Learning Library

← Back to Library

AI-Powered Excel: Prompts and ROI

Key Points

  • AI integration in Excel (via Claude and Microsoft Copilot) is a game‑changing development that lets large‑scale, complex spreadsheet tasks be handled automatically.
  • Claude’s newest Sonnet 4.5 model can extract and analyze multi‑currency data from a simple screenshot, but the strongest features currently require the pricey “max” plan.
  • Microsoft Copilot’s built‑in Excel builder, while a bit less powerful than Claude with optimal prompts, is readily available to most users and still offers substantial automation.
  • The transition from text‑only LLMs to numeric‑aware models forces a new ROI mindset: we must decide which calculations are best left to AI versus human effort and budget accordingly.
  • Practical prompt techniques demonstrated in the video can shave hours off daily Excel work, and these capabilities are expected to become broadly accessible within months, so early adoption is advised.

Sections

Full Transcript

# AI-Powered Excel: Prompts and ROI **Source:** [https://www.youtube.com/watch?v=xCrd3ajdlzw](https://www.youtube.com/watch?v=xCrd3ajdlzw) **Duration:** 00:24:56 ## Summary - AI integration in Excel (via Claude and Microsoft Copilot) is a game‑changing development that lets large‑scale, complex spreadsheet tasks be handled automatically. - Claude’s newest Sonnet 4.5 model can extract and analyze multi‑currency data from a simple screenshot, but the strongest features currently require the pricey “max” plan. - Microsoft Copilot’s built‑in Excel builder, while a bit less powerful than Claude with optimal prompts, is readily available to most users and still offers substantial automation. - The transition from text‑only LLMs to numeric‑aware models forces a new ROI mindset: we must decide which calculations are best left to AI versus human effort and budget accordingly. - Practical prompt techniques demonstrated in the video can shave hours off daily Excel work, and these capabilities are expected to become broadly accessible within months, so early adoption is advised. ## Sections - [00:00:00](https://www.youtube.com/watch?v=xCrd3ajdlzw&t=0s) **Unlocking AI-Powered Excel** - The speaker showcases how Claude’s new Excel capabilities and AI prompting can automate complex multi‑currency analyses, save hours of work, and reshape decisions about investing time and money in AI tools. - [00:03:46](https://www.youtube.com/watch?v=xCrd3ajdlzw&t=226s) **Claude: AI Financial Modeling Assistant** - The speaker highlights Claude’s ability to quickly generate and organize complex financial statements, ROI models, and note‑taking tasks, stressing that effective prompting delivers significant time savings and value for finance teams. - [00:07:10](https://www.youtube.com/watch?v=xCrd3ajdlzw&t=430s) **Crafting AI Prompts for Excel** - The speaker outlines a step‑by‑step method for designing clear, intent‑driven AI prompts that preprocess raw accounting CSV data and produce a monthly P&L in Excel, emphasizing modest upfront effort and modular prompt construction. - [00:10:25](https://www.youtube.com/watch?v=xCrd3ajdlzw&t=625s) **Managing Ambiguity and Excel AI Integration** - The speaker explains how to flag uncertain cases for human review while demonstrating the new Excel co‑pilot’s ability to create spreadsheets, apply formulas, and streamline budgeting tasks. - [00:13:45](https://www.youtube.com/watch?v=xCrd3ajdlzw&t=825s) **AI‑Powered Financial Dashboard Guide** - The speaker explains how to input data into Claude to generate growth analysis, cash‑runway projections, visual charts, and professionally formatted key‑metrics dashboards for use in presentations. - [00:16:56](https://www.youtube.com/watch?v=xCrd3ajdlzw&t=1016s) **AI-Assisted Assumptions Sheet Creation** - The speaker outlines creating a master assumptions sheet, manually refining seasonal and hiring inputs, then using AI prompts to generate a cleaned spreadsheet and build a revenue model with cohort stacking. - [00:20:05](https://www.youtube.com/watch?v=xCrd3ajdlzw&t=1205s) **Briefing LLM for Data Cleaning** - The speaker outlines how to describe a messy data export—detailing its inconsistencies, specifying precise cleaning steps, quality checks, and output expectations—to give an LLM a clear, actionable briefing for