Blog

Speak the language of your audience, or watch your AI pilots stall

AI pilots rarely fail because the model was not good enough. They fail on the human side, when the experts who own the problem get lost in a fog of jargon. Translation is the lever.

Business ValueAI AdoptionAI FluencyClaudeGPTChange Management

This past week I spent time in two very different rooms. One was the Stockholm Tech Show, full of keynotes and demos. A lot of it was impressive, and a lot of it sailed right over the heads of the very people who are supposed to put AI to work, because it was pitched at engineers. The other room was a workshop I ran for an executive-search and recruitment firm. That one felt different. People left with things they could use the next morning.

The difference between those two rooms is, I think, a badly overlooked lever in getting business value from AI. It comes down to speaking the language of your audience.

Pilots fail on the human side, not the technical one

When I talk to leaders about why their AI efforts stall, they usually point at technology. The model, the platform, the integration challenges. But pilots do not fail because the models were not good enough. They fail on the human side, because the functional experts who actually own the problems, the recruiters, the finance leads, the HR partners, were never brought in, or were brought in and got lost in the translation, buried under a fog of "tokens," "embeddings," and "orchestration" diagrams.

Much of our industry talks about AI in a way that practically excludes the people who would create the value. We dress capability up in technical language, and then wonder why adoption is not working. You cannot extract value from a tool that the person with the business problem has been made to feel too unqualified to touch.

The workshop succeeded for reasons that had very little to do with how advanced the AI was, and everything to do with translation. Three moves did the job, and you can copy all three.

Move one: build everything from their context

I did not use generic demos. The examples were real recruiting and search workflows, tasks these people do every day. And I built the materials the way I was preaching. I used Claude to research their industry, update my slides, rewrite the exercises, and sharpen the prompts. I was practising what I teach, in front of the people I was teaching. That is a small thing that lands hard, because it turns the abstract claim "AI can help you with this" into a demonstration they just watched happen.

Move two: meet them in the tool they already had

Their organisation had approved ChatGPT, so that is what we used. I did not say "you should really be using something else instead." I took the tool sitting on their desks and showed them capabilities and small shortcuts they had never tried. The barrier to value was simply knowing what was possible with what they already owned. For most teams, the next bit of value is hiding inside a tool they are already paying for.

Move three: make the difference visible

This part stuck with the CEO, so let me give it to you as something you can run yourself.

I walked the group through a simple prompt formula I call RIGHT, for Role, Information, Goal, How, and Tweak. Then I added the idea of giving the model one or two worked examples, which is known as one-shot and few-shot prompting. If you want the mechanics of why a couple of examples lifts quality so much, I broke it down in what is few-shot prompting.

Then I had everyone do this:

Open two temporary ChatGPT chats, side by side. Pick one real task, something you actually need, like drafting a LinkedIn post or analysing a report. In the first tab, write a basic prompt, the kind you used to try until today. In the second, use the formula and add an example or two of what good looks like. Then compare the two outputs.

The gap was immediately noticeable and undeniable. Same tool, same person, same minute, wildly different results, with the only variable being how they asked. The CEO looked at his two tabs and said, "wow, this is so telling." That sentence was his whole takeaway from the day, and it cost nothing to produce. A visible before-and-after that made the value obvious.

Translation is becoming a core leadership discipline

Translating AI into the language and context of non-technical leaders is increasingly a discipline in its own right. AI Sweden runs a programme called AI for Executives, built explicitly for senior leaders who lack a technical background, in partnership with the universities of Stockholm, Gothenburg, and Uppsala. The stated goal is to give executives the judgment and practical footing to lead AI in their organisations. When the national centre for applied AI builds a multi-week programme around exactly this gap, that should tell you how central it is.

Translation does not mean dumbing things down. It means putting AI inside the person's own context and tools so that the value becomes self-evident. It is the practical, adoption-altitude version of the same lesson I drew from the Innovate Stockholm panel in this issue's piece on trust and diverse perspectives. The bottleneck was never the technology. It is human, and it is unlocked by translation.

It also happens to be the antidote to a problem I write about in this issue's term of the week, AI theater. A flashy keynote that nobody in the room can act on is theater. A two-tab exercise that changes how a recruiter works the next morning is value.

Your action step

Pick one team that is "not technical" and has been left out of your AI conversations. Find the tool they already have approved. Sit with them for an hour, take one real task from their week, and run the two-tab exercise. Show them the difference between a lazy prompt and a good one on a task they care about. Then watch what happens to their curiosity.

If you want this run properly across your organisation, a workshop tailored to a specific function, in the tools they already use, is exactly the kind of session I deliver. Get in touch and we can design one around the team that has been left out so far.

Frequently Asked Questions

Why do AI pilots fail in most organisations?
Pilots rarely fail because the models were not capable enough. They fail on the human side, because the functional experts who own the problems were never brought in, or were brought in and lost in a fog of technical jargon like tokens, embeddings, and orchestration. You cannot extract value from a tool the person with the business problem feels too unqualified to touch.
What is the RIGHT prompt formula?
RIGHT stands for Role, Information, Goal, How, and Tweak. It is a simple structure for writing better prompts: tell the model what role to play, give it the information it needs, state the goal, describe how you want the output, and tweak based on what comes back. Combined with one or two worked examples, it reliably lifts output quality.
How do you show non-technical teams the value of AI?
Run a visible before-and-after. Open two chats side by side, pick one real task the person actually needs done, write a basic prompt in the first and a structured prompt with an example in the second, then compare. Same tool, same person, same minute, with the only variable being how they asked. The gap makes the value self-evident.

Originally published in Think Big Newsletter #33 on Amir Elion's Think Big Newsletter.

Subscribe to Think Big Newsletter