Run Amazon
with Claude:
Course File

---

## Lesson 03: Amazon's official MCP, put to the test

Video: https://www.youtube.com/watch?v=WrOcHjpOd1o (13:43)

### What it covers

I installed Amazon's official SP-API MCP in Claude Code, ran it on a live UK account, and it still could not tell me my profit.

Amazon ships an official SP-API MCP server. This is what it actually takes to get it running: Claude Code, two credential files, a developer profile you had to apply for, and the install fighting back on camera. Nothing staged, nothing cut.

Once it works you can pull seven days of sales and request a Sales and Traffic report, which lands after a 45-second wait and three files of unpacking. Then the question that matters. Orders and Amazon fees, yes. Cost of goods, no. Ad spend, no, that sits in a second API. The same question goes to a data layer at the end, side by side, while the official one is still running.

The most technical lesson in the course. None of it is required to follow the rest.

### Chapters

- 0:00 What you need to run Amazon SP-API in Claude Code
- 1:06 The two credential files SP-API needs
- 2:35 Installing Amazon's official SP-API MCP
- 5:13 When the install goes wrong, live
- 6:25 Pulling seven days of Amazon sales
- 7:32 Requesting a Sales and Traffic report, and the wait
- 9:55 What SP-API can and cannot give you
- 11:15 Official MCP against a data layer, side by side

### Materials

- Amazon's official SP-API MCP sample: https://github.com/amzn/selling-partner-api-samples/tree/main/use-cases/sp-api-dev-mcp

### Prompts used in this lesson

Prompt 1:

Note: To use this prompt you need your own SP-API credentials first: an Amazon developer profile, an LWA app and a refresh token. Want an easier way? See lesson 04.

```
Hey, I want to connect to the Amazon SP API MCP. This is the URL. https://github.com/amzn/selling-partner-api-samples/tree/main/use-cases/sp-api-dev-mcp

Go ahead and check how to install it, and install it here for me. Tell me which SP-API credentials you need and where to put them, then run a quick test and let me know once you finish.
```

Prompt 2:

```
How much did I sell in the last seven days? Give me totals by day.
```

Prompt 3:

```
What was my profit last month?
```

### Transcript

#### 0:00 What you need to run Amazon SP-API in Claude Code

All right, let's build the connection to Amazon SP-API so you can actually run it from your favorite AI tool. I'll walk you through it step by step. One thing to mention is this is the most technical lesson in the whole course. So if you are not feeling comfortable, there's no need for you to really do it yourself. I just want to show you how to do it, what's possible if you actually want to really do it on your own, and then I will also show you the comparison to the shortcut that will be continuing in the next lessons. So let's do that.

So for this one you actually need Claude Code. There's no way for you to do it just from the normal chat, as we need to actually build something. We need to actually do a little bit more complex stuff. Fortunately enough, still, AI will do it for us. So I'm opening my Claude Code right now, the fresh new directory. You can actually see here, I just created this directory, run Amazon with Claude, and this is the Amazon SP-API dev MCP. Now how.

#### 1:06 The two credential files SP-API needs

To do it is basically very simple. So we need two files to begin with. Basically those are our SP-API credentials. So as soon as you actually get access to SP-API and you will have the application, the developer profile, then you will be able to retrieve the very important credentials. So this is pretty much the access token. The token that allows you to connect and get the data and do things with the SP-API. For the ads is literally the same thing.

So I have two files here: .env and mcp.json. Now I'll show you this, of course those will be blurred. So the first one is .env. So you need the SP-API client ID, client secret, refresh token that is very large, SP-API base URL, and the SP-API region. I have the UK account, so that's why I'm using here the EU and this URL. If you are using the US or others, there are different links, and everything is in the documentation. So this is the most important file. Without it, it will not work.

And the second one is the mcp.json. So this is the file that gives information to our AI about the MCP servers that we are going to use. And we have SP-API dev assistant and SP-API workflow. Most likely we'll just use one of them, but this is basically what the documentation says. You need those two. So we have them here stored. Now I've opened a new project, and we'll do it step by step. So we can literally just.