automated data remediation. - [00:23:13](https://www.youtube.com/watch?v=xCrd3ajdlzw&t=1393s) **Unlocking AI‑Powered Excel Efficiency** - The speaker urges users to adopt Claude Max–driven AI prompts for Excel, emphasizing that even modest time savings can justify the cost and that AI has the potential to accelerate spreadsheet work by orders of magnitude compared to traditional methods. ## Full Transcript
0:00If you are not using AI for Excel, you 0:03are sleeping on the power of what it can 0:05do. It is one of the biggest changes I 0:07have seen in the last year for AI. This 0:09video is all about AI and Excel. It's 0:12about Claude. It's about C-Pilot. It's 0:15about what we can do that we couldn't do 0:17before. And I go farther than that. I'm 0:20going to talk about specific prompts and 0:22show you specific prompts and how I use 0:24them to save hours of work. And I'm also 0:27going to give you a sense of how we 0:30should think about these new emerging 0:32capabilities in AI and how they change 0:34our own calculations of what's worth 0:37doing for people versus what's worth 0:39doing for AI and how much we should pay 0:41for AI, invest time in AI when it 0:44continues to gain capabilities like 0:46this. So, lot to cover. Let's dive in 0:49first to what launched and why I'm doing 0:51this video now. Claude has launched 0:53Excel capabilities for the last couple 0:55weeks. It's extraordinary. It's really 0:57good. It got better with the launch this 1:00week of Sonnet 4.5. I was able to 1:02oneshot complex Excel analysis out of a 1:06screenshot for multiple currencies, like 1:08four or five different currencies. This 1:10week with Sonnet 4.5. It's gotten 1:12really, really good. But people don't 1:14know about it because it's in Claude's 1:16max plan, which is expensive. We're 1:18going to get to that in the ROI part and 1:20how you think about that. So just put a 1:22pen in that for now. If you're not on 1:23the max plan, this is still for you 1:26because one, Claude is bringing those 1:28Excel files down. You can look for them 1:30to come down market soon. Two, Copilot 1:33launched this week also an Aentic Excel 1:37builder, right? Like it can be in Excel, 1:39it can be your assistant, it can build 1:41Excel for you, can edit Excel for you. 1:43I've played around with it. To me, it 1:45reads as about a number two compared to 1:47what Claude can do with a great prompt. 1:50But it's the tool you got if you're in 1:51co-pilot and it's fantastic and you 1:54shouldn't sleep on it. And so I think 1:55it's great. It's also another reason to 1:57talk about this. The larger story is 2:00clear. We had LLMs doing words for a 2:04long time and now they do numbers. Now 2:06they do numbers. And that means that you 2:09need to think about ROI and value 2:11differently. And it also means you 2:14should assume that this video matters 2:16for you. this capability. The prompts 2:19I'm going to show matter for you 2:20regardless of what plan you're in today 2:24because the capabilities are coming to 2:25you. They're coming to you in three 2:27months. Everybody's going to have this 2:30maybe in a month. Everybody's going to 2:31have this. And frankly, if it's in 2:33copilot, most people have it now 2:34already. That is how quickly this is 2:36changing. So, how do you think about the 2:40value of Excel work? I feel like it's 2:43very very different from words. When we 2:45are prompting with words, primarily what 2:47we are doing is we have a thought and we 2:50are saving ourselves the hands-on keys 2:53time to write it all out. And whether 2:56you do that well or badly, that's the 2:58idea. Now, you can prompt badly and get 3:00malformed thoughts or thoughts that feel 3:02like AI slop that aren't really your 3:04thoughts, or you can prompt really well 3:06and get stuff that is your thoughts just 3:08written out really, really fast. It's an 3:10accelerator, right? That's fundamentally 3:12what words are. But it's different with 3:14Excel and I think the leverage is in 3:17some ways higher because what you get 3:20with Excel is the leverage to compute 3:23numbers and with Claude and with the new 3:26co-pilot launch that leverage is 3:28essentially production grade. It is not 3:31as good as a CFO that has been doing 3:33spreadsheets for 25 years. I don't want 3:35to overstate it. It is not as good as 3:38the person who has maintained the 20 3:39sheet tab for 15 years in your office. 