{
  "video_id": "EDRKA1EXsfc",
  "channel_slug": "ai-makerspace",
  "channel_handle": "ai-makerspace",
  "title": "Dysprosium Financial Assistant by Derek Yimoyines - AIE9 Demo Day",
  "duration_seconds": 593.0,
  "url": "https://www.youtube.com/watch?v=EDRKA1EXsfc",
  "upload_date": "",
  "transcript": "All right, good evening everybody. So,\nwe're going to talk about FPNA,\nfinancial planning and analysis,\naka the thing that your CFO should be\ndoing. Um, but they're probably not.\nThey're probably worried about billing,\ncollections, payroll, accounting,\nHR, leases, leases, contract\nnegotiations. Um, but they should be\nstrategic. the CFO, the office of\nfinance should be, you know, doing FPNA,\nfinancial planning. They should be\nworrying about should we open a new\nwarehouse? Should we be selling our\nproducts in Target? Should they be in\nCostco? Are we going to launch a new\nproduct? You know, should we is our\npricing model right. So, this is a\nfunction that typically is underserved\nin businesses. Um, it ends up being very\nreactive. As I said, they worry about\nproblems that are just like payroll and\nexpenses. It's typically not scaled with\nsoftware. It's scaled by adding six\nfigure financial analysts to the team.\nLet's add another analyst. We need more\nhorsepower. It's also really manual.\nIt's errorprone. And it's done in\nspreadsheet. Be surprised how many $500\nmillion revenue companies. That's their\nfinancial model. All right. 2025 budget\nfinal final version 23 official. That's\nit. Nobody touch it. That's the budget\nfor 2025.\nSo I'm targeting the FPNA or financial\nplanning and analysis user. In startups,\nthey're your CEO. They're the visionary.\nThey're driving all the growth. They\nhave the vision, but they're also\nthey're the business person. So, they're\ngoing to run with this business. As they\ngrow, they get more successful. They\nbring in a finance team. They might have\na CFO, somebody who can like keep tabs\non the market. Um, and they're trying to\nget, you know, trying to get insights\nand trying to grow the business. And\nthen other people that are touching FPNA\nare people that are, you know, demand\nplanners, uh, worrying about what\ninventory to buy. uh how much stock are\nwe keeping and then folks marketing\nworry about like how are we acquiring\ncustomers. So as far as a TAM there's a\n$5 billion TAM and CPG. CPG is consumer\npackaged goods, chocolates, lipstick,\nwhatever you can buy on store shelves in\nTarget, Walmart, Costco, that's a\nconsumer consumer package good. Um\nShopify 5 billion a year sold. And\nthere's about two billion users of Excel\nand G sheets. So I'm going to focus on\nthose communities. your sellers of those\ngoods and your finance folks using G-S\nsheets. We're do a little case study\npowder drink city hydration powder\nmultiple flavors. They use an overseas\nformulator which is very common. Steve,\nhe's the founder CEO. He's got 15\npeople. He's also the CF CFO because\nthere is no formal finance function.\nSteve's got a problem. October 2026,\nhe's got a rudimentary financial model.\nIt shows him that's when he goes cash\nnegative. So, he needs a plan. and\nyou're sitting there going like, \"Okay,\nI want to make this business work. I've\nborrowed all the money I can borrow. I'm\nnot borrowing anymore. How can I make\nthis work? Maybe next year we'll do some\ngummies, but until we get them, we're\ngoing to do it.\" So, one of the weird\nthings about\nsmall businesses or small CPG businesses\nis it's possible to be profitable and go\nbankrupt at the same moment, which is a\nbit hard to hold those two things in\nyour brain. But basically, because he\nuses an overseas formulator, it takes 20\nweeks for the goods to get to him. So,\nhe's got to outlay a couple hundred\nthousand dollars and he gets those goods\nin 20 weeks. In the meantime, he's\npaying his staff. He's paying his\nwarehouse. He's marketing. He's\nadvertising to customers. He's dealing\nwith returns. He's dealing with all\nsorts of issues. So, he's got to have\nenough cash to get through that period\nbefore he gets those products to sell.\nAnd usually, he's got to put down 50%\nupfront. So again, you can be selling a\nhundred bucks to each customer for\nsomething that cost you 30 and spending\n$50 you acquire customers. You can be\nprofitable and at the same moment you\ncan go bankrupt. So\nsince I've been working on this, a\nlittle thing called claude and excel has\ncome out and open and excel have come\nout. There's a little bit of a, you\nknow, check mark analysis on it. If you\ngo to cloud and excel, a bunch of our,\nyou know, folks at work use it. It's\npretty awesome. Open a Excel pretty\nawesome. They can do basic stuff. Um,\nyou can ask them for basic analyses. You\nknow, what's going on, what's going on\nthere. They can do some external data.\nUm, they can do some variance analysis.\nBut the thing I'm going to show today is\noutcome optimization, which I don't\nbelieve they can do. They may eventually\ndo it. These things are moving extremely\nquickly. As I said, I don't think cloud\nin Excel was around 3 months ago, but\nman, it has hit the market. So,\ninfrastructure, AI stack, um, Verscell,\nGoogle Sheets, Google Slides, AI\ntoolkit, Langchain, Lang Smith, Tava\nQuadrant for a little little bit of rag,\nand then, uh, the optimal optimizer.\nBasically, what it does is it uses Latin\nhyper cube sampling. The user puts in\ntheir goal and it does a goal seek. And\nI'm using a a hyperform library which is\na Python library which basically\nexplores and runs um an in-memory\ncalculation as if it was Excel or Google\nSheets engine based on formulas and it\ngets instead of having to go to an Excel\nsheet and plug in numbers and get\nresults, it does it in memory. It does\nit in a couple seconds and can run a\nthousand different permutations and\ncalculations. It saves a ton of time um\nwhich I'm going to show. So anyway,\nwe're a little little quick little demo\nhere.