Transcription
The year 2026 is the very year when a revolution in your work will happen. And now artificial intelligence and neural networks are truly helping a lot, for me, for example, in my work on almost all issues. I have written a lot of programs, done a lot of work with analytics, conducted a lot of research. And now I realized that it's time to consolidate all this knowledge into a whole series of videos, which will be called, in essence, Trends for 2026. These are the very skills that you need. So, in essence, in 2026, to fully implement them into your work. And the first video in the series of trends will be related to neural networks in working with tables. I have worked a lot with analytics, with tables, and, of course, I have gone a long way here, during which I made a lot of mistakes. That is, I made all possible mistakes, which now, of course, I will show you in the video how to avoid them right at an early stage. Today I will show you in general how you can work with and implement artificial intelligence in tables. This video is practical, so you will see all the time codes below and can rewind to the method that will be of interest to you. Let's start with the main problem. In order to analyze any table and in general to create a pivot table, set all the necessary formulas, look on the internet for what formulas exist. This can take from one to two hours to ten, maybe even fifteen to twenty hours. And this is only for analyzing some file. Now neural networks can work with tables. Many still do not know this and do not use it in their work. They still don't trust it, they are afraid of confidentiality. And for many other reasons, they don't turn to neural networks to solve this request. And in vain. Today I will show you visually in general how this issue can be completely resolved, so that you are not even afraid of data leakage. So, in general, we have outlined the problem, and let's structure our video today as follows. Today I will show you several ways. First, it is working simply classically with a neural network, when we send it either the Excel file itself, and it has to analyze it or write some formulas, or when we want the data to be, well, manageable by us and we don't leak anything anywhere just in case, I will show you an alternative method. I will show you a method how you can use AI agents. Yes, in 2026, this is also essentially a trend, agents. We will have a separate video about this a little later, so subscribe to this channel right away and follow the updates. But we will consider AI agents that can perform automatic calculations in Excel tables, or Google Sheets. We will also consider specialized services that work with your data, can see it, can look at it, insert all possible formulas, can create pivot tables, or even write program code to analyze all of this. Well, and in the finale, I will show you a paid, well, an interesting tool that I generally love to use very much now, namely the use of a neural network to solve complex calculations. And only this neural network is capable of performing complex calculations. Therefore, friends, I think now is the time for us to move on to practice. And we will start with the first block, namely the use of a simple classic neural network in working with tables. Let's get to practicing. So, we will now use absolutely free methods to test and see how all this will work in practice. I have this table, which is called a summary of courses. This is about the sales of my courses. Well, this is all test data. So, we have the date, course name, price, number of students, and the month when this course was sold. So, I made a potential export. There are 61 records in total. What will we do? We will now go to ChatGPT. It can read Excel files, it can perform calculations, it can work directly within the table itself. But now we need, in general, some of its basic analytical skills. Therefore, let's go to it. The first thing we need to do is send our Excel table here. What do we do next? Yes, what do I want to do in general now? What do I need help from artificial intelligence for? I want to understand what my average check is for the products in general. I want to understand which courses brought me the largest number of students. That is, well, I made the most revenue from these courses. I want to understand how sales have changed over time, that is, from month to month, if there is any correlation between them. And I want, in general, to ask it to give me some recommendations. This table can be large, that is, 100, 200, and even 300 rows of data. ChatGPT can read this. But if, for some reason, your table has 500, 600, or more rows of data, then, friends, you will need other methods and other models. That is, ChatGPT will not cope with this due to the fact that it has a context window that it can process. This is the first block of information that I want to announce to you right away. The second block of information, what is very important to consider, is that all calculations that we will do using neural networks, that is, ChatGPT, Gemini, and other models, you should ask them to perform calculations through Python code. Or just write in your prompt, "do all calculations for me through code." Let's go to ChatGPT. And now let's write this simple prompt. I am uploading a table with my course sales