3:42No, but it is an extremely good 3:46professional assistant that can prepare 3:49dozens of different financial statements 3:53or models or attribution analyses or ROI 3:58calculations. We'll get into all of 4:00them. Like you you'll see the prompts 4:01that I'm building. Like it can do all of 4:03the standard business spreadsheet stuff 4:0630 to 40 times faster than a human in a 4:09better organized way. And I want to call 4:11out Claude in particular here. Claude is 4:13a good notetaker. Claude is so obsessive 4:16about having a little tab that has all 4:18of your notes there. And I love that so 4:20much. And it's going to do that faster 4:22than than anybody in the office, 4:23including the CFO. And so really, when 4:26you think about the ROI for these 4:28investments, is it worth it going to 4:30Claude Max? Now, I don't really think 4:32about it as am I going to get the money 4:35back per se. I think about it as am I 4:38going to get the time back and do I need 4:40it now now now or can I wait a month for 4:42it to come down to the right plan. There 4:44is no question I will get value back in 4:46terms of what it can do for me if I know 4:48how to prompt it. And that's exactly 4:49what we're doing here is we're going to 4:50talk about how you prompt it so you get 4:52the value. There's no question at all. 4:54the the prompts you are able to build 4:57the revenue analysis the board ready 4:59models these are going to save dozens of 5:01hours for finance teams right they'll 5:03save dozens of hours for PM teams who 5:06are trying to build ROI models there 5:08Excel 5:10is extremely hard for humans to build 5:13because it's all predicated on the idea 5:15that you have to have control over every 5:17individual cell and know what it says 5:19and know how the formulas compute and 5:22that is inherently a very high compute 5:23task For us, it goes in our brains. This 5:25is why I lose my hair. I had to do a 20 5:26sheet Excel when I was a marketer back 5:29donkeys years ago. And like the gray 5:31hair is real. You have to think of these 5:36as multiplied accelerators when it comes 5:38to Excel. So if they just accelerate 5:40linear production of words already LLMs 5:42are amazing. But with Excel, they are 5:45giving you not just quick thoughts, they 5:48are giving you full models that you can 5:50use to do real work with. To me that is 5:53actually more time leverage. It saves 5:55you another order of magnitude more 5:58time. So if I am trying to write a 6:02product requirements document and I can 6:04get it done now in 20 minutes instead of 6:063 hours that's great. If I am trying to 6:09build a set of board level analyses and 6:12I can get that done in 60 minutes 6:14instead of 8 days in a team of three 6:16it's another order of magnitude of 6:18improvement. It's a big deal. Now, now I 6:20do want to emphasize that the entire 6:23workflow shifts and moves the data 6:26collection burden to you early in the 6:29process. You still would have to collect 6:31the data for Excel regardless to get all 6:34of this work done. That doesn't change, 6:36right? Same data, got to get it. You're 6:38building an analysis. You need the same 6:40input numbers. They got to be right. But 6:41you have to get them up front now. And 6:43so for people who are used to saying, 6:44"Okay, I'm going to go to the next 6:45sheet. Going to go get a cup of coffee. 6:47Going to go pick this up off of Daryl's 6:49desk." and then come back and look at 6:50it. That's different now, right? You got 6:52to get it all in advance. You got to 6:53make sure you have it all. Now, you can 6:56give relatively messy inputs. Like I 6:59said, I gave a screenshot with numbers 7:01in multiple currencies and I got frankly 7:05surprisingly good outputs from sonnet. 7:08It was like one mistake and I corrected 7:10it quickly and that was it. I think we 7:12are sleeping on the on the that we get 7:15if we are willing to do just a little 7:17bit more work up front. So do the little 7:19bit of work up front, collect the data, 7:21and without further ado, let me show you 7:25four examples of prompts that I am 7:27building and using for AI in Excel and 7:32why I think they're so powerful. I 7:33actually built a bunch more of these. I 7:35think I have like 18 of them and I'm 7:36putting them all together into a prompt 7:38pack and and they'll be in the post, but 7:40but I want you to at least get a sense 7:42of the structure because my goal is for 7:44you to understand how I put them 7:46together. And there's not magic to it. 7:47It's really about communicating intent. 