\nof the product. So, here we are. And my\nman Steve saying, \"All right, when do I\nrun out of cash?\" And this is a bit of a\nsplit screen, so I'm going to show you\nwhat's going on. So, we've got his\nfinancial model. And his model says he's\ngot business levers, things that he can\nchange, you know, how many orders is he\ntaking, what's the average order value,\ncost to acquire a customer, how many\norders is each customer making per\nmonth, and so on, discounts. So, these\ngreen things are things you can change.\nThe rest of these are just sort of like\nother outputs of the model. And down at\nthe bottom, we got the critical numbers\nof his business. The EB arriva, what's\nhis profitability look like and how much\ncash does he has? And as for mentioned,\nOctober, bad times ahead. He's in the\nhole for cash and he gets in a million\nbucks in the hole by March. Not good.\nThat's where Steve is asking for help.\nSo Steve can ask the model,\n\"Hey, when do I run out of cash?\" and\nbang it has the answer is because it's\nusing the G sheets API to communicate\nwith Google. So the data source here is\nhis financial model and it can handle\nmultiple different financial models.\nRight now it's got his budget but it has\nhis base, his growth, his cost\nreduction. So he can kind of do that's a\nvariance piece of it. He can be\ncomparing different models as he goes.\nSo that's the first thing he can ask\nstuff like all right give me three ways\nto improve cash flow in my business. So\nthat's where the rag part comes in. He's\ngoing to, you know, I built it a little\nuh database of financial terms and\nfinancial literacy so it doesn't go off\nto the internet and like find random\nstuff, but basically how do you improve\ncash flow, you know, incr improve your\nmarketing efficiency, lower CAC,\nincrease return on ad spend. But this is\nlike not really his business. This is\njust like generic. Yeah. Yeah. Okay. I\nknow this like you know CFOi training\nsheet sweet. You can kind of get this\ninfo anywhere. All right. Here's where\nit gets interesting and this is the\noutcome optimization. find me three\nscenarios where cash is over 200 grand\nby January 2027. Now, we know his\ncurrent model says he's underwater at\nthat point. So, now we're looking and\nsaying, \"All right, when do I get in\nJanuary 2027, can I have over 200\ngrand?\" So, it comes back great. It runs\nabout a thousand scenarios in a few\nseconds, pumps it back to the LLM, and\nin this case, Claude rips through them\nand says, \"Okay, what are the most pro\nwhat are the most feasible ones?\" So,\nSteve's sitting here going, \"Okay, this\nis great. like I want 800 grand in cash,\nbut I can't, you know, I can't increase\nmy average order value by 50% in just 6\nmonths. That's crazy town. I can't do\nthat. All right, next take. So,\nbasically, he's going to work with the\nagent like it was a junior analyst. And\nagain, a junior analyst would take a\nhalf a day or a day to go away and run\nscenarios by hand, but he's doing it um\nin the agentic interface. So,\nlet's take scenario two and change\nreturns only by 10% and let's throw\norders per customer into the mix. So,\nlet's add another variable. So, he's\nincreasing orders per customer. Again,\nhe loves the cash, but like this is\nstill a little too much. I don't think\nSteve can do this to his business, but\nloves the cash number. Let's try another\nturn. All right, let's do one where\ndiscounts CA and AOV change by less than\n20%. Model comes back and says, you know\nwhat, sorry, can't do that. Best I can\ndo is a96,000\ncash balance. Again, it goes in the\nbackground using the formulas from his\nlive model, runs a bunch of\ncalculations. In this case, it couldn't\ngoalsek anything. It just told him,\n\"Look, man, I can't do that.\" All right,\nfine. What if I can change these by\nmaximum 25%. And boom, we've got a\nscenario. That's a little more than he'd\nlike to change, but Steve says, \"You\nknow what? Since I only have to reduce\ndiscounts by five points, I probably can\nget my CAC and AOV in line. I'm going to\nhave 250 grand of cash.\" The other one\nit gave him was 45 grand. I can't really\ndo that. So, next thing Steve says is,\n\"All right, great. implement scenario\none in my model. At this point, we call\nback out to the G Sheets API, implements\nthe model, and as we can see, October\n276 was our out of cache date. We are no\nlonger out of cash. We're in good shape.\nWe got our 200 grand in January. So,\nbasically\noutcome optimization. We're finding\nsolutions. We're doing goal seeks. It's\nusing his actual financial model in the\nbackground, running calculations. Again,\nthis would be days of work for an FPNA\nanalyst. No need to hire that six-f\nfigureure hire. You get the answers in\nminutes in using some rags, some web\nsearch, Latin hyper cube, and we had\nmultiple different agents. So, you want\nto go play with the demo, it's at\ndprosium.ai.\nUh this is also what I do kind of for a\nday job, which is dried.io. We help\nbrands like Graza Olive Oil, Oats\nOvernight, and other big brands solve\nthese exact kind of problems. And\nthere's my LinkedIn in case you want to\nconnect later. Thanks,",
  "transcript_chars": 10623,
  "ingested_at": "2026-05-15T10:48:58.162761+00:00",
  "source": "channel",
  "yt_meta": {
    "view_count": 65,
    "like_count": 3,
    "channel_id": "UCbDZFHUjTCCUKyXgcp3g50Q",
    "categories": [
      "Science & Technology"
    ],
    "tags": []
  }
}