for half a year. I need to understand what my average check is, which five courses brought in the most money. How did sales change by month, and give a couple of ideas on what to improve. Use code for data analysis. Thank you. Well, thank you is already my, let's say, pleasant habit, which is automatically always in the prompts. Not that it significantly affects anything. Let's send the prompt. The model will now launch the Pandas library. If you see that any free model you choose has launched the Pandas library, then this is a very good sign. That is, it really does calculations well using this library. So, we are waiting for the analysis and will already look at the results. And here's what it gave us. It wrote that the average check is 13,490 rubles. It calculated it differently, right? That is, I, for example, thought, probably, you need to take the total price from the cost per course, but it wrote that it is actually calculated like this: total revenue divided by the total number of students. Let's see. And it also calculated the top 5 courses by revenue. AI for business 574,000, ChatGPT in practice 329, productive, and so on. That is, it calculated everything, but we want to check if it did it correctly. And for this, we now need to do the same thing, only manually. We go to our table, calculate the total revenue. Let's sum the entire column. This is 937. And here we also look at the total number of students. Now let's calculate the total revenue. That is, it's already suggesting that I do this and writes that it's 5,715 rubles. And this is incorrect, because I now realized that I made a mistake. This is not total revenue, this is the cost per course. And we have the number of students. Therefore, let's add another column on the right and write "total revenue per course." We calculate it using the following formula. We multiply the price per course by the number of students. Thus, we get a total revenue of 2,120,000. Accordingly, we now need to set E63 / D63 in this way. And this comes out to 13,490 rubles. It calculated everything correctly. Now we need to see how much revenue we actually got from a particular course sold. We apply a filter, that is, we click on the column, select filter, and after that we can filter our courses. Let's leave, for example, "AI for business," as it is our best-seller, as it showed us, and get the total cost, revenue, rather, 574,770 rubles. Let's also look at "ChatGPT in practice." 329,670. Everything matches. Thus, we have sent him the file, he has broken it down, calculated it, and did it absolutely correctly. Be sure to double-check him. Without this, you can make a very big mistake, especially if you ask him to calculate it just like that without code, he will 100% make mistakes. Therefore, we are now seeing really cool analytics. What else can ChatGPT do? You can ask it to compile all of this into a single table, for example, with formulas, and it can do it for you. Since there is not much data, this task is not super difficult for it. If you have a paid subscription to it, you can use the Chat model. For example, the classic one, without the "on" or "min" suffix. That is, we do not choose small models, we only work with large ones. And it will do it without any problems. A method how this can be done more safely, if you are worried about, well, in general, data security. Let me show you that too. For this, we will use an alternative method. This is a safer method for your data. I will now write a pre-prepared prompt here. It looks like this. I am uploading a screenshot of a table with my course sales for half a year. I need to understand what my average check is, which five courses brought in the most money, how sales changed by month. How can I do this within Excel? Describe and provide the necessary formulas. You have the data and column names. After that, I go to the table, take a screenshot. If you are very worried about the columns, replace a few rows of data that contain confidential information with, well, some neutral ones. Be sure to highlight it so that the column names are visible. That is, A, B, C, D names, for example, row names. But row names are already optional. This is how we do it and send him this screenshot. A screenshot is enough for the neural network to start processing your file, but it will process it not from the perspective of calculating something, but to give you recommendations and in general, well, essentially, a route map. Go here, enter this formula, and you will get this. And here it writes: "Average check." There are two options. Option A, recommended. Average check per student. Sum of E divided by E, well, that is, the entire column E divided by the sum of D, which does what? It sums up all revenue, divides by the total number of students. You will get how much one client pays on average. Let's check this formula. And let's insert our formula into some empty column. Everything, as you can see, it has already been applied automatically, and the average cost has been calculated. And correctly, just as ChatGPT calculated it before. Now I want to look at the courses, what it offers me with a breakdown. Top five courses by revenue, with one formula without a pivot table, but it will still look like a pivot. This is really nice. We copy this entire large formula, go here, delete the data from