7:50And so, let's go look at our first 7:52prompt. Okay, this is actually a 7:54relatively short one. These are an 7:55ascending order of complexity. I thought 7:57I'd start you out easy. This is not 7:59multiple hundreds of lines. This is just 8:01a few dozen lines per step. Our goal is 8:04to get a monthly P&L from raw data. 8:06Let's say you have a CSV export from 8:07your accounting system. It's a little 8:09bit of a mess, but let's put it 8:11together. My recommendation is that you 8:14actually do these in pieces. And so 8:16you'll notice that these Excel prompts 8:19do come in pieces. They presume that you 8:22will get a deliverable at the end. If 8:24you notice at the end of this initial 8:26step, it is return the clean data in a 8:28new sheet called clean data. Just in 8:30case, for example, you run out of 8:32context window on one of the larger 8:34prompts. This allows you to continue 8:36your work and it doesn't become one of 8:37those nightmares where like the thing 8:38runs out of context and it does what 8:40Claude at least does in a very 8:42frustrating way and just says, "I I 8:44can't do this. were out of context and 8:45then you lose like half an hour. It's 8:47designed to prevent that. So it's very 8:49simple. You declare your columns and 8:52then you just declare your intent. 8:53Please create a clean data table and 8:55then you start to define cleanliness, 8:57right? Dates must be formatted properly. 8:59Amounts must be formatted as currency. 9:02Remove duplicate transactions. This is 9:04an example of something where like if I 9:06knew what was in the data set, I would 9:08probably be much more declarative about 9:10what counts as a duplicate to avoid 9:12false positives. Uh, and so you can 9:14expand that, make it yours. Number five, 9:16add a column for month. That's something 9:18that would be relevant for most people 9:19to be able to filter. And then flag rows 9:22with missing data. You can identify 9:24other steps in your data hygiene 9:26process. Most organizations will have a 9:29data hygiene process that they can go 9:32through that actually lays out all the 9:35steps. Like it's an SOP. It's a it's a 9:38standard operating procedure. You can 9:39literally encode that as a prompt, 9:41right? doesn't have to be a manual thing 9:44that I do every time I open up the sheet 9:46because this is what I would have to do, 9:48right? I used to have to do this all 9:49manually. Then second step, new prompt. 9:52You use the sheet and now you start to 9:54do work against it. Now you basically 9:56take it and start to categorize it now 9:58that the data is clean. And so you're 10:01going to define categories. These would 10:02be up to you, right? I made them up, but 10:04you can define what you want them to be. 10:06Right? This this seems standard but you 10:08can sort of adjust them. Uh and then use 10:11the description. This is something only 10:12LLMs can do, right? This just was not 10:15possible before with rules-based uh 10:17categorization. I know I've written 10:18those sort of if thens and they're 10:20nightmares. This is much much easier. 10:23The examples are good. If you feel like 10:25you have some areas of concern where 10:28it's like I don't know if this is 10:30correct or incorrect or a duplicate or 10:32not, you can include counter examples. 10:33You can include tiebreakers or rules 10:35around ambiguity. Um, and I love this at 10:39the end. Mark that it needs review and 10:41add a note explaining why it's 10:43ambiguous. Yeah, that makes a lot of 10:46sense because you don't want it to try 10:48and declare something ambiguous. You 10:50want to give it back to a human. All 10:51right. Then we get to the monthly P&L. I 10:53don't want to spend too long on this. We 10:55have three others to go into. You create 10:56a new sheet. Again, you're declaring 10:58your rows. You're telling it what 11:00formula to use. One of the breakthroughs 11:02that Claude launched into the world and 11:04co-pilot with Excel does this too