Start with context, not with question
The neural network does not see your file, so it has no way to guess the structure. Describe three things: what is in the columns (for example, “in A - dates, in B - cities, in C - sales amounts”), what should be the output and what program you are working in - Excel or Google Sheets, because the syntax and argument delimiters are different there. Specify the language of the functions: Russian SUMIF or English SUMIF. The more accurate the input, the higher the chance of getting a formula that can be inserted without editing. Make it a rule: first the context and the desired result, only then the request “give me the formula.”
Collect formulas for typical tasks
Most work requests are a few repeating scenarios. To summarize sales for a specific manager is SUMISF or SUMILIMS with several conditions. Pull up the price from the reference book by article - VLOOKUP or a combination of INDEX and SEARCH. Calculate the number of orders per month - COUNTIFS with a date range. Describe the task in words, and the neural network will select the appropriate function and assemble the syntax with your ranges. Ask for two options at once - a short one and a more error-resistant one - to understand the logic. This way, in an evening, you will complete tasks that previously took hours of searching on forums.
Describe the summary and formatting rule in words
Pivot tables and conditional formatting are intimidating interface for beginners, but they can also be ordered as text. Write: “You need a summary: rows - cities, columns - months, values - amount of revenue,” and the neural network will give a step-by-step procedure for your version of the program. For highlighting, ask for the rule: “highlight in red the lines where the remainder is less than 10,” and get a ready-made condition with a formula. Check the version - the interface of 2016 and 2021 differs in the location of the buttons. This approach saves time not only on formulas, but also on visual analysis of data, which usually takes the most effort.
Bring the entire error and get an analysis
Errors like #VALUE!, #REF!, #N/A, and #DIV/0! scary, but each has an understandable reason. #VALUE usually means that text was included in the calculation instead of a number; #LINK - that the cell on which the formula was based was deleted; #N/A says VLOOKUP did not find the value. Copy the formula along with the text of the error and describe what is in the cells - the neural network will show where it broke and offer a corrected version. Ask not only for a patch, but also for an explanation of the reason, otherwise the same error will return in the next report. Analyzing your mistakes is the fastest way to stop being afraid of them.
Check the results on your data
The neural network confidently produces the syntax, but does not know your data and sometimes confuses ranges or function names, especially in the Russian locale. Never insert a formula into an important report blindly. Test it on a small piece where the answer is known in advance: count the same three lines by hand and compare. Pay attention to delimiters—semicolons or commas—and absolute references with dollar signs, which are often lost during copying. Treat the answer like a draft from a knowledgeable colleague: almost always correct, but you are ultimately responsible for the numbers.
Ask for a formula and learn as you go
A ready-made formula solves a problem once, but understanding solves it forever. After each answer, add a request: “explain, piece by piece, what each argument does.” Then SUMILIMN will cease to be magic, and next time you will assemble it yourself or quickly correct it for a new task. Gradually, you will begin to formulate requests more precisely and notice that you describe tasks structurally - this is the prompting skill that can be transferred to any tools. You can practice setting precise queries in a free introductory lesson - they will show you how to turn a vague task into a clear query to a neural network.