here, and here, attention, we insert this formula not like this, otherwise we will insert everything into different rows. We must do all of this in this way. That is, we insert it all directly into one line. Press Enter. And here, please, we already have a breakdown with our courses in general. If you want there to be six courses, make it six. If you want all of them, ask ChatGPT to create an automatic formula. Thus, as you can see, it's not difficult at all. You can work with your tables. The most important thing is that it will send you ready-made formulas that you can use, and will also help you create pivot tables. The disadvantages are that it is almost manual work, that is, you will spend time on it. The advantages are that it is still automation, but you will know for sure that it is calculated absolutely correctly, because all data will be taken into account from your entire table. And of course, it is also safe. With this, we are ready to move on to a more advanced way of working with tables. Method number two. Method number two is working with AI agents. There are already many of them on the market. There is Manus, there is G Spark, there is even Chat Z, a Chinese model with JLM on board, and there is Chat LLM or AC. All of them have good functionality related to tables. We can also use these tools for more advanced work. What needs to be done? We now open Manus and write this prompt. I am doing this on a free subscription. As you can see, you will get 300 tokens every day. This will be enough for you to perform simple, quite fast tasks. We write this prompt here. I need an Excel file to track my expenses. Make me something like this: columns, date, category, food, transport, entertainment, and so on. Amount, note. Below, automatic sum for each category. A separate tab with a graph. how I spent by month. And also add the percentage of the total budget for each category. My budget is conditionally 100,000 rubles per month. I take this prompt, copy it, and now we will even conduct such a test and see how different AI agents cope with this. While MAUS is thinking, we move on to Genk. We will also use a free subscription. This is important. That is, we have 100 credits for performing tasks of this kind. This is also more than enough. We go to the AI Sheets section here and paste our prompt here. Thus, we will already have a prepared agent with tools for working with tables. Let's launch it. We do the same with EBACUS or chat LLM. That is, here we have a paid tool among all agents. This is the only one where the entry is immediately paid. It costs from 10 dollars, but it has very advanced work with tables, which I absolutely love about it. We go to the Deep Agent section. And paste our prompt here. Let's launch it. We will now see how each of the agents coped with this task and what table it prepared for us. So, manus has already prepared everything in less than a minute. I have already downloaded this file, and this is the file it made for us. There are tabs, there is in general a prepared table, percentage of budget, expenses, and analysis, expenses by month. It's unclear how it will account for expenses by month. Apparently, we will need to create additional rows below. Let's try to check how this will work. I will now add one row of data for food and add a large budget there. Currently, it's 1.00 ruble to drive. Let's copy this row Ctrl C and place another one right below it. Let's choose the category "food" here or write "food." And here you can choose, these are drop-down lists. Great. And let's make it that we spent 40,000 rubles on food. Products for the month. We see that food has been calculated in the overview. We see that food is also accounted for here, and the percentage of the total budget has also been recalculated. In the analysis, we see that it has also taken this into account. The only thing is that it wrote that it is February. Let's try to change the month here. Let's set it to January. Let's see. And nothing has changed here. That is, we will need to somehow change something with formulas here. That is, manus coped, but we will have to delve deeper into understanding how this whole system works. Now let's go and see what Gen has done for us. Genpark is already showing us everything within this table. Here we see expenses, and here we see a graph by month. Let's look. Here everything is already calculated using formulas. And the percentage of the budget is also calculated through formulas. Let's check if we can create new rows. And so that it is taken into account, let's do that too. For example, let's still choose a date for food, change it to Let's make it the fifteenth, for a purchase of 3,500. Everything is recalculated. The category is food. Well, is it added? Yes, it's added. Everything is great. Everything works here without a hitch. To download the result, click on export and save it as an Excel file and take it into service. Great. Let's move on to our last tool agent, namely Deep Agent from ACUS. It will always ask us a few questions: the format of the graph by month, the percentage of the budget, and pre-filled data. Let's now answer all of this automatically off-camera, and we will look at the finished result. I have sent answers to three questions in general. I want a breakdown by category for each month. I want to see the