is we 11:07have fluency with formulas finally and 11:10you can format it with bold headers, 11:11subtotal rows. Can you tell I've been 11:14having fun since Claude launched this 11:15Excel thing? That was the beginning for 11:17me, right? I was like, "Finally, it can 11:18do Excel. How much can it do?" Well, 11:20you're starting to see how much it can 11:21do. And and then we got it's like 11:23Christmas for nerds. We got to have the 11:25co-pilot launch with Excel, too. And so 11:27now we have multiple companies bearing 11:30down on this Excel problem space. It's 11:32going to be really exciting for folks 11:33that do any kind of work with 11:35spreadsheets in the next couple months. 11:37We get to budget variance. And so you 11:39can actually start to compute budget 11:40variance here. And you'll notice at each 11:43step we're just saying this is a new 11:45sheet. This is a new sheet. It will come 11:47back with a new sheet. And even if it 11:49runs out of context window, you're okay. 11:51And now you're doing one operation on 11:53it, right? Just add the columns. By the 11:55way, Copilot is really really good at 11:58this like just like do this one thing in 12:00the sheet. It lives in that embedded 12:02sheet whereas Claude treats it like an 12:05object and manipulates it from the 12:07outside. And actually Claude is still 12:09very very good. Like I said, I think 12:10it's actually better overall in my 12:12experience than Copilot with Excel in 12:14the sense that it has more ability to 12:17follow complex longunning prompts and 12:20get high quality results. But if you 12:23want the inline edits, like the ability 12:25of like the Excel magical AI, that's 12:27just something Microsoft has a 12:28distribution advantage on and they're 12:29just going to do it. And so this is an 12:31example of something where I sort of 12:32privately think Copilot is going to do 12:35well here. Uh and then you can validate 12:37it, right? You can add a validation 12:38section at the bottom and you can make 12:40it prove itself and display it as a 12:42checklist. And then this is what you get 12:44out of it. And so I do more, right? I 12:45have a sales pipeline dashboard. There's 12:46other ones, but this should give you an 12:48idea if you were an individual 12:50contributor of what you can do with 12:52Excel. And you might think, well, Nate, 12:54there's a lot, but like I'm saying, 12:55like, how often do you touch monthly 12:57P&L? You have to touch it every single 12:59month. And let's say it takes you hours 13:02and hours a month, and now it takes you 13:045 to 15 minutes. That's going to be ROI 13:06that pays itself back really, really 13:09fast. Let's get to the next. Okay, we're 13:11going to go through this a little bit 13:12quicker now that you know the idea. I 13:14don't want you to get bored. This one is 13:16for a board. Let's say you're a founder. 13:18You're trying to put this together. You 13:19want your chief of staff to put this 13:20together, what have you. You need a 13:22comprehensive financial package. What 13:24does that look like? You can go through 13:26and have it just create a revenue 13:28analysis sheet with comprehensive 13:30revenue tracking. And you're defining 13:31that all here. You're declaring it. This 13:33is what you want. By the way, these are 13:34very standard. And that's good. You 13:37actually want standard if you're talking 13:38to the board. So, this is this is a 13:40great thing. And it's just going to come 13:41back with this. Like, it's going to come 13:43back with the overall perspective. 13:44you're going to need to supply all the 13:45inputs for this. So, you should add that 13:47in when you write this prompt, but then 13:49it's going to come back with the growth 13:50analysis, the calculated metrics, and 13:52all the rest of it. And now you have a 13:53separate step for cash runway. So, you 13:55can go in and you can have inputs at the 13:57top for cash balance, for month, for 13:59burn rate. You get your projections, you 14:01get your alerts, your milestones, and 14:03you even get a visualization. And if 14:04you're wondering if it makes 14:05visualizations, yes, Claude in 14:07particular is very good at 14:09visualizations in a way that I am still 14:11surprised by. When I ask it to make a 14:13chart, it makes a good chart. And then 14:15you can actually give it this little 14:17instruction format for easy copy paste 14:19into board slide decks. Uh which is nice 14:21and touching. I you you can do all kinds 14:24of things people don't realize. I have 14:25told I have told the AI where to bold 14:28and it will just bold it in the right 14:30place, right? Like there's all kinds of 14:31stuff you can do people don't realize 14:32that make it seem more professional. 