#### 2:35 Installing Amazon's official SP-API MCP

Create quick prompt that we want to start the integration, and I will just talk to the AI. Hey, I want to connect to the Amazon SP-API MCP. This is the URL. So go ahead, check how to install it, help me. And you can actually install it here. I already have my credentials in this directory, in those two files. You can check the directory. So you should be actually able to just install it, and we will be able to test it very quickly. So go ahead, do it for me, and let me know once you finish.

All right, so this is pretty much the prompt. It doesn't have to be necessarily like the same prompt, but we will do it in this way. So I just want to show you really that it's super simple. Also, even though I said it's for technical people, you know, it's still very easy with tools like Claude Code. Those are those two, right, the SP-API dev assistant and the SP-API workflow, and this one is more complex, and most likely we don't need it. This is mostly for building much more complex workflows and code pieces for advanced projects.

All right, so it seems like it's working, everything is built, we have access to the MCP servers. Let's try it with the first use case. So this MCP is not only for like working with your data, but it's also for like building more complex tools and automations, literally like creating the code on SP-API. So I will start with a quick check, like what are the rate limits for orders API, and the rate limits are pretty much information how frequently you can do requests to retrieve certain information, in this particular case orders. If you exceed those limits, then you will be blocked and you have to actually wait. So if you actually want to, for example, get a lot of orders over the last few months, you need to create a special script with the algorithm that is actually waiting and only doing the request that is defined in this rate limit. And this is of course for all of the APIs, because otherwise everyone would be just doing tons of requests and Amazon servers wouldn't handle that.

Okay, I see that it actually didn't use the MCP. So let's do it again. I will just use the MCP, because sometimes you actually have to refresh the tool. So let's do it again. Okay, seems like it's not loaded to this session. So let me just restart my Claude Code, and let's try one more time. Okay, now they are loaded. So we should get it very quickly. Seems like something is still wrong with it, but it's actually fixing right away.

#### 5:13 When the install goes wrong, live

Okay, it's actually doing some very weird stuff. Why it's downloading a half gig file? Okay, this one I'm actually doing the first time, and I don't know why it's doing it like that. So sometimes, if you actually just go to your session and say hey install it without the best practices, as I did here, you may get into some troubles like this. I mean still, like, it will figure out the whole thing. And in this particular course, I just wanted to show you, you know, the quickest and the easiest way possible, without going to like the details on Claude Code itself. So, let's see, I hope it will actually figure it out and give us the answer right away. And I just want to also show you, you know, if you're not very technical, that you will end up fighting with the AI tool sometimes to install all of those things, and I want them to be live on the video. All right, so we had to install a few dependencies and restarted our Claude Code, and now let's try it again.

Okay, so we got the answer very quickly. So those are the rate limits for the orders API, and some of the other endpoints here in the orders API. We got it right away. I know it's not very useful for you. I just wanted to show you like that it works and what's the use case for it. Now, if you want to actually work with this MCP and get the data that is interesting for you, we'll.

#### 6:25 Pulling seven days of Amazon sales

Go ahead with this prompt. How much did I sell in the last seven days? Give me totals by day. And it will need to figure out through the MCP which endpoint to ask for a data. Then it will start doing the request, and also it will iterate over different pages, because there's a limit on like what you can get from the API. So those are pretty much those bottlenecks, disadvantages of using those raw MCPs from Amazon and not the data layer, and you will see it later when I show you the shortcut. But still, if you want to build it yourself, use the official way, this is what you have to go through. And for example, in this account I have quite a lot of sales, so this may take some time, and I want you to all see it, so it's now, like, getting the endpoints and figuring out from where it should get what data, and then it's actually executing this.

Okay, so here we are still on the 16 EU marketplaces. So it doesn't know from which marketplace, and I'll just tell it UK marketplace. Okay, so pulling daily order metrics from Amazon, last full 7 days, and we got it here. So this is pretty much, yeah.

