Where It Is Truly Useful
AI is good at explaining how your heating bill is calculated, how different billing schemes work, and why in some buildings your actions affect the bill while in others they do not. This is exactly where you should start, because most people calculate their potential savings without understanding whether they can influence them at all. Instead of asking AI to calculate the savings, ask it to create a formula for a spreadsheet where you will enter your own data. A formula does not make arithmetic mistakes, it can be checked, and it will work every time you change the input figures. AI is good at structuring comparisons, meaning it shows which parameters change under different options and what you should look at. If you need to write an application to your condominium association about installing metering or requesting a recalculation, AI handles this task well because it concerns text rather than facts. You can paste the text from your bill and ask it to explain what each line means.
Where It Makes Mistakes
Arithmetic. This is the most dangerous issue because an error can look just as confident as a correct answer. AI systems regularly make mistakes in multi-step calculations, and the only way to check the result is with a calculator.
Tariffs. The cost of thermal energy differs between cities and suppliers, while electricity tariffs are reviewed over time. A model may either provide an outdated figure or invent a plausible one.
Rules and regulations. References to specific sections of documents, resolution numbers, and regulatory figures are the category where fabrications occur most often.
Your specific situation. AI does not know whether your building has a heat meter, what distribution system it uses, the condition of your insulation, or how much energy you actually consume. Without this data, any calculation is a fabrication based on average values.
I wrote more about the mechanics of such errors in a separate article about fabricated answers.
What to Collect Before Asking
Here is the list without which there is little point in asking AI. The billing scheme in your building, meaning whether there is a meter and what type it is. Your heating bill from the previous season, with the amounts for each month. Your heat supplier's tariff. The area of your apartment and the number of rooms. The type of building and the condition of its insulation. The cost of the measure whose savings you are calculating, such as the price of window insulation, thermostatic radiator valves, or a meter. Without these six points, you will get not a calculation, but a nicely formatted assumption.
How to Formulate the Request Correctly
The key principle is to provide the data yourself rather than asking AI to find it.
First, describe the situation, including the area, type of building, whether metering is available, and your actual expenses from the previous season with specific figures. Then formulate the task in one sentence, for example, to create a formula for calculating the payback period of a specific measure. Then specify exactly what you want to receive, meaning a step-by-step calculation with intermediate values that can be checked.
Also separately ask it not to use assumptions about tariffs and regulations, but only the figures you provided, and if some data is missing, to say so directly instead of inventing it. This last point is the most important because it noticeably reduces the number of fabricated figures.
How to Check the Answer
Ask it to show the calculation step by step rather than only the final result, because intermediate values make it possible to spot an error. Check the arithmetic with a calculator, and this is not excessive caution but a mandatory step. Ask the same question again using different wording, and if the figures differ, neither answer should be trusted. Any reference to a regulatory document should be checked against the original source. Any tariff should be checked on your supplier's website.
An Example of What Not to Do
A request such as “How much will I save if I insulate the windows in a two-room apartment?” will give you a confident answer with specific percentages. That answer will be fabricated because AI does not know your tariff, billing scheme, the condition of your windows, or whether your actions affect the bill at all. Worse still, if your building does not have a heat meter and billing is based on floor area, the correct answer is that there will be no savings on the bill at all, and AI will not tell you this because it does not know your situation.
When AI Will Not Help at All
When the question requires knowledge of local tariffs and decisions made by your local community. When legal interpretation of the terms of your contract with the supplier is required. When the actual technical condition of a specific building needs to be assessed. When exact regulatory values and standards are involved. In these cases, other sources are useful, namely your supplier, condominium association, relevant specialists, and official documents.
What to Do Instead
The most practical approach looks like this. Collect your data according to the list above. Ask AI to explain the mechanics and create a formula. Transfer the formula to a spreadsheet and enter your own numbers. Check the result with a calculator. Confirm the tariff and billing scheme with the supplier or condominium association. Then you will have a calculation based on your actual data rather than average figures from the internet.