percentage of the monthly budget for each category separately. And I want it to give me pre-filled data, so that there is at least some. That's it. And let's wait. Now let's look at the final result. Click "Allow editing." And let's look. We see the date, category, amount, and note. We also have a summary by category. Percentage too. Everything is there. There are graphs. The graphs also look quite interesting. That is, it has done everything. Let's add some large expenses now. That is, we have the health category, for which we spent 4,495. Let's add a new row. Let's copy the date here. On the twenty-third, we did something there. Let's enter "health" here and add, say, 15,000 rubles. Everything. So, it has been recalculated automatically for health, we also have 19,000. Everything is great. In the graphs, this should also be here. The only question I have is what these categories are here, right? What dates did it add here? It's unclear how to work with this. That is, you need to ask it for instructions. Now let's move on to analyzing the results. It's important to understand, this is AI. They work with Excel files, they can create pivot tables, they can analyze large data, but this is still, well, for example, 20-40,000 rows. If you want it to calculate everything itself, then with such a number of cells, rows of data, it will not cope. But it can work with your Excel file, automatically write formulas into it, create pivot tables, and have it calculated within it, it will definitely cope with that. Therefore, you can create large tables in AI. The main thing is to ask it to create Excel files, so that everything is verified by formulas. Thus, you will be able to get 100% results. Among the disadvantages, I can say that, as with neural networks, you essentially transfer all the information, so it's better not to transfer surnames, first names, patronymics, any, well, confidential data here, or somehow encrypt them, and then you will have, essentially, safe work, essentially, fully automated. And now, before we move on to the third method, friends, I remind you that these are AI trends. This is only the first of many different videos, so subscribe to the channel right now to stay informed of all events and know all the main trends. Also, subscribe to my Telegram channel, as all the main AI announcements, all the main models, all possible functions, life hacks, and of course, news from me, including the prompts that I used in this video, including all the tables, that is, all of this will be attached. Look for the post from February 3, 2026. We are moving on to method number three. In order to move on to method number three, let's first look at what kind of table we have. We have a breakdown by month for 1 year, revenue, number of sales, average check, source. This is either organic or, essentially, advertising. Everything is very simple. Let's download this file and go to a tool specifically designed for working with tables. It's called Quadratik HQ. This is a tool that works, including for free upon registration, but in general, it's not expensive. That is, a subscription costs only 20 dollars. And you can work with any amount of data, whether it's 20, 30, 50,000 rows of data. It doesn't matter at all. Each month you will be given five messages for free, and you can work directly with tables. And it's as if ChatGPT is there. In the Pro plan for 20 dollars, you have an unlimited number of messages with artificial intelligence, an unlimited number of automatic connections to pull databases. And you can upload as many files as you want. I have been working with these tables for a long time, so I can confidently recommend it to you. Click import local file and upload our Excel file here. Then click Next everywhere. It essentially guides you through the training. And here, essentially, all we need is the ability to write messages here. The mode is set to auto. The model Clot OPUS 4.5 is automatically selected. At the time of recording, this is the most advanced model for analytics, which works very well with these tables. And moreover, you can use it for free five times. Here you can attach a PDF, an image, a CSV, some additional file, that is, work with two tables simultaneously. You can connect to a specific source from which the database will be pulled, if you want to upload it from somewhere from, well, from outside, for example, from a website. You can also use a reference list. That is, you can set different tasks between sheets. And there are also additional options. This is essentially what should be used as the language for working with tables. And here we have formulas, which are automatically enabled. Python code for more accurate analysis, when it launches program code, or JavaScript. I recommend, since we are working with Excel, to leave only one tab. This is formulas. Let it calculate everything through formulas. We also have AI Rules. This is essentially a pre-prepared prompt, with which you can essentially write who it is and what it should do, how it should, in general, perform all possible calculations. In general, we also have a Privacy tab. And here it is very important that you also take this into account. Here there is an option to disable Quadratik from learning from your prompts. That is, this is exactly the conversation about data security. Enable