#### 7:32 Requesting a Sales and Traffic report, and the wait

Those are the dates, units, average unit price. Okay, nice. Now let's do and request a report. So I'll request a Sales and Traffic report for June, or for the UK also. And this is actually using the Reports API. For the Reports API, there's a little bit more complex logic if you want to do it by your own, because you have to request the report, wait for the report, and pull it once it's ready. Some of the reports you have to actually really wait a lot of time. Some of the reports, they come very quickly, and also depends on the timeline that you want to do it, like one month will be quicker, for example, than three months. So it's using the Reports API, create the report, pull until it's ready, then download the document. So it's a lot of operations, and we also burning the tokens, and we also have to wait.

Okay, so the report is created, now it's pulling. So it's in the queue. This is basically, once you request it, you got to wait. So now we actually got to wait 45 seconds, because we need to wait till it's done. Okay, so the report is done, now we are getting it. So we got a URL. This is also like the gzip-compressed report. So it's actually building a code under the scene that is actually unpacking it and getting us the data. You can see, download, decompress, validate report JSON. Okay, we got it, 4.5 megabytes. Now it's computing the June summary and daily breakdown.

Okay, report is ready. It's taking so much time. All right, we got three files. But finally, we managed to actually get it here. So those are our metrics: order, product sales, unit order, order item sessions, page views, average selling price, conversion, and this is, you know, just one report for one month, just for one account. Imagine doing it every day. And now those are all the data from Seller Central. There's like no ads, because it would be a different MCP, you would have to combine those two together. So I'll just give you and show you this final use case here, right, and the limitations. So for example, what was my profit for June? In order to get your profit you really need the numbers from multiple sources. So all of the information from the Seller Central, from the ads, because you also want to know how much you spend for ads, and then also, for your real profit, you need the information about your cost of goods. This is the actual true profit and loss, and I want to show you here that it will be pretty much very limited.

Yeah, this is what they said. So straight answer, the API can tell you the true profit on its own. I actually want to include the cost of goods from DataDoe.

#### 9:55 What SP-API can and cannot give you

Though already, but let's just do this one for now. I just wanted to show you this table. So from this one MCP you can get orders, Amazon fees, but cost of goods, not possible, ad spend, not possible, also like VAT, shipping, overhead. So basically, there's no way of really getting the information just by using this one MCP.

And even if you were using, like, the second MCP, the ads, you would have to do a lot of request, combine those two sources together, and do a lot of computation. And if you want to do it daily, you can already see how long and not optimized this process would be. So if you really want to go with this option, you can definitely do that. But it will take you weeks of really building a proper infrastructure, a proper automation, like I would say even a platform that is do having a lot of mechanisms in the background, the database. And of course, you can do it all in Claude Code, but I'm just saying, this is just a lot of time, and your time is the most valuable thing. You should be actually scaling on Amazon and running your business, not fighting with the AI to get the numbers.

And trust me, also, those numbers, most likely you wouldn't be happy about them, because what you see in Seller Central, for example, is a little bit different from what you get from their endpoints. And we've been working with the Amazon data for those last years, and constantly fighting with this problem. So this one is actually now doing the work. We can leave it run.

#### 11:15 Official MCP against a data layer, side by side

And I want you to show you the shortcut that you can just get the data right away. Doesn't have to be even Claude Code. I'll just do it here in a pure Claude chat, and I'll just get those information here. So I'll just run: what was my profit from June, Delto UK account? So this one already knows that we are going to use the shortcut, which is DataDoe, that I will tell you more in the next lessons, and it's just directly getting this information, without all of the work that you can see on the right side, when it's also like burning a lot of tokens.

Okay, it's like running in the background, since it's a 60 pages and at the rate limit. And guys, this is not staged. This is actually how it works. And you can already see on the left side, on the other hand, that if you're using the data and action layer for your Amazon account, you just get this information right away, like that, and for all of the cases, right, for all of it, Seller Central information, Ads information, across all of your brands, all of your marketplaces. So we got it here, the breakdown, total sales, cost of goods, Amazon fees, you can see even like a small breakdown, ad spend, and you got the profit here.

