Introduction

When companies begin exploring the use of AI, they usually take one of two paths. Some focus first on tools and technology, launching platforms or features in the hope that people will follow and value will emerge. These people assume that seeing the tool in action will naturally drive engagement. In reality, this rarely happens as expected.

Others take a more people-first approach. They run workshops, initiate conversations, and offer training to help teams better understand what AI can do for them. And then start design and build AI-enabled use-cases. This route aligns well with common change management advice, where ‘successful transformation starts with people’.

However, the way that AI works is different from technology innovations that we’ve faced in the past. Training people to use AI solutions in their workflow, is different from training for traditional software or systems. It involves more than teaching where to click or how to follow a structured workflow.

In most of my previous projects, training involved a logical sequence. For example, first you explained the value of a system. Then you walked users through a set of screens and showed them how to perform tasks. With AI, that simplicity disappears. The interaction becomes more open-ended and conversational. People are not just following steps, but they are learning how to think with the tool.

AI has rules, but these are hidden

AI systems do follow rules, but these are often hidden behind layers of complexity. Most software tools operate within clearly defined logic. You press a button, something predictable happens, and you can trace the system’s behaviour through visible processes.

With AI, that transparency is missing. You write a prompt, receive a result, and the logic behind the output is not easily accessible. For example, to truly understand what happens when you talk to ChatGPT, you would need to dive into how large language models work - concepts like context windows, fine-tuning, and reinforcement learning. That level of technical depth is not always something most users want or need to explore.

This makes AI feel less like a tool and more like a black box, because so much is happening behind the scenes. That complexity can be intimidating and discouraging for people who are used to systems where cause and effect are much easier to trace.

The same applies to trust: people often struggle to feel confident using AI because they cannot see or understand how it works. Even when the output feels accurate, it can feel unsettling if the path the machine followed is unclear.

That is why training cannot just focus on outcomes. It also needs to help people build a basic understanding of how AI operates, where its limitations lie, and how to evaluate results critically. People need to feel empowered to ask questions, give feedback, and develop a sense of control, even if they do not fully grasp the underlying model.

Quality In, Quality Out

There is also the issue of expectations. Due to the hype around AI, people tend to overestimate its capabilities, often assuming that AI will simply figure things out. They give it a vague or underdeveloped prompt and expect a polished result. Or use a specific AI tool for a use-case that it's not developed for. When the output is weak, they assume the tool has failed and quickly disengage.

But the real issue is often the input. A prompt like “write a strategy” does not give the model much to work with. The result will be generic because the request was generic. Or the prompt is good, but to for example receive a high-quality analysis it should have been given to Manus instead of OpenAI.

This is a reminder of something many of us have heard repeatedly in data teams: garbage in, garbage out. And, for example, for tools like ChatGPT: Weak prompts produce weak results.

Effective AI training helps people learn how to craft better prompts, test different inputs in different models, and understand how to shape interactions in ways that lead to better outcomes. This is a skill like any other, and it improves with practice and feedback.

For example - this article was, of course, written with the help of AI. But it wasn’t a one-click solution - it required multiple iterations and manual refinements. And that's the reality of using AI effectively and where user’s expectations should be set: it’s not about replacing quality with automation, but about enhancing quality through collaboration.

If you don’t already know what "good" looks like, AI won’t magically get you there. It can generate content, suggest solutions, or analyse data, but the final quality depends on your ability to guide, assess, and refine.

Reflect on your own tasks. Are you using AI to go from 0% to 10% (i.e., just getting started?) Or are you using it to push from 80% to 90% (i.e., shaping something already strong into something excellent?) Especially at the beginning of shaping a new workflow, a quality output with AI comes in increments. It’s a process of building, testing, adjusting, and learning.

No one has all the answers

AI is evolving quickly. So quickly, in fact, that even the people who work with it every day are constantly learning and adjusting. There are no final answers and best way of doing something. No one has it all figured out.

In this, the goal of AI training should be to get give people the mindset and tools to define hypothesis on AI usage for their workflow and tasks, create their own structured experiments that helps them learn and improve through experience, staying adaptable as the landscape changes. This shift toward experimentation, learning, and continuous improvement is one of the most valuable mindsets people can bring to AI.

How we deliver AI training at Riverflex

At Riverflex, we do not have a perfect formula, but we are experimenting with approaches that feel honest and helpful. And honestly, the feedback we got so far has been encouraging.

Our trainings begin with simplicity and genuine conversation. We break down complex concepts - like large language models - into clear, everyday language, avoiding jargon and assumptions. Next, we introduce practical examples and use-cases that relate directly to participants’ industries and roles, making the material relevant and relatable. We then ask thoughtful questions such as: Which use case could genuinely make your day easier? What might spark curiosity for you or your team and inspire the energy to explore further? Most importantly, we create a pressure-free space for experimentation with different tools and use-cases. The goal isn’t to get it right on the first try; it’s to encourage participants to start exploring how AI can be applied to their work and workflows. We help them engage with a selection of tools, define meaningful use-cases, and develop a structured approach to ongoing experimentation so they can continue learning and evolving.

If you’re trying to get your team thinking with AI, reach out to Maite Zanasi at maite.zanasi@riverflex.com!