‘Stop chasing AI and start chasing profit instead,’ says AI expert

Zwingmann said what publishers should really be asking themselves when considering an AI project are questions like these:

• What's at stake?
• What's the impact?
• What are the current costs of what you are doing that you want AI to help with / replace?

For example: "Why should you spend 80K on an AI project if there's a human who can do this for 50K a year? It's not a good business case." Zwingmann said. "But that's what I see so often happening in real life - that you have this immense gap between what AI can actually do realistically, and what your business needs practically."

"This gap is what leads to so many failed AI projects that will just end up in a waste of time and waste of budget," he told participants at WAN-IFRA's Frankfurt AI Forum in April.

Zwingmann, the author of books such as "AI-Powered Business Intelligence," (2022) and "Augmented Analytics" (2024), also publishes a free weekly newsletter called The Augmented Advantage.

"Don't start with what AI can do," he advised. "Start with profit opportunities in your business."

And while it can be difficult to be precise about the level of profit or impact that a publisher wants to achieve, Zwingmann offered another consideration.

"It's much easier to ask the question, 'What's the minimal number you need in order to get moving?' " he said.

Having a common number in mind helps keep everyone on the project aligned: "If everyone has the same hurdle, you will have much more actionable projects," he said.

For example, Zwingmann said, he recommends people consider an actual hurdle, like the ones used in sports for track and field events.

"It makes a huge difference whether you want to break the Olympic record, or whether you just want to get over that without breaking your leg," he said.

It's important to have this same kind of alignment in AI projects, he added, because if someone on the team wants to break the Olympic record, and someone else involved in the project is just happy if they don't break their leg, then the project will ultimately fail.

"You need to have this alignment at the beginning of saying, 'What's the minimal hurdle to jump over?,' Zwingmann said.

That's also the first step in a five step process, he recommended for creating profitable AI projects, "Define a threshold"

Zwingmann recommends starting off with a minimum reward that the launcher wants to see from a new project. He calls this a "10k threshold" because "10k is a number that is roughly equivalent to what you pay for one professional or specialised full-time employee per month," he said, which translates roughly to approximately 50 or 60 euros per hour.

"This is a baseline amount, and you can justify that," he noted, and added that using such a cost-basis is much easier than trying to determine the exact outcome you're looking for.

Analyse your business problems

The second step in Zwingmann's recommended process is to look at the problems to be addressed within the scope of the given project. And not only current problems but also those likely to arise. He calls current problems "pain points," and these are issues that are currently annoying or costing the business money right now. "Bottlenecks" are future problems or issues, such as if you have a number of employees in a given area who will all be retiring within a few years.

Publishers should then lookat every pain point and bottleneck as part of the process they define in order to help set their priorities for the project.

WAN-IFRA Members can access all of Tobias Zwingmann's presentation slides, as well as those from other speakers at the Frankfurt AI Forum, on our Knowledge Hub.

Apply your threshold

The third step in Zwingmann's recommended process is to "Apply the threshold." This step is very straightforward, he noted, because by this point you have already identified the problems you want to address.

"You know what minimal impact you're looking for, and now you just need to apply that," he said.

For example, you can think about things like, will this help to unlock additional business opportunities worth 10K plus for me? Likewise, will removing this bottleneck prepare us for this upcoming regulation and minimise the risk associated with it by 10K plus, he said.

There are lots of different questions you can ask, Zwingmann noted, but essentially the team should look at each pain point and bottleneck inside the process that they want to define.

Map AI capabilities

Once the project team has identified what their business actually needs, they can start considering what AI can do, he said.

"Not every pain point that you prioritise will actually be a good fit for AI," Zwingmann said.

There are multiple ways to find that out, but one way Zwingmann likes is "AI Design Sprint," a methodology invented by a Danish company called 33A.

A key part of the AI Design Sprint framework is a set of cards that offer descriptions of AI capabilities in what Zwingmann calls "business friendly language."

These cards provide help in determining what AI can do at various levels, including details of use cases and examples of how that is applied in real life, as well as some of the technologies that are involved, he said.

Implementing the AI roadmap

Coming to the fifth step, Zwingmann said "If you follow this process, what you typically will end up is a list of use cases or business case fact sheets."

Typically, he said, teams will have 10 to 15 high impact opportunities they can now get to work on.

"These opportunities will have clear financial targets because you started with these targets upfront," he said. "They have well-defined outcomes because you know what to expect from the AI. And they are optimised for ROI."

He then recommended that the team summarises each use cases on a single page, such as a PowerPoint document, and make some straightforward statements about things such as:

• What are you trying to achieve / what problem are you trying to solve?
• What are the value anddata dimensions?
• What kind of data do you need for that?
• What's the solution dimension?
• Which AI cards do you use for that?
• Who could be stakeholders?

After answering these and similar questions, Zwingmann said you will be able to take the original 10 or 15 opportunities and rank them by their potential impact for your organisation and have a clear plan for your project.

Brian Veseling

brian.veseling@wan-ifra.org