So this is just one quick example on building it on your own, at least, you know, trying to build like a quick start, because this is basically how you would start. You would have to install this MCP, install the Amazon Ads MCP, and then start really building those functions to get something useful. But you can see that this is really not the way that you want to work with your Amazon data. And on the left side, you can see a shortcut at DataDoe, which is basically the extension for your AI to work on your Amazon business, to get the information right away, to push it back to Amazon, to analyze the data, run skills, workflows, and everything. So we can keep waiting for this one, but I feel like it will be a waste of time.

So in the next lesson, I'll show you how the shortcut works. We'll create the account. We'll connect the Amazon account. We'll connect DataDoe to Claude, and then we'll have a lot of fun in the course, because we'll be doing amazing things that normally would take a lot of time, a lot of different tools, Excel spreadsheets, most likely even multiple people involved in the organizations, if you're running a serious brand. And we'll be just automating crazy processes, analyzing data, building dashboards, using skills, and also writing stuff to Amazon, so optimizing the listings, changing them, and also working on ads PPC directly from Claude. So let's jump to the next lesson.

---

## Lesson 04: Connect Seller Central and Ads to Claude

Video: https://www.youtube.com/watch?v=Egk196HF4Ew (10:03)

### What it covers

Connecting Amazon Seller Central and Amazon Ads to Claude, start to finish: sign up, authorise, pick your access level, ask the first question.

The whole connection is a few minutes of clicking. No developer registration, no tokens to rotate, no rate limits to work around, none of what lesson 3 went through.

Seller Central and Amazon Ads both get authorised, and the choice that matters comes right after: read-only, or read and write. Read and write is what makes the last two lessons of this course possible, where Claude edits listings and runs PPC.

Then the part nobody mentions. The first sync takes about 24 hours. Your account is connected long before your data is ready, and that is normal. History runs from the day you connect.

At the end, DataDoe goes into Claude as a custom connector and answers the first real question: last month's sales, orders and net profit.

### Chapters

- 0:00 Sign up and connect, without the API route
- 1:56 Connecting your Amazon Seller Central account
- 2:34 Connecting Amazon Ads
- 3:08 Read-only or read and write: which to pick
- 3:25 Why the first sync takes 24 hours
- 4:22 Uploading cost of goods for a real P&L
- 7:48 Adding the connector inside Claude
- 9:07 First question: last month's sales and profit

### Materials

- Sign up: https://app.datadoe.com/register
- MCP connector URL: https://mcp.datadoe.com/mcp/v1
- Claude setup guide: https://www.datadoe.com/hub/docs/datadoe-mcp/claude
- Setup guides for every AI client: https://www.datadoe.com/hub/docs/datadoe-mcp/overview

### Setup steps from this lesson

1. Create an account at https://app.datadoe.com/register. It starts with a 14-day free trial, then one plan at $97/month.
2. Under Accounts, connect Seller Central (or Vendor Central) and Amazon Ads. Pick read and write access to follow lessons 08 and 09. The first data sync takes up to about 24 hours.
3. In Claude, open Customize > Connectors > Add. Name: DataDoe. MCP Server URL: https://mcp.datadoe.com/mcp/v1. Keep the detected authentication, click Add, then Connect and sign in to DataDoe.
4. Test it with: "Show me my Amazon Seller Central sales for the last 7 days."

### Prompts used in this lesson

Prompt 1:

```
Help me connect DataDoe to Claude so you can work with my live Amazon data. Walk me through it one step at a time and wait for me to confirm each step before moving on.

1. If I don't have a DataDoe account yet, send me to https://app.datadoe.com/register to start the 14-day free trial, then have me connect Seller Central (or Vendor Central) and Amazon Ads under Accounts.
2. Add DataDoe to Claude as a custom connector: Customize -> Connectors -> Add. Name: DataDoe. MCP Server URL: https://mcp.datadoe.com/mcp/v1. Keep the detected authentication and OAuth options, click Add, then Connect and sign in to DataDoe.
3. Once it's connected, test it: show me my Amazon Seller Central sales for the last 7 days. If the account was only just connected, remind me the first data sync can take up to 24 hours.

If anything goes wrong, like no Connectors option, a sign-in loop or a "not authorized" error, help me fix it. The full guide is here: https://www.datadoe.com/hub/docs/datadoe-mcp/claude
```

Prompt 2:

```
Hi there, I want to use DataDoe. Tell me which Amazon accounts I have connected.
```

Prompt 3:

```
What were my sales last month?
```

### Transcript

#### 0:00 Sign up and connect, without the API route

All right. So, let's actually connect Amazon accounts to Claude, or to any AI, as a shortcut, pretty much with the simplest way possible, without doing the hard way of getting the data directly from the Amazon APIs, sticking it together, cleaning it, and burning tokens. Basically, it's very simple. We'll sign up, drop the code, connect our account, Seller Central and ads.

Optionally, you'll be able to upload your cost of goods. This is not necessary, but highly recommended. And I'll give you, of course, the quick tour of DataDoe, and I'll show you how to wire it to Claude to start a conversation.

So, first thing you have to do is go to datadoe.com. You can also just check our website. We have tons of nice use cases, so you can actually learn more about the tool. But of course, I will show you the best use cases in this course. You can also go to platform, check what DataDoe is, how it works, what are the use cases. We even have this nice demo that you can actually check on your own.

But yeah, let's not waste the time here. Let's just click get started for free. So I'm going to register here. So make sure to put your email address, strong password with all of the necessary requirements and click sign up. Once you create your account, you have to confirm it on your email account. And as we can see here, DataDoe: verify your email to start your DataDoe trial.

We're going to click here, verify account. Now the account is verified. We can now go ahead, click sign in and just put our credentials. Click sign in. And we are on a first page.

So basically as we land here, we have to activate our free trial. So make sure to read our terms of services and privacy policy. Hit start free trial. Now put the card details. All right. So once you put all of your card details, you're just going to hit start trial. All right. So we successfully created our account,

#### 1:56 Connecting your Amazon Seller Central account

Started our free trial. Now in order for you to continue, the first step is to add your Amazon account. So hit connect your first seller or vendor. I'm going to connect my UK seller and I'm going to log into Amazon. Sign in. Now we need our code, or if you have two-factor authentication, of course, it's a different way.

Hit sign in. Now you need to pick your actual account. Connecting my UK account. Hit select account. Now we need to authorize our application. I'll hit this one. Sign up to DataDoe.

#### 2:34 Connecting Amazon Ads

And we've successfully connected our seller account. And now we will connect our ads account, so we have the whole pair. Connect Amazon ads. All right, we got our match here. I'll just hit continue.

All set, go to accounts. And as you can see here, we successfully connected our Amazon account. We are in the account section. If you have more marketplaces, more accounts, you can go ahead and add them as well. And it's best that you do it right away, because each new account actually takes some time to connect all

#### 3:08 Read-only or read and write: which to pick

Of the initial data. So we have this account here. We have the read-only access. We can change it to read and write, because in this course we will be doing also the write operations on Amazon. So we'll be optimizing our listings, we'll be managing our PPC, all from Claude. And one more thing to

#### 3:25 Why the first sync takes 24 hours

Mention, this is already what you can see here. This is the data sync in progress. We are actually collecting the first data. It takes around 24 hours, maximum 48 hours. You'll get an email that will notify you that your data is ready to use, and from there on you'll be able to actually start using it in your favorite AI tool.

One thing to mention here is you can also upload your cost of goods. However, in order to upload cost of goods, we first need this initial data sync on the account, because DataDoe needs to know what products you actually have on this account, and therefore we'll be able to match it. So I'll show you this a little bit later.

Now if you go back to home, you can already see here that we are still in the onboarding process. We are syncing the data. As I mentioned, you will get an email notifying you that your account is actually ready to use. And now I will show you how your account will look like after this first sync is done. All right, so after the first initial sync is completed, this will be a home screen that you will see.

#### 4:22 Uploading cost of goods for a real P&L

First thing you can do, but this is purely optional, you don't have to do it, is to set up your cost of goods. You can do it for each of the account, and simply head here, download our CSV. We also have a Sellerboard CSV template. So you can actually go ahead and download it, and you will just have to basically delete those and replace them with the actual cost of goods that you have.

Then save it and upload it back here. Click submit, and DataDoe automatically match this information to your account, so you'll be able to retrieve full profit and loss for your account. Again, this is purely optional, you don't have to do it especially right now. I would say just follow the course, build stuff with me together, and then once you feel comfortable, put your cost of goods there.

So, very quick overview of DataDoe, this is basically the first screen you can see, and it's very simple. Really, all you should actually understand now is that in order to manage your accounts you should go to Accounts. You have sellers and vendors. I suggest you to put the read and write access, because we'll be doing the write operations in this course. Then we have the Integrations page, where you can set up DataDoe with your AI, Claude, ChatGPT and much more. We also have other options like REST API, direct Google BigQuery, webhooks, and recurring exports. For now, there is no need to go deeper into this.

We have the Actions, where you can actually see all of your operations that you've done to Amazon. We'll cover this later in the course. Files, if you want to work with your listings and change your images. Again, not needed for now.

Data schema, where you can actually see all of the tables that we have supported in DataDoe. You can actually see the description of each table, all of the columns. And we have Amazon Seller Central, Vendor Central, and Amazon Ads. And we keep adding more and more tables every week.

Then you have the Report view. This is pretty much a quick visualization of your data if you want to see it within the app. Exports, where you can export the data. This is not needed at all for now.

We have new features coming soon. Plugins, that I'm not going to talk about at the moment. I can just tell you that this will be the additional context for your AI, so you can actually give instructions to DataDoe how to properly run your brand, and additional informations like SOPs of your company. Again, we are launching this feature very soon, so I won't spoil too much. Now we have Skills, and also on the Skills I will tell you more in next videos.

We have the whole Settings, where you can actually manage your subscription. You can see your AI tokens. We have a Data page where you have a full option to manage your data connector, meaning you can pick the retention and decide if you want to build a historical database forever, if you want to do a hard cut, or if you want to disable certain tables and not collect them at all.

And we have Actions, where you can actually manage your actions, enable them and disable them, so your AI can actually do certain operations on Amazon. We will cover this module fully in the next lesson, so don't worry right now. You can also add more team members to your account.

And one of the most important pages is actually the Documentation. This is the DataDoe Hub, where you can actually read everything about what DataDoe is. We have full documentation about integrations, access, users, benefits, data fetch periods, data sources, managing data tables, actions, which is the write operations to Amazon that we will cover

#### 7:48 Adding the connector inside Claude

Later, and some tutorials on how to connect with your favorite AI tool. Now let's connect DataDoe to Claude. So all you have to do is go to Integrations, click Claude, open guide, and this is pretty much the full tutorial. Basically, one thing you have to do is copy this URL, and for everyone who is actually using a different AI we have the tutorials also listed here, as I mentioned. But you can also just go to MCP, and basically the principle is that you just need this MCP URL.

It's exactly the same from here. So go ahead, copy this URL. This is the MCP URL for your AI. So head to Claude, go to Customize, Connectors, and just click Add custom connector. Name it DataDoe and paste this URL. This is literally the only thing you have to do.

Click Add. I already have this URL, so I will just go ahead, click here, DataDoe MCP. Click connect. And you just need to log in. And basically you're now connected to DataDoe, meaning you are connected to all of your Amazon accounts and data.

So let's do a very quick test. Hi there, I want to use DataDoe, tell me what accounts do I have connected, and basically I'm asking what Amazon accounts do I have connection to. So we are just checking this first connection, and we can already see we have four

#### 9:07 First question: last month's sales and profit

Accounts connected, meaning our MCP is successfully working. I'll just do a very quick check to show you that it works. What are my sales last month for Delto UK seller? I'll just hit Always, and we got our answer back. So, June 2026, full month, we got the sales, units, orders, net profit, and we got the cost breakdown, meaning it fully works and we are ready to start doing the exciting things and build amazing stuff on top of it.

So that was the setup. Now is the fun part. If you stick till the end, make sure to just wait for this initial sync, 24 hours. You will get the email, and once you have it, you can connect to Claude or any other AI, and you'll be able to follow the next lessons of this course.

