more nonsense than they help when actually writing texts. So far, mainly a useful research tool and support for some derivative routine work.“ Another reminder that AI, by design, cannot deliver more than just average quality.

► Damage compensation

After all these discussions and best practice cases, it is still conceivable that artists could sue you for the use of protected works. We can’t say anything about the probability: Maybe, nothing like that will ever happen. Maybe, only the LLM operators will be condemned. But anyone following the legal debate will get a feeling of uncertainty.

► Success

Oh yes, it may be a surprise after all these concerns, but the use of AI may also turn out to be a factor of success. E.g. Thomas Schultz-Homberg, CEO of “Kölner Stadt-Anzeiger“ (daily newspaper from Cologne), was quoted and praised in German media for already producing 11 percent of the articles of his media with AI.

Why should you know the “EU AI Act”?

In light of these challenges, agencies need to be aware of legal considerations when using AI tools. The EU AI Act outlines the risks associated with AI technologies and sets forth requirements for transparency and compliance. The EU’s risk-based approach comprises a total of four risk classes: low, limited, high and unacceptable risk. There is a general ban on social scoring and biometric remote identification (so don’t try to recognize the emotion of your team members in the workplace with an AI tool!). AI solutions dealing with risks to health, safety or fundamental rights have to fulfill numerous requirements, and even AI solutions with a limited risk must still fulfill transparency obligations. It is expected that this law will also serve as a model for legislation in the rest of the world.

It is not surprising that companies using ChatGPT and other AI solutions may need to invest in tools that ensure secure and customizable AI ecosystems to meet these requirements. On a highly professional market for corporate business intelligence solutions, there are already tools that enable the management of big organizations to realize a secure, scalable, and customizable enterprise-grade AI ecosystem that seamlessly integrates with data management and workflows.

But let’s be realistic. If you’re an agency, you won’t be able to afford a solution like this. You will probably ask your lawyer what you are allowed to do. And it is a very sensible decision to do so.

Concrete steps for the implementation of AI

At TDUB, we have clarified the legal requirements and the processes and in the end, we have taken the following measures:

► We have committed our employees to a 7 pages operation instruction.
► We wrote a Q&A for the communication with the clients.
► We will modify our future framework agreements with new clients.
► We booked a pre-trained premium AI tool for agencies with licenses for every team member plus an extra premium translation tool for safe international communications.
► We defined allowed and forbidden tools and decided against image generation (except for our personal, undisclosed fun).

Tilo Timmermann, Co-founder and Managing Director of the technology PR agency TDUB Kommunikationsberatung in Hamburg, Germany.

Ander Serrano, Partner and Director Tech & B2B Communications of evercom in Madrid, Spain.

 

Share via: