Get practical about AI: Join our free webinar on June 4, 2025. REGISTER NOW
Today, it may seem that AI is discussed everywhere. Probably, this statement is really not far from reality. Artificial intelligence and its transforming power are an extremely popular topic. The Innovantage podcast hosted by Sigli’s CBDO Max Golikov, also makes its contribution to raising public awareness about the potential of this technology. In the new episode, Max decided to concentrate on the real-world impact of AI on the business world.
To talk about this, he invited Maxime Vermeir, Senior Director of AI Strategy at ABBYY, who agreed to share valuable insights with the podcast’s audience.
Maxime is one of those experts who “were studying AI before AI was cool”. Over the years, the industry has changed a lot, and he has seen its evolution from many angles. In his career, he worked with enterprise tech, SaaS solutions, and mobile apps, and even launched a startup before the startup scene became mainstream.
These days, his focus is sharply on artificial intelligence. Now, he helps people understand and harness it in meaningful, productive ways.
AI is a very trendy technology. But what is behind this hype? Maxime acknowledged that hype has always been a part of the tech industry from Web 2.0 to the cloud. AI is no different in that regard.
Artificial intelligence has entered the mainstream, sparking both excitement and concern. On the one hand, more people are becoming curious and educated about AI. This opens up new opportunities for businesses.
On the other hand, this surge in attention often leads to inflated expectations and sometimes misconceptions. For example, there are people who still worry that robots may take over their jobs.
What stands out here is how rapidly the AI hype cycle is compressing. Unlike the slow and chaotic early days of cloud adoption, the AI space is maturing much faster. Regulations are coming in quickly. Companies are looking for transparency and ask AI vendors to provide them with detailed information about security protocols and the way the technology is built.
For Maxime, that’s a sign of real progress. The industry is beginning to get rid of the “one-size-fits-all” approach and is moving toward a more thoughtful, pragmatic perception. The hype may still be there. But also, there is a growing sense of responsibility.
He also emphasized the importance of practical value over hype. For instance, in some cases, it can be simply not feasible at all to build your own large language model, which can be a very expensive and time-consuming project. Instead, it can make more sense to rely on much simpler, classic technologies like regular expressions.
There is real innovation happening in AI, and powerful tools are emerging. But not every problem needs a big solution. It takes time for people to understand that the “next big thing” isn’t always the right choice. It is much more important to match the right tool to the right use case, with a clear understanding of both their capabilities and limitations.
As the hype around LLMs exploded, many rushed to apply them to document data extraction and even made that one of their first use cases. Maxime found this to be not the best idea, given the technology’s broader creative potential. This led to a widespread belief that LLMs could fully replace established solutions like Intelligent Document Processing (IDP), a mature technology used by enterprises for over 20 years.
At one point, an analytics firm even declared that “IDP is dead” and compared it with RPA (Robotic Process Automation), which was overshadowed by newer tools.
But today, the narrative has changed.
As Maxime noted, businesses have realized that throwing PDFs into LLMs often leads to a lot of issues, like hallucinations, context limitations, and inaccuracies. This is especially problematic when extracting critical data. The consequences of such mistakes can be quite tangible and costly.
With decades of experience in document processing, ABBYY has always taken a different approach. Long before the LLM boom, they were already leveraging transformer-based models tailored to particular tasks, such as document segmentation or value extraction. The key is combining technologies within a purpose-built platform, designed specifically to deliver reliable results in real-world business environments. This approach can bring true value to businesses.
Maxime highlighted that there are several kinds of AI that can serve different goals. For example, generative AI shouldn’t be the first choice for tasks like data extraction. Instead, it would make sense to rely on more traditional, proven methods like machine learning, convolutional networks, NLP, and transformer-based models. All of them have been around long before ChatGPT entered the game.
For example, ABBYY also integrates large language models into its platform, but with a clear purpose. They are used to deal with more complex, probabilistic use cases where generative AI can truly add value. When used appropriately, it creates a so-called “1+1 = 3” effect, where multiple technologies are used together to deliver better outcomes than any single one could do when applied alone.
Solving real-world problems always requires a combination of AI methods, each chosen in the right context.
As Maxime mentioned, while GenAI might not be a go-to option for document processing, it has great creative power. It makes such tools highly helpful when it is necessary to overcome the blank page problem in writing and content creation.
Maxime explained that it is possible to transform generative AI into a powerful creative partner for your personal and business workflows. While AI doesn’t always perfectly understand users’ intents, its ability to offer thoughtful, predictive responses can help to refine and evolve your own ideas.
He also sees generative AI as a good social equalizer in creativity. In the past, expressing creative ideas often required technical skills or specific tools. Such things were serious barriers for some people. But now, with intuitive AI interfaces, users can simply explain what they want and get results.
This shift is unlocking creative potential for a much bigger group. More people can now bring their ideas to life, regardless of background or skills.
Today, many executives approach AI tools with very high expectations. After implementing a solution, they want to see a quick, transformative impact. But the reality is more complex.
As AI becomes more mainstream and visible in everyday media, there is a perception that it can do everything. This leads some decision-makers to underestimate the effort and resources required to implement it effectively. While generative AI demos are undeniably impressive, turning that potential into practical value is much more challenging.
Maxime pointed out that while creative applications of AI are usually quite simple to understand and leverage, enterprise use cases are a different story. Business processes involve collaboration, integration across systems, and clear ROI expectations.
The true impact lies in automating end-to-end processes, not just speeding up separate tasks. Executives need to understand their processes first, identify the real bottlenecks, and only then choose the right technology to improve them.
The wrong way to implement AI is to focus only on speeding up individual steps or reducing human input without understanding the full workflow. While those changes may show short-term gains, they often create new issues in other stages.
Success in AI adoption should be measured by improvements in the overall process. That’s where process intelligence plays a key role. It allows you to visualize, test, and predict the impact of changes across the entire workflow. Without that insight, automation alone won’t deliver meaningful results.
Time often becomes a key metric in measuring AI success. However, it shouldn’t be the only one.
For example, automating an insurance claim process too quickly with the wrong tech, like an LLM, may initially seem efficient but could lead to costly errors and rework. A better measure of success includes fewer mistakes, faster and more accurate outcomes, and compliance with regulations like the EU AI Act.
Maxime strongly believes that the real challenge with AI isn’t the technology itself, but how we implement and use it. He views the EU AI Act as a necessary step in regulating this technology, although it could move faster. The Act doesn’t hinder innovation. It ensures transparency and accountability, which are essential for building trust in AI systems.
This regulatory framework helps to clarify AI’s role and prevents its misuse. As a result, businesses can stay compliant and avoid a lot of risks.
Organizations can either self-assess or work with third parties to ensure compliance with transparency requirements to promote a more responsible use of AI in the business space.
The reality is that AI is being promoted in exaggerated ways, similar to how RPA is now being rebranded as "agent technology" to attract more market attention.
Nevertheless, when selecting AI tools, enterprise executives should focus on real outcomes such as cost savings or process improvements backed by customer testimonials.
Today, there are a lot of vendors that name well-known technologies like GPT or cloud services behind their solutions. At the same time, they don’t provide insight into any unique features and approaches that determine the value and differentiation of their tools.
For businesses, it is vital to balance old and new technologies. While a well-established technology introduced over 30 years ago offers reliability and a proven track record, the downside is that it may lack flexibility or fail to address newer, evolving needs.
It’s essential to combine different technologies strategically and create solutions that deliver real value. AI should be implemented thoughtfully, not because it is one of the latest trends or buzzwords, but because it can change processes for the better.
According to Maxime, AI agents are currently both a hype and a useful technology. While Salesforce has contributed to their promotion, there is potential for AI agents to address issues that RPA couldn’t. For example, they can automate complex processes and enable broader access to technology.
AI agents are designed for practical automation, and they can orchestrate existing tools without human intervention. The reason for optimism is that more people now understand the importance of process intelligence. This awareness is increasing, making organizations more cautious and focused on properly implementing AI to avoid failure.
It’s interesting to mention that 80% of AI projects today still fail. But of course, people do not want to be this 80%, they want to be the lucky 20%.
How to reduce the chance of failure? Maxime mentioned that AI can only be effective if you provide it with the knowledge of your organization. Here, it is necessary to focus on the data about how your business operates and the information locked in your documents.
Additionally, you need to clearly explain how your processes work. That is where process intelligence can help. By providing a blueprint of your current workflows and what changes will lead to the desired ROI, you can guide AI to deliver results.
Maxime stated that AI could be the next big step in enterprise automation. People are tired of dealing with countless APIs and bypass services. They want to have integrated platforms. It is still a bit early for agent frameworks to be fully enterprise-ready, but this can happen quite soon.
Every major tech shift brings both job creation and job displacement. It absolutely doesn’t mean that designers or artists are no longer needed just because AI can generate visuals. There will always be two groups: those who resist change, fearing it will replace them, and those who embrace it, using it to amplify their skills. AI is just another tool. It doesn’t replace expertise, it enhances it. Just like a paintbrush doesn’t make someone a painter, AI doesn’t make someone an artist without the underlying skill.
New roles, like prompt engineers, are already emerging. At the same time, some repetitive tasks are being phased out. It’s the natural cycle of innovation. Those who adapt, learn, and evolve will continue to find new opportunities.
According to Maxime, this technology shift is no different from those before it. When RPA was introduced, people were afraid that automation would replace their manual tasks. And it did. But it also opened the door to more valuable work.
AI can perform some very tangible, very clear tasks. For example, if somebody is bad at writing emails, AI can help this person to do it at an average level.
But for experts, the value lies in how well they guide the tool. The quality of the input directly determines the quality of the output. “Garbage in, garbage out” still works. If you provide vague prompts, you will get generic results. But if you take the time to explain structure, context, and intent, the output can match the quality you aim for.
At the end of their conversation, Max Golikov asked his podcast guest to share practical recommendations for AI leaders and founders.
Maxime said that they must start by clearly defining the problem they are trying to solve. One of the most common mistakes is adopting technology simply for the sake of innovation, especially during intense hype cycles, which are becoming more frequent and extreme.
Whether building a new product or implementing AI within a business, the core principle remains the same: clarity of purpose is critical. For enterprise leaders in particular, a deep understanding of their existing processes is non-negotiable. Without that insight, applying AI effectively is nearly impossible.
With clear goals and process visibility, leaders can choose the right tools with precision. This will help them avoid the common pitfalls, like using overly complex solutions for simple problems.
Identifying the proper use of emerging technology is one of the things you can learn from the episodes of the Innovantage podcast. Don’t miss the next episode to stay up to date with all the latest trends in the business world.
.svg)
.svg)
.svg)
.svg)
.svg)
.svg)
.svg)
.svg)
.svg)
.svg)
.png)
.png)