---

## Lesson 05: TACOS, ACOS and a full P&L in one chat

Video: https://www.youtube.com/watch?v=1zV_spSQFvk (11:16)

### What it covers

Amazon TACOS, ACOS, stock cover and a full P&L, pulled out of one Claude chat in the time it takes to open Seller Central.

Over 100 tables sit behind Seller Central, Vendor Central and Amazon Ads. This lesson gets at all of them in plain English, with no exports and no spreadsheet in the middle.

Top products by sales with margin against each one. A full profit and loss after cost of goods, Amazon fees and ad spend, charted. Which SKUs lose money. What is running low and how many units Amazon says to ship, by when. Then the ad account: 135 campaigns scored, blended ACOS, and every campaign plotted so the ones above the line are obvious.

The last part is a ready-made Skill. A Weekly Business Review, cloned into Claude and run against a live account, which also flags why last week's profit number is not trustworthy yet.

### Chapters

- 0:00 What you can ask once your Amazon data is connected
- 1:41 Top 10 Amazon products by sales, with margin
- 2:18 A full P&L after COGS, Amazon fees and ad spend
- 3:00 Finding the SKUs that lose money
- 4:26 Stock cover and what Amazon says to ship
- 5:18 Which ad campaigns are losing money
- 7:49 Running a ready-made Skill in Claude
- 9:11 Why last week's profit is not reliable yet

### Materials

- Free ready-made Skills: https://www.datadoe.com/hub/ai-agents-and-skills

### Prompts used in this lesson

Prompt 1:

```
What Amazon accounts do I have connected?
```

Prompt 2:

```
What were my top 10 products by sales last month?
```

Prompt 3:

```
What was my full profit last month after cost of goods, Amazon fees and ad spend? Visualise it.
```

Prompt 4:

```
Which SKUs are losing money right now?
```

Prompt 5:

```
What's running low on stock, what should I reorder, and what does Amazon suggest I ship?
```

Prompt 6:

```
Which of my ad campaigns are losing money, with high spend and low or no return?
```

### Transcript

#### 0:00 What you can ask once your Amazon data is connected

All right, so finally we are all set to do exciting things and really start optimizing your business, saving a lot of time, getting to the data that matters in a matter of seconds with the help of AI and DataDoe. So in this lesson, I will show you what's possible when it comes to getting the Amazon data from Seller Central and from ads, combining it together and finding the information that matters.

First of all, in this course, we just cover the basics. I just want to show you how to start, what's possible, and then I recommend you to go on your own, play with it, check different data sources, check different prompts, combine sales and ads data in one prompt and see what's possible.

So DataDoe has access to more than 100 tables. All of the tables you can find in a data schema in our Hub, so you can really check on your own what's possible. You can also just ask AI and it will guide you through it. In a moment I will show you a demo, and we'll be actually checking information about our business just by using prompts. Additionally, I will also show you ready skills that you can plug to Claude or any AI and run them to go even deeper. So let's jump to the demo.

All right, so let's start using it. Let's just ask it first: what Amazon accounts do I have connected? It's a good start. And obviously, make sure to have your DataDoe MCP connected, as I showed you in the last video. So, we have four accounts connected, that's good. We'll be working on our Delto UK account. Let's start with the first question.

#### 1:41 Top 10 Amazon products by sales, with margin

What were my top 10 products by sales last month, Delto UK seller? And just like this, we got our table of SKUs connected to the ASINs, of course, with the sales, units, profit, and the margin. And we can already see from here that our best seller is actually losing money.

Just so you know, this is also done on purpose, because I uploaded a very large cost of goods for this one, so you can actually see it. Obviously, on the real account, I would optimize it right away. But here I just wanted to show you that you can literally see anomalies like this within seconds just by asking. Now let's check our full

#### 2:18 A full P&L after COGS, Amazon fees and ad spend

Profit, also after last month, after cost of goods, Amazon fees and ad spend. And also I'll just ask it to visualize it. All right, and just like this we got the answer. We have, first of all, here this small table with the revenue, cost of goods, Amazon fees, ad spend, and here we have our full profit, and we can see how much percentage this is of the total revenue.