14:34Step three, you want to get to a key 14:36metrics dashboard. At this point, 14:37honestly, you're just kind of dunking on 14:39the competition because like you don't 14:41have to have this as a separate 14:42dashboard. It can be but but I wanted to 14:44show you what you could do, right? So, 14:45you can go through and you can calculate 14:47these metrics in a new sheet. Um, and 14:51you define how you talk about 14:52benchmarking here. You can define your 14:54metrics. Um, again, these are very 14:57standard metrics, so there's nothing too 14:58surprising, which is good because you 15:00actually want AI working off of standard 15:02metrics and also because you want your 15:03board working off standard metrics. 15:05Department spend breakdown very similar. 15:07You want to go piece by piece. Are you 15:09getting the idea? This may be hundreds 15:10of lines in total, but we're breaking it 15:13out so that it doesn't break the context 15:15window. And it it's going to save you 15:19days of work. It is worth figuring out 15:21how to use it so that you can save the 15:24days of work. And there's a ton here, 15:26right? And now you're now you're 15:27formatting it, right? Here's a board 15:28summary sheet. You talk about sort of 15:31what's in the overall report. you give 15:34it very specific formatting specs and 15:36export steps. Um, 15:39and honestly like this is this is very 15:41cloud specific because Claude can work 15:43across PowerPoint and PDF and Excel 15:46within one prompt whereas Microsoft's 15:48Excel agent won't. So I wrote this for 15:50the more complicated use case, but you 15:51could obviously drop it out if you would 15:53like. Is there other stuff? Yes, we can. 15:55The 13week cash flow model is a totally 15:57different thing. You you get the idea, 15:59right? It's a longer prompt. It's broken 16:01out in ways that the context window 16:02works for. It uses standard terminology 16:04which is good for the LLM and the board. 16:07And it presumes that you can gather the 16:09inputs that you need. Let's do a third 16:13one. We're going to go up yet another 16:14level here. Let's say you want to build 16:16a three-year business plan. Now, you 16:18might think, well, only founders need to 16:20do that. But honestly, everybody should 16:22build a business plan if they're doing 16:23anything at all related to a side gig. 16:26and everyone should be able to read one 16:28if you are in a business because boy 16:30does it shape your world whether you 16:32like it or not. Um, and so at least you 16:34should be able to understand this. And 16:35by the way, one of the things that's 16:37really interesting about AI as a sidebar 16:39is that the availability of material 16:42like this makes AI a phenomenal teaching 16:45tool. You can go into AI and say, you 16:48can go into cloud, chat, GPT, any AI and 16:50say please tell me why the three-year 16:53business plan has the elements it does. 16:55And you can get a whole lesson. You just 16:56get a whole lesson on it. Anyway, we 16:58create an assumptions sheet which serves 17:00as a source of truth. And so we start to 17:02do that first. We go through each of 17:04these. You'll notice we are getting very 17:06clear with our seasonable seasonal 17:07patterns. You will need to go through 17:09and sort of hand hand brush these over 17:12so that you have the right adjustments 17:14for your business. Right? Similarly with 17:16hires, right? You want to sort of be 17:18able to hand hand adjust this. We get to 17:20expenses etc. Now, the good news is, 17:23let's say you were like, "Oh, Nate, I 17:25don't want to go through and touch every 17:27single line." Well, you don't have to. 17:30You can go in and you can copy this 17:33prompt and then you can get all of the 17:36the the spreadsheet, the dirty CSV, the 17:39picture of the cocktail napkin where you 17:41have the data and you can say, "Put A 17:44and B together in a smart way, right? 