Ippen Digital, Mediafin share how they created customer-facing GenAI products

By Paula Felps

INMA

Nashville, Tennessee, USA

Connect      

With generative AI becoming so prevalent, news publishers are either looking at how to use the technology to improve the customer experience while also integrating it into internal workflows.

During a “show and tell” segment of a recent INMA Webinar, GenAI products: getting them right, INMA members shared what customer-facing GenAI projects they’re working on and how they use AI in their media companies.

Ippen improves summaries and images

Alessandro Alviani, product lead/AI at Ippen Digital in Germany, showed members how the company uses AI in reader-facing applications.

The first product he introduced is an app that allows users to get a bullet-point summary of an article at the click of a button. In these early stages, Ippen Digital is also asking for feedback to understand what needs to be improved. That’s important, he said, “because we really would like to roll it out for additional apps.

Another feature of the app is that readers can select the amount of content that fits their situation at the moment: “Let’s assume you have just a couple of minutes, then you can generate a three-bullet point summary. If you have a couple of more minutes, you can get a longer summary, you can get a version in simplified language, and so on,” Alviani said.

To prevent hallucinations, Ippen is developing a feature to evaluate the content being generated and assess the quality of the bullet points.

“If the bullet points do not meet a specific threshold or specific metrics that we have established — something like completeness or factual correctness — the tool is able basically to reprompt itself and generate a new set of bullet points to help our editors [fight] against hallucination.”

Ippen Digital has taken a responsible, proactive approach to creating AI-generated images.
Ippen Digital has taken a responsible, proactive approach to creating AI-generated images.

The company is also working with AI-generated images and wants to do that responsibly. That means not generating photo-realistic images, using only photo illustrations instead. The prototype it has created allows editors to generate images just by pasting an article into the tool and telling it to generate an image based on the article.

If editors aren’t happy with the picture, they can manually change things such as the colours, add to it, or even change the prompt to change the image.

“Then, at the end, when you’re happy with the final results, you can easily download the picture as a JPEG file,” Alviani explained.

Mediafin’s super search tool

Isaac Van Der Straeten, product owner at Mediafin, showed one of the projects the company developed in 2023 in the wake of ESG (enterprise, social, and government) regulations being implemented.

Combing through the documents related to the lengthy regulations was a nearly impossible task for humans, but Mediafin recognised AI would be up to the challenge.

“We contacted a partner, LegalFly, a company that is used to transform legal documents and digest a lot of legal documents,” he said. LegalFly uses AI to review and draft contracts, providing users with the correct information and to ensure a contract is valid or that it isn’t missing anything.

“We thought maybe they could help us digest all these documents.”

Working with the company LegalFly allowed Mediafin to create a tool for easily searching ESG documents.
Working with the company LegalFly allowed Mediafin to create a tool for easily searching ESG documents.

The partnership worked and resulted in a tool that allows users to search all the documents related to ESG. They can also search the pages of individual companies and their documents related to ESG, and LegalFly wrote prompts so users can search them more easily.

Van Der Straeten acknowledged it’s a niche product that hasn’t received a great deal of use, but the company is happy with this initial venture into AI.

“It’s been up and running for a couple of months and we didn’t have any accidents so far — no hallucinations,” he said, explaining the LLM was trained to use only the data contained in the documents, which prevents it from giving wrong answers.

How do INMA members feel about AI?

To wrap up the Webinar, INMA Generative AI Initiative Lead Sonali Verma conducted an online poll to take the temperature of INMA members.

First, she asked how confident members are in building a GenAI-based consumer-facing product. The results showed that a majority of attendees — almost 60% — were “reasonably confident” and are cautiously experimenting with it, while 25% still felt it was too risky. The remaining 17% said they were “very confident” and had built-in safeguards.

Van Der Straeten commented that working with LegalFly had given Mediafin greater confidence because it was clear the company had guardrails in place to prevent hallucinations and false information.

“I think the lesson in that then is to partner with someone who’s building the tool who is extremely careful with what they do there,” Verma said.

The biggest concern most INMA members have about a customer-facing GenAI product is accuracy.
The biggest concern most INMA members have about a customer-facing GenAI product is accuracy.

For her second question, Verma asked attendees their biggest concerns about building a customer-facing GenAI product. The No. 1 concern, according to 56% of the poll participants, was that it will provide accurate information. Cost is the biggest concern for 28%, while 6% were concerned it wouldn’t be used.

About Paula Felps

By continuing to browse or by clicking “ACCEPT,” you agree to the storing of cookies on your device to enhance your site experience. To learn more about how we use cookies, please see our privacy policy.
x

I ACCEPT