this tab when you have a paid plan for 20 dollars, and you can work with your tables without fear that your data will leak somewhere. But not so much the data itself, but rather the prompts. That is, the system will not learn from your prompts. Regarding AI functions, you see here that our data is not stored in OpenAI, in ChatGPT, in Cloud, AVS Bedrock. That is, any model you choose will not store your data within itself. But yes, it will still run them through its servers. Therefore, again, even here I want to tell you right away that it is better to delete all conventional data or encrypt it, replace it with some unique identifiers. Then you will be absolutely sure that you are working with data safely. Let's close all this. And I have already pre-prepared three tasks for it that I want it to do for me. Find months where there were some oddities, sharp drops or surges. Explain what this might be related to, if you see patterns, and forecast sales for the first 3 months of 2026. Let's send it. That is, here we have analytics, as well as forecasting. Let it do all of this through Well, with forecasting, of course, I now understand that I made a mistake. I should have given it the opportunity to connect Python. For forecasting, it's better to use it. And so, it will now give us analytics and a slice through formulas. Here it has calculated everything. March, sharp drop, revenue 245,000, drop of 36%. It has entered the forecast separately here, and it is calculating it now. Let's wait. So, what do we see here? We see that it has calculated everything. It has added formulas here. We won't see this here until we download the file itself, but it's very easy to do here. Here it has added several additional requests for me. For example, build a revenue graph for the year. It can create pivot tables, it can forecast more accurately. That is, if you add Python to the languages, it can do all possible visualizations, graphs here. For example, in one of my cases, for example, I had a report on Yandex Direct for a whole year of advertising campaign work. It cleaned up the platforms for me, that is, it helped clean up the platforms. It suggested to me, advertising campaigns, which ones, accordingly, need to be optimized and how to optimize them. That is, this is indeed a practical tool in the work of almost every person who works with analytics. Here it has drawn me, you see, a graph, right? That is, revenue minimum, that is, it has indicated here even with some additional comments. Minimum of the year -36% to February, maximum of the year 45% to October. That is, this looks nice, beautiful. Now let's download this file. Click on the file, download. And you download it as Excelx. That is, you download it as an Excel file. Let's check if everything has been downloaded. Everything is here. Everything except visualizations. We can download visualizations separately, just as an image, and insert it here. That is, I will now close this file with the right mouse button. Ah, and you can copy it as PNG, right? Copy as PNG format and Ctrl V, insert this graph here. That is, you essentially worked, saved the graphs you need, and transferred them to the file you have. This is a really cool method. Quadratik HQ is needed by those who constantly work with tables and analytics. That is, this is a very practical tool. It can be used, including for super complex tasks, because here it can read absolutely all Excel, including working with all sheets. That is, it can take into account all sheets in its work and create pivot tables for you. And now, friends, we can move on to the fourth method of working with models. And this is precisely the Clot model, model 4.5 OPUS. If you want to work with a model that is capable of creating entire Excel books for you, in which there will be very large and complex interdependencies, well, essentially, calculated and all this is automated, then you can use the Clot neural network. And let me show you what we will give it now. We now have three tables. That is, attention - this is marketing, all channel costs, with a breakdown by year, by each channel. How many expenses do we have here? [music] Here we have specific sales for 2024. And here we have sales for 2025. Our task will be as follows. I want to know through which channels my return on investment has grown from 2024 to 2025. To understand which channel is the most profitable for us. I want to understand where I am overpaying and what can be, for example, suspended. I want to understand now how and where I can shift the budget, that is, to which channel. And additionally, I want to get recommendations from it on what is best to do. Therefore, I save all three Excel files. In this case, we will use Clot on a subscription, that is, 25 dollars. Switch to the OPUS 4.5 model. To perform complex calculations, this model is needed. Attach all three Excel files that we have here. I need to understand which marketing channels work best. Specifically, through which channels has ROI increased from 2024 to 2025 and through which has it fallen? Where am I overpaying? Where does it make sense to shift the budget in 2026? Give me the top three recommendations on how to optimize marketing. If you can, visualize the comparison of 2 years in a clear way. All calculations must be done in Excel with formulas, and visualization