17:45Like, here's my prompt. Do not follow 17:47the prompt. Instead, fill in the prompt 17:49with this information." and the LLM can 17:52do it. It kind of doesn't matter. Any 17:54cutting edge AI will be able to do that. 17:56And you go through and you say, "This is 17:57what I want, right? These are all of my 17:59assumptions." You go down to step two 18:00and now you're building a revenue model. 18:02It's a separate sheet. And you're going 18:04to go through, you're going to create 18:06the revenue model and stack the cohorts 18:08all the way through. Make sure that it 18:10understands how you want to build your 18:12build your business. And again, you'll 18:13need to adjust this for your business. 18:15We're assuming implicitly that there's 18:17almost a B2B format here because that's 18:19just become the standard for a lot of 18:20these the the way we talk about business 18:22in the software world. If you're not an 18:24MR business, don't use MR, right? Use 18:27use the the metrics that make sense for 18:29you. Then you go through and you can 18:30create the visualization, right? The 18:31point is to show you what you are 18:33capable of doing in these models. And 18:35then step three, you get to expense 18:37models by department. This is another 18:39level deeper. You can see already you're 18:40another level deeper versus the exact 18:42the exact ready board props. I don't 18:44want to bore you. This is a long prompt. 18:47The good news is I wrote it already for 18:49you. All you have to do is modify it. 18:51Super super easy. And so you go through, 18:54you can do that. You can do the 18:55headcount model. If you're a tiny 18:56startup, you don't need that. This that 18:58is part of why I broke this up. It's all 18:59optional. You can do a P&L rollup. 19:02Again, it's more complex in the board 19:03one, but it's good for a business plan. 19:07And you can do finally a cash flow 19:09statement. So there's a lot there, but 19:11that's good, right? Because at the end 19:13of the day, you actually get your entire 19:15business plan done. This is the kind of 19:16thing a bank needs or a venture capital 19:19firm needs to think about investing 19:21right. It's much more in-depth. But this 19:23kind of thing saves you hundreds of 19:26hours if you do it right. Let's look at 19:28one more. And I saved the best for last, 19:30guys. I saved the edits for last. Yes, I 19:34want to talk about the possibility of 19:36editing with Excel. That has been really 19:38challenging for models in the past. And 19:40I am so excited to share what I've 19:42worked on here. Okay, it is really, 19:44really important when you're editing to 19:46be clear about what you want. And so 19:50this is going to presume certain kinds 19:52of mess. My goal here is to show you how 19:56to clean up as a set of principles so 19:59you can clean up your own mess with 20:01prompts that are forked from this. 20:03Right? So for example, I have a messy 20:05data export from and then you'll have to 20:06specify where it's from. uh this is the 20:08current state. You want to describe this 20:10as specifically as you can. So 20:12inconsistent formatting, duplicate 20:13entries, text and number fields, unclear 20:16column headers, and this is what I want. 20:18And then specifically, this is how you 20:20clean the data. You want to be as 20:22specific as you can here. And it should 20:23map back to current state. This is what 20:26categorization looks like if that 20:27matters to you when you're cleaning the 20:29data. So again, you're going to specify 20:30that as cleanly as you can. Um this is 20:33how you start to work against it. And 20:36then this is your quality checks. The 20:38output is a clean data set with a 20:40summary dashboard showing total items by 20:42category, monthly trends and items 20:44required review. Ensure all formulas are 20:46error-free and totals reconciles to 20:47source data. You can actually do more 20:49than this in validation. If you get 20:50paranoid, you can make it produce a 20:52separate document showing how it did all 20:55the work and so it can go through and 20:56sort of explain to you how it did it as 20:58a way of auditing. 