When you work with AI every day, you get used to how powerful it has become. But most SME leaders I meet still think of it as a quick helper for emails or a shortcut for admin tasks. Useful, yes. Business changing, no. It is not something they would trust with deep thinking or genuine analysis.
That gap between how AI is used and what it can now do is enormous.
So I decided to run an experiment to show the other side. Two weeks before the Chancellor delivers the Budget, I asked four leading AI models to predict what Rachel Reeves is most likely to announce. Not guesses. Not surface level commentary. Full analytical work. Deep research, fiscal argument, political context and historic comparison. The kind of report a policy analyst or journalist would normally spend days producing.
Every model received the same long form brief. Ten pages of structured instruction, clear research tasks and specific data requirements. It was a serious test. After they produced their full reports, I pushed each one further with a follow up prompt to create a public-facing summary of its findings.
Most SME leaders never see AI operate at this level. They see the convenience tools. They rarely see how far you can push these systems with the right structure. This experiment is designed to show exactly what they are missing.
Why I Ran the Experiment
Many businesses tell me the same story. They have tried AI, but it fell flat. It helped with small tasks. It took the edge off the admin. But it never changed their decision making or their thinking.
In my experience, the problem is rarely the model. It is the prompt technique, the structure or the lack of depth. Leaders give AI shallow prompts, vague questions or tiny fragments of context, then expect deep analysis in return. AI is not magic. It is a tool that needs proper instructions. When you push it with structure, depth and clarity, the quality goes up dramatically.
This Budget project sits at the opposite end of the spectrum from the usual quick tasks. One complex national event. Four different models. One detailed brief. A clear benchmark for accuracy when the Chancellor stands up on Budget Day.
In a few days, we will see how close they got. More importantly, we will be able to see how each one thought.
The Four Models and What I Expected to See
The four models I used were ChatGPT, Claude, Gemini and Perplexity. Anyone who works with AI regularly will know they behave very differently.
ChatGPT is excellent at structured reasoning.
Claude has a real talent for long form analysis and nuance.
Gemini works well on economic patterns and data.
Perplexity brings speed and breadth.
I expected each one to take a very different path through the same problem. Some would go heavy on tax. Some would lean on political risk. Some would look closely at bond markets or productivity forecasts. One might prioritise public sentiment or polling.
Those differences matter. When you use AI for decision making, you do not want one model’s blind spots to become your blind spots. Four models in parallel are like four analysts in a room, each with their own lens. You get more clarity, more challenge and far fewer surprises.
The Process Most SME Leaders Never See
Here is the part that surprises most SME leaders.
Each of the four AI models has its own way of handling deep research, but they are all capable of working at a level most people never tap into. I used a full deep research prompt. Ten pages. It set the scope, the constraints, the rules of evidence, the depth of analysis required, the format for predictions and the evaluation criteria.
Each model then produced a long report, often thousands of words, drawing on its own reasoning patterns and research methods. Some pulled information and analysis from twenty to thirty sources. Others gathered insight from hundreds of sites and documents. The variation in their research behaviour was fascinating and showed just how much range there is across modern AI systems when you raise the bar.
Once the full reports were complete, I wrote a second prompt for each model asking it to produce a short form summary of its findings. These are the versions I will share publicly in this series so you can see how each system thinks without needing to read the full research documents.
What Comes Next in This Series
Over the next few days, I will publish each model’s Budget prediction as a standalone post.
You will see how they think. Where they agree. Where they go in different directions. Which issues each one highlights and which it quietly parks in the background. You will also see the gaps. The blind spots that only become obvious when you place four perspectives side by side.
Then, on Budget Day, I will rate them using a Strictly style scoring system. The scoring will be serious. Each model will be judged on accuracy, insight and how well it read the political moment.
Once the Budget has been delivered, I will publish a final wrap up comparing all four models against what Rachel Reeves actually announced and, just as importantly, reflecting on what this experiment tells us about using AI as a decision support partner inside real businesses.
If You Want to Use AI at This Level
This project is not a party trick. It is the same technique I teach in my training sessions.
Deep prompts. Structured reasoning. Multi model comparison. Practical methods that help businesses cut through noise, analyse complex situations and make better decisions faster.
Used properly, AI stops being a clever assistant and starts acting as a serious analytical partner. It can help you explore scenarios, pressure test strategy, understand markets and challenge your own assumptions with a level of rigour that was out of reach for most SMEs even a couple of years ago.
If your business wants to use AI at that level, you can book a training session or contact me for a conversation.
AI is not the future. It is already here. The real question for SMEs is how well they choose to use it.