Bring Messy Company Data Into the Report

[00:00:24] exists. And the last video, the last lesson,
[00:00:27] we produced the owner's report,
[00:00:29] a live living system that someone,
[00:00:33] a client, a company would actually pay for.
[00:00:36] And today, what I want to go over is something that nobody really shows
[00:00:40] you in AI demos. What happens when clients send you something and
[00:00:45] a real money question is on the table,
[00:00:47] where there's anxiety, there's stress,
[00:00:50] there's pressure, and how we're going to settle that from the mess of the
[00:00:55] data that the client, in this case, has provided to us.
[00:00:58] So for a second, I want you to put yourself in the owner's seat
[00:01:03] for a second. Let's say it is Tuesday morning and
[00:01:08] a customer called Noreen,
[00:01:10] called your front desk. And says,
[00:01:13] I think you've charged me twice.
[00:01:17] Instantly, your stomach drops.
[00:01:19] You pull up the register and there it is.
[00:01:22] You see right there, two identical charges.
[00:01:24] The same day, two minutes apart.
[00:01:26] $225,
[00:01:29] both of them. And now,
[00:01:31] as a business owner, you're stuck in a limbo.
[00:01:34] You're stuck between Too bad. Options.
[00:01:37] One is, you could refund her, but what if it was already refunded?
[00:01:42] And you're paying her twice now. Option number two is,
[00:01:45] you do not refund her. But,
[00:01:48] in this case, if Noreen,
[00:01:50] the client who just called us, is right,
[00:01:53] you look like a thief. You are risking
[00:01:58] the reputation of your business online.
[00:02:00] In front of clients and customers,
[00:02:03] because, simply put, this is about money, guys.
[00:02:06] And when the conversation is about money,
[00:02:08] there is stress and this pressure involved. So,
[00:02:12] the not knowing feeling is the worst part of running a small business.
[00:02:22] So, this is a straight-up continuation of the previous lesson.
[00:02:27] I still have the exact same chat open over here,
[00:02:29] where we had created the reportproof.md file.
[00:02:33] And, once again, what we need open is this page over here,
[00:02:36] which you already are taking a look at. Cursor with the same setup and
[00:02:39] the owner's report in a browser tab. So,
[00:02:42] let's take a look at what the client, or in this case,
[00:02:45] the company, the medspa, has actually sent to us.
[00:02:48] Let me move this down over here, and let's take a look at the
[00:02:51] raw intake. So, first, I'll just quickly go through all of these images.
[00:02:54] And this is genuinely what it looks like at times,
[00:02:57] guys. It differs, but when you're working with small businesses,
[00:03:01] this is not uncommon. Messy systems,
[00:03:04] messy data, messy processes.
[00:03:07] And this is the opportunity for us,
[00:03:09] and where our expertise comes into play. She sent it us a screenshot of
[00:03:14] her ads manager, a screenshot of her booking calendar,
[00:03:18] a intake notes, the voice memo that she had sent,
[00:03:22] an invoice export, where,
[00:03:24] once again, if it's a PDF on cursor,
[00:03:27] this is not legible. What you need to do is right-click and click on
[00:03:30] open in the browser. Well now we can see it is parsed and here
[00:03:34] we have the invoice, the owner's voice memo and
[00:03:43] this is what we collectively have. So now let's go back into the exact
[00:03:47] PDF of the registration report where,
[00:03:50] if we zoom into this, we can see at the footer that she has
[00:03:55] sent us page 1 of 4.
[00:03:57] So a screenshot of one single day of the booking calendar.
[00:04:00] That we have over here. And adds screenshot
[00:04:05] and the one that says,
[00:04:09] okay, cancel that when I say the one that She says.
[00:04:58] And these folders, ahem,
[00:05:01] cancel that. And these screenshots,
[00:05:04] files, and PDFs are exactly the data that we have access to.
[00:05:08] And this is always the data. In this case,
[00:05:11] there was no API, no clean API. No export and no specifications.
[00:05:15] It is simply an edge case or a case scenario of messy,
[00:05:18] dirty data. And what we're going to be doing over here,
[00:05:23] cancel that entire section. I'm redoing it from scratch.
[00:05:32] And what we have over here, collectively, is the data that has been provided
[00:05:36] to us. A couple of screenshots, a voice note,
[00:05:39] and a PDF document. This is the data.
[00:05:42] And that is always the data, guys.
[00:05:45] But the beautiful part is that these screenshots,
[00:05:47] the AI can simply see this image,
[00:05:50] extract every piece of the information,
[00:05:53] and transcribe it. There's no API,
[00:05:57] there's no clean export, there's no specification.
[00:06:00] If our system always needs clean inputs,
[00:06:03] it is not really a system, and it is not prepared for the actual
[00:06:07] dirty world use case of real business data.
[00:06:10] And finally, what we're going to be doing over here is something really cool,
[00:06:14] where we're finally graduating from using copy-paste prompt templates,
[00:06:19] and really going into using and just talking to the AI,
[00:06:24] which is the right way to be, and this fast,
[00:06:26] modern way of using AI. So what we're going to be doing is,
[00:06:34] just watch me ask. Okay,
[00:06:36] cancel that.
[00:06:54] And guys, and guys, you've completely earned this.
[00:06:56] At this point, I truly believe that you have enough experience to wield and
[00:07:01] use AI freely, fluently,
[00:07:04] and just speak. So watch me just ask the Cut that out anyway.
[00:07:13] Again, completely. And guys,
[00:07:16] cut that out. And guys,
[00:07:19] you've completely earned this. At this point, I truly believe that you're well and
[00:07:22] prepared to be able to simply communicate to the AI.
[00:07:27] So watch me just ask the AI the way you talk to a Sharp
[00:07:30] employee. So what I do is I come back over here.
[00:07:33] Let me just close these folders over here, these files to clean it up.
[00:07:37] And take a look at this. I'm going to turn on WhisperFlow.
[00:07:43] So what I want you to do is read everything in the raw intake
[00:07:47] folder. Over here,
[00:07:49] the PDF documents, the screenshots,
[00:07:52] the voice memo, the intake notes.
[00:07:55] The owner over here has sent me a message thinking that she double charged
[00:08:00] a customer. Named Noreen,
[00:08:05] this is the spelling, and what I want you to do is trace every
[00:08:09] Noreen line in the register against the calendar and settle it for me.
[00:08:14] Was she double charged on her end? And what did she actually get?
[00:08:19] then also add a section to the owner's report telling me
[00:08:24] or telling the owner's report, sorry, what each of these files proves,
[00:08:28] what it cannot prove, and what to ask her for next.
[00:08:31] Do not change any company files.
[00:08:38] There we go. As you see guys, this is how I communicated to the
[00:08:42] AI. I just spoke to it like I was speaking to an employee.
[00:08:45] I even and messed up,
[00:08:48] I believe, when I was saying this. Specific line. And I just repeated myself.
[00:08:52] Something interesting you might have seen me do while this AI is generating everything.
[00:08:56] I just dragged the file into it. This is a practice I would highly
[00:09:00] encourage each and every one of you to incorporate,
[00:09:03] where when you're asking the AI to go over a specific file or folder,
[00:09:08] simply just drag it. So let's I wanted the AI to go over 4
[00:09:12] specific folds. Let's see. Let's say specific folders,
[00:09:15] the raw intake, the mybrain folder,
[00:09:17] the entirety of it, and the brain. What I would do is I would
[00:09:20] just dump this. Because simply put,
[00:09:23] the AI guys, it does not understand.
[00:09:26] If you just tell it the raw intake. In this case,
[00:09:31] assuming that there's only one folder or file in this entire system
[00:09:36] called raw intake, it was. It will be able to find it.
[00:09:39] But let's say we say something such as go into claw.md.
[00:09:42] As we know, there are multiple claw.md files.
[00:09:44] There's one over here and there's one over here. So what you want to
[00:09:47] be doing then is just drag the exact one into it.
[00:09:50] And another interesting thing you may have noticed is I typed the word Noreen
[00:09:53] over here. Because when we're using a tool such as WhisperFlow,
[00:09:57] or any of these dictation tools, it's highly likely or it happens very at
[00:10:01] times, I would say, not highly likely, but when you're using employees' names,
[00:10:06] when you're using words that are not something you would really expect
[00:10:11] the AI to pick up with 100% certainty,
[00:10:15] it's might as well just type it. Because let's say I said Noreen,
[00:10:19] but WhisperFlow wrote it as Norien or it picked it
[00:10:24] up as Maureen for example. For instance,
[00:10:26] what happens then is now the AI is searching for a person called Maureen
[00:10:30] for the double charge. And that instantly deprecates the quality because now the AI
[00:10:34] is ingesting information that is inaccurate,
[00:10:37] which is going to lead it to producing reports that are inaccurate.
[00:11:02] So you saw me just drag the files in.
[00:11:04] No paths, no setups, just drag it.
[00:11:07] If it's an image, dump it into it. If it's a PDF,
[00:11:10] ZIP file, you can pretty much drop anything into it.
[00:11:13] It's really simple. It's really cool, but even if you drop an entire ZIP
[00:11:15] file, the AI will unzip it and parse through the information.
[00:11:19] And if you do not have a dictation tool yet,
[00:11:22] the same ask as you saw me do is perfectly fine.
[00:11:25] You can simply type into it. Or,
[00:11:27] if you want to try, you can use the written version of this prompt,
[00:11:30] even though I would highly encourage you to begin using AI autonomously,
[00:11:35] by yourself, instead of relying on these prompts.
[00:11:39] And I did not do anything special. I just really described the exact situation
[00:11:43] that the client, in this case, messaged me.
[00:11:46] Nothing else, really. So,
[00:11:48] let's take a look at what the PDF has come out,
[00:11:50] and let's give it a read, actually, first. So,
[00:11:53] Noreen was double charged at the register, and Brin already caught and reversed it
[00:11:57] the same day. No company files were changed,
[00:11:59] and the trace is full over here.
[00:12:01] As we can see, what she actually got, a hydrofacial and her net,
[00:12:06] and her net at the register for $2.25.
[00:12:09] So, this also answers Marisol's voice memovery.
[00:12:11] I don't remember which sale number was right.
[00:12:14] So, the fix is done. What the files cannot prove,
[00:12:17] and added to the owner's report, where,
[00:12:19] if we check the owner's report, we will find that that information has been
[00:12:23] ingested into the system.
[00:12:25] Where I'm reloading the page,
[00:12:30] and let's scroll down to the bottom, where,
[00:12:33] here we go. We can see the answer,
[00:12:36] and it is much better than a simple yes or no.
[00:12:38] We see the exact date at which this thing happened.
[00:12:43] The exact timestamps, 1.30pm,
[00:12:47] charge, 2.20pm.
[00:12:50] 1.40pm, 2 minutes later, same charge,
[00:12:52] looks bad. But next day, at 5pm,
[00:12:56] we can see a minus 2.25 of a refund,
[00:13:00] that points at the duplicate charge by name.
[00:13:03] And the refund line itself points at the second charge as the duplicate.
[00:13:07] The one calendar screenshot, that also adds to it.
[00:13:11] It gives us additional confidence over here.
[00:13:14] That the calendar appointment, it names the exact charge that it belongs to,
[00:13:18] and it refunded it.
[00:13:26] We know the exact spe- Cut that.
[00:13:30] We know the exact specific treatment that this client came for as well.
[00:13:35] And they checked out at 1.30pm.
[00:13:38] Or sorry, 1.38pm,
[00:13:40] in this case, over here. And yes,
[00:13:43] she was double charged for about 27 hours specifically.
[00:13:48] Then someone on the team caught it and refunded it.
[00:13:51] And the net is one service, one charge,
[00:13:53] and the mistake has already been fixed. So now,
[00:13:56] our client in this case, the owner of the med spa,
[00:13:59] they call Noreen back with the entire appointment.
[00:14:02] The whole the whole holistic answer instead of a guess.
[00:14:05] And that exact call builds trust for us with our own client.
[00:14:08] And that allows our own client to call back Noreen and build trust with
[00:14:12] her customers.
[00:14:21] And let's take a look at what the report now says about the inputs
[00:14:24] itself, the images. The PDF document we're given.
[00:14:28] The PDF, page one of four,
[00:14:31] ten days, so no monthly conclusions about this exact system and the information
[00:14:36] for the month of June. The calendar,
[00:14:39] only one specific day. The ad screenshot,
[00:14:43] three months, or sorry, three months of ads,
[00:14:46] but the purchase column is completely blank.
[00:15:14] cancel that,
[00:15:16] I'm retaking it. And for each one of these constraints,
[00:15:20] the system names the smallest files to ask her for next.
[00:15:23] Never a complete login, never just give me access to your entire business and
[00:15:28] your tools. The smallest export that allows us to answer the question.
[00:15:32] And that really, really simple restraint is what allows us.
[00:15:37] And why or not? Sorry, why owners hand over their businesses
[00:15:42] to us. So now that the system eats
[00:15:47] what the clients actually sent and turns it into answers an owner can act
[00:15:52] upon plus a straight list of what it still needs.
[00:15:55] We have made some really, really clean progress and this really encapsulates
[00:16:01] a simple yet high-quality highly effective use case of the business brain in a
[00:16:05] real operating manner and environment working
[00:16:10] with businesses. So what we have after this is a 15-minute task on
[00:16:14] your page where what I want you guys to be doing is tracing Nadia
[00:16:18] from the calendar to the register and prove while no file here can say
[00:16:23] which ad got the credit for it. And the next video will be the
[00:16:27] last technical lesson. It is a little bit different and before this system ever
[00:16:31] touches a real client, we're going to attack it.
[00:16:34] We're going to stress test our own system on purpose.
[00:16:38] We're going to interrogate our own AI the way an owner
[00:16:43] and a business CEO will be interrogating ourselves.