21:01What would you change with your mess? 21:04That is my question for you. Would you 21:05change the data has issues like so it's 21:09more specific? Would you maybe for 21:11example say 80% of our issues are 21:13inconsistent formatting and these are 21:14the six ways that we see it. Um those 21:17are all ways to give the the LLM more 21:20clarity. Basically think of this as the 21:22LLM's briefing before going into the 21:25spreadsheet. And you want it to be as 21:27comprehensive as you can so the LLM gets 21:30no surprises and doesn't hit a point of 21:33ambiguity. And so if you think it's 21:35going to hit a point where something 21:36might have inconsistent formatting or a 21:38mixed date format and there might be two 21:40correct answers based on your prompt, 21:41your prompt is not clear enough. And 21:43that is one of the reasons why edits 21:45fail. And I've come through and I've 21:46done some other edits like you can get 21:47the fix the broken formulas and all of 21:49that. These are complex prompts and 21:52these ones because they're edits are 21:54going to require another level of 21:56involvement from you because everyone's 21:58spreadsheet is broken in its own special 22:01way. And so you will have to go in and 22:03talk about how your spreadsheet is 22:05especially broken to get this stuff 22:07fixed. But this is really, as far as I 22:10know, the first real effort anyone has 22:12made to make AI do edits and not just 22:16create from scratch. It is it is harder 22:18because they're trained to create from 22:20scratch. That's how reinforcement 22:21learning works. it is still worth it to 22:24figure out the edit piece because if you 22:26can get a cleaner version of your 22:28spreadsheet, if you can fix your 22:29spreadsheets, again, dozens of hours 22:32saved. It's it's a huge time savings. 22:34Okay, you have been patient enough to 22:36stare at four different kinds of 22:38prompts. You have looked through IC 22:40level prompts. You've looked through 22:41executive ready board meeting prompts. 22:43You've looked through business plan 22:44prompts. You've looked through how to 22:45edit with Excel, which is net new. I 22:48hope what you walk away from is 22:50realizing that you have the ability in 22:54your pocket now to save hundreds of 22:57hours on complex business computation. 23:00To me, what has launched in the last 2 23:03weeks has the same leverage and power 23:06for numbers as the launch of the pocket 23:08calculator. That's how big a deal it is. 23:11and we're sleeping on it because most 23:13people either don't have the claude max 23:14plan, haven't really sat down and 23:16thought about prompting the co-pilot AI 23:18for Excel or they just don't have the 23:21time to come up with these complicated 23:22prompts and they've always done it this 23:23way and so they're just always going to 23:25do it. Let me challenge you. It is worth 23:28it to go a little bit differently to try 23:30a new prompt, to adjust one of my 23:31prompts. If it saves you that much time, 23:34if it saves you a dozen hours this 23:36month, the Claude Max plan is probably 23:38worth the money because you're going to 23:41spend that amount of time, right? Like 23:43your your time has some value to it. And 23:45so I I would love it if you guys would 23:49share ways you are using Excel and if 23:52you guys would talk specifically about 23:54what that speed up feels like when 23:56you're using it well versus the speed up 24:00you get from using AI for words because 24:04I continue to come back to this idea 24:06that AI for words was huge. It was like 24:08a 10 or 100x speed up. But the potential 24:10with Excel is a thousandx because of how 24:14complex these numeric structures are. 24:18They're not lists of numbers. They're 24:20complicated multi-heet hundreds of 24:23formula, multi-ellular structure messes. 24:27And now we can finally deal with them. 24:29So that's why I'm excited for AI for 24:31Excel. This has been I I really think 24:33this has been the biggest week for for 24:36numbers for leveraging numbers and going 24:39farther in business since the personal 24:41calculator. Like it's huge. Maybe since 24:43the original Excel. The original Excel 24:45was pretty cool. Maybe we'll give them 24:46credit there. But it's a big deal. So 24:47have fun. Go dig into Excel. As you can 24:50tell, I'm an Excel nerd. And so I hope 24:52you enjoyed this one. Share it with 24:53someone who loves Excel.