Discussions around AI are awash with hype, complex jargon, and sometimes unrealistic expectations. While technologies like generative AI and concepts like ‘agentic AI’ capture headlines, achieving tangible results requires a clear head and a practical approach.
Luckily, Andrei Papancea, CEO and Co-founder, NLX, recently joined Kane Simms on the VUX World podcast and shared some valuable strategies for cutting through the noise and making AI work for your specific challenges.
Get your house in order first
Often, the biggest barrier to successful AI implementation isn’t the AI technology itself, but the underlying infrastructure. Before you try to build a sophisticated AI application, ensure the foundational elements are in place. Does the AI system have reliable access to the necessary data? Are the enterprise systems it needs to interact with (e.g., databases, APIs for booking appointments or processing transactions) available and functional?
For example, if AI needs to perform an action like replacing a lost credit card, but there’s no robust, accessible API for initiating that process, then no amount of AI sophistication will solve the core problem. Addressing these foundational system requirements is often a prerequisite for AI success.
It’s also important to ensure the people at your organisation are invested in AI! AI projects will struggle to get off the ground without their support because AI will need to connect with multiple business departments to ensure it is genuinely useful for customers.
Start with your problem, not the technology
It’s tempting to chase the latest AI trend, but a more effective path begins with identifying the specific problem you need to solve or the goal you want to achieve. Are you trying to reduce operational costs, improve customer response times, increase sales conversion, or enhance user experience?
Define that objective clearly. Then, adopt a ‘working backwards’ approach, similar to philosophies used by companies like Amazon and Apple. Once the goal is clear, break it down into smaller components. Analyse what’s needed to achieve each part. Only then should you explore potential solutions.
Crucially, recognise that the best solution might involve AI to varying degrees of complexity, or perhaps no AI at all if it doesn’t achieve your goals. This grounds your efforts in real needs, preventing you from pursuing exciting tech (like the often vaguely defined ‘agentic AI’) just because it exists.
Define buzzwords before you chase them
Terms like ‘agentic AI’ are frequently used but poorly defined. Before investing time and resources based on such a term, define precisely what agentic AI means in your context and how it relates to the problem you identified.
Many applications currently labelled ‘agentic’ are, upon closer inspection, sophisticated systems that follow predefined rules or playbooks, using AI components to handle specific steps more flexibly. This isn’t necessarily bad – such systems can be very valuable. The key is to understand the mechanics. Are you building a system that follows a defined process using AI tools, or are you expecting a truly autonomous agent that sets its own goals? Being clear about this helps set realistic expectations and choose the right architecture. Considering ‘agentic functions’ within a larger workflow might be a more useful framing, because you can consider which parts must follow the script, and which parts can suitably be handled by an AI agent that will try to improvise a solution for the customer by itself.
Insist on ROI to filter out the hype
With new AI tools and platforms emerging constantly, how do you separate genuine opportunities from fleeting trends? One powerful filter is a relentless focus on potential Return on Investment (ROI). Before embarking on any AI project, even an experimental pilot, ask: what is the tangible business value we expect to gain?
If you can’t articulate a clear, measurable potential outcome – whether it’s cost savings, efficiency gains, revenue increase, or other concrete benefits – it’s a red flag. The initiative might be driven more by market buzz than by a genuine business need. Many impressive AI prototypes fail to transition into production precisely because this link to value is missing. Making potential ROI a prerequisite helps ensure resources are directed towards efforts most likely to deliver meaningful results.
Understand what AI can (and can’t) reliably do
Different AI technologies have different strengths and weaknesses. Large Language Models (LLMs), for instance, show great promise in understanding the nuances of human language, leading to better intent recognition in customer service scenarios compared to older systems.
However, generative AI (the part that creates text, images, etc.) comes with inherent risks, especially in customer-facing situations. Because these models work on statistics, they are fundamentally probabilistic and occasionally produce incorrect or nonsensical outputs (“hallucinations”). Relying on guardrails isn’t a foolproof solution, as they are often statistically based themselves.
Therefore, it is crucial to assess the tolerance for imperfection in your specific use case. For internal tasks like summarising meeting notes or assisting employees with finding information, a small error rate might be acceptable. But for external interactions, especially in sensitive areas like finance or healthcare, the potential brand damage and legal risks associated with incorrect information can be significant. Choose your applications wisely based on this risk assessment.
Pragmatism powers progress
The potential of AI is undeniable, but unlocking it requires moving beyond fascination with the technology itself.
By starting with clear problems, demanding tangible ROI, understanding the real capabilities and limitations of different AI approaches, defining terms carefully, and ensuring your foundational systems are ready, you can build AI solutions that deliver genuine, lasting value. This pragmatic, problem-focused approach is key to turning AI’s promise into a practical reality.
Thanks to Andrei for sharing his time and thoughts on the VUX World podcast. You can check out the full interview here.