should also be done in it. Create a pivot table. Even with a paid subscription, well, unfortunately, the model does not allow you to go too far. But if you have well-equipped your prompt, even working with things like tables, and even a lot of data and comparisons between them on the paid Pro plan, this should be enough for you. To perform super complex calculations, pay for the plan, the plan is already max, it costs around 100-200 dollars. And you will work with it. That is, for those who want the maximum, this is an additional method. Just so you know that there is also such an alternative method. And there is only one model that can work with such complex calculations under the hood. This is opus 4.5. While all this is being done, I will immediately give you a list of some recommendations. Most models are capable of helping and prompting you, as we did with the ChatGPT model. Here's a screenshot of our table. Write me formulas, and it will write formulas. Almost all models, including Chinese models, will cope with this very well and will prompt you, write good formulas for you. ChatGPT, Gemini, well, Clot on free subscriptions will cope with this perfectly. Chinese models can in most cases, but not all. For example, DeepL does not always write formulas correctly for me. As for paid models, these are either AI agents or Quadratik itself. This is a complete solution for working with tables. Here I can only say one thing. If you want more advanced tools, unfortunately, you have to pay for them. That is, Acus copes best among all AI agents. Among the free ones, Genspark copes better than, for example, Manus. That is, it understands from half a word: I can test it, but I can't always work with it fully within the free tokens in one day. Therefore, sometimes you need to work in Genk not for one day, but for two or three, until these tokens are renewed. Or Ebacus for 10 dollars starting price will cope with this 100% very well and almost the first time. Clot OPUS is just an additional example that we will look at now. It's good when you have the opportunity to pay for a subscription, and it will be more than enough for you every day as well. And now let's wait for the final result and evaluate it. And here it has just finished everything, it has already outputted a table for us. Well, here we have different sheets in which it seems like there is nothing. In fact, we are not worried. This is an Excel file. Separately, it has built graphs, everything is very convenient here. Let's download this file now and see what we got in the end. Click "Allow editing." And here our data is already here. We see expenses, revenue, number of clients, ROI percentage, return on investment, and cost per action. 2024-2025. Separately, it compared 2024 and 2025. It brought everything here. It did an efficiency analysis. Efficiency, if a channel brings more share of revenue than it takes a share of the budget. We, accordingly, see that there are only two channels. Effective are Telegram and Direct. Efficiency less than one, ineffective. Overpayment is occurring. Good. It has written recommendations for us. Let's see. Google Ads. Optimize, reduce, increase budget for Direct, we have a drop for Telegram. Optimize, reduce. Top three recommendations. Scale Yandex Direct. Direct showed an ROI of, well, an increase of 112% to 152%. This is the best result among the channels. Efficiency 1.4 brings 28.8 revenue with 20% costs. Increase budget by 40-50%. Optimize or reduce Google Ads. Optimize Telegram. Save Telegram. So, look, we have received quite good recommendations on a separate sheet, even with visualization. This is wonderful. And here is our visualization. That is, everything, as you can see, is highlighted in the table. We see the graphs perfectly, that is, everything is very detailed, very clear. In this regard, of course, Clot OPUS is currently 4.5, but I heard that soon there will be some model 5, well, Clot 5. In the near future, if it comes out, expect a review from me, as usual. So I highly recommend you use it. That is, with a paid subscription, it is enough to perform tasks of this kind. And in general, the conclusions here are probably superfluous, because I voiced them when Opus was working. Use free methods. You know how to work safely with models. You know how to work with advanced tools, with artificial intelligence. These are AI agents. But remember that in general, to bring the work and the agent to the result, you need, if we use it for free, several days for everything to be taken into account. If we go the paid route, then in general, Genk on a paid subscription or Ebacus or Chat LM will cope with this perfectly on a paid subscription even for 10 dollars. Specialized solutions like Quadratik HQ, well, you are given five tokens for free, five prompts. And Clot 4.5 OPUS is used when the task is really complex. Friends, all the main prompts that I used will be in my Telegram channel, so at the bottom in the pinned message, be sure to subscribe to the channel. All the main news, newsletters, and all, well, all the most interesting things will be there. In addition, of course, subscribe to my [music] channel, give it a like, write comments about what you liked the most, what you use in your work. That's all for today. See you in the next video about trends. And for now, bye.