And here we have the dashboard. So we got the revenue, net profit, net margin, TACOS, we got this nice chart over there, and we got net profit by day as well, so we can actually see each day how much profit we actually did. So this is just a very quick example. Now I

#### 3:00 Finding the SKUs that lose money

Would like to actually check my products that are losing money. So I'll just ask for it as well: which SKUs are actually losing money right now? And just like this we got very useful information, without exporting data from Seller Central, sticking it with Excel files, using different tools.

We literally just got everything here in a chat, and we just know that we have more than 460 SKUs that are actually showing negative profit. That's roughly £8,400. And basically, we also know that one of the SKUs is almost 60% of this. We got here the concentration of SKUs and the loss.

And we can see here the one that really matters. So, we got this one that is losing almost 5,000, basically. Here's the second one with minus 500. And we have some more. And this is something that we actually can start and act immediately, and you should actually do it.

Obviously, no one wants to have the inventory that is losing money, like this particular one, of course, just for the demo, this one is actually fake, because I put extra cost here so we can really visualize it and see it as a nice use case.

But those are definitely the ASINs that we should be working on, especially this one, which sold 132 units last month. And then Claude, of course, gives us a little bit of the recommendations, analysis. We can definitely use it, go deeper, analyze it, and start changing things on our

#### 4:26 Stock cover and what Amazon says to ship

Account. The next use case that I want to show you is about stock. So I want to check what's running on stock now: what should I actually reorder, and what Amazon suggests. So let's check it out.

All right, so we got the answer. We got the SKU selling 25 more units in the last 30 days, those are the numbers from my FBA. And we can actually see here the SKU, the product on hand, sold, days left, and Amazon says ship.

So we should actually, yeah, here ship 38 of this product by 26th of July, but of course this is already overdue. Here we have some other numbers: so this one, for example, we should actually ship 32 items today, here 25 by 7 August, and much more. So basically you can also ask about your stock and get recommendations, so you actually know what to do next.

#### 5:18 Which ad campaigns are losing money

Now I also want to show you some use cases with the advertisement data. So I'll just ask which of my ad campaigns are losing money, so they have high spend and low or no return. All right, and we got the answer.

So, first of all, short answer: your ad account is broadly healthy across 135 campaigns. In July you spent almost 20k, got 90k in attributed sales, blended ACOS of almost 20.5%. There is no single campaign that is pretty much hemorrhaging money, so there is no single campaign that is actually really losing us a lot of money.

And we got every campaign plotted against that line, so we can actually see the breakdown. Those are above the breakdown, we can actually see, and here we have the 12 campaigns above the line, so those are the ones in red, and we can actually see the campaign name, its type, spent, sales, and ACOS.

So literally we just got this whole summary, like the actual analysis of our ads, just by asking the AI, and it got the answer, and it's really good. So I highly recommend you guys to run those prompts also with your own accounts, because I'm very curious to see the results, and basically you can keep asking it more and more questions. I will basically leave it all to you. So it's all about really your business question, something that you always wanted to check and ask. You can actually do it right now.

I wanted to show you the skills, because overall those are ready for you to use instructions for the AI that you can actually reuse. You can also create your own skills. Those are pretty much the instructions for the AI, so you don't have to always create those advanced prompts, but you can actually put them as a package, basically a skill, and reuse them. And later in this course I will show you basically how to run scheduled functions. So, for example, you'll be able to basically get the reports on a daily or a weekly basis without writing those prompts over and over and asking manually for it, because the AI will basically execute it periodically.

And we have a lot of them that are pre-built. So I'll just show you here, in DataDoe we have to go here, go to Skills, and we can actually check our library. We are adding more and more skills. We have the skills that are read only, we have the skills that are also write. So, some of these skills might actually change things on your Amazon accounts, such as optimizing your bids, creating campaigns, changing your titles in bulk.

Some of these skills may even create dashboards. Some of the skills might even execute external functions to send emails or notifications. I mean, there are just endless possibilities. We will be also

#### 7:49 Running a ready-made Skill in Claude

Building our own skill later. I just wanted to show you very quickly how easy it is to really use one of them. We can actually go and search for weekly business review, and we're going to use it. So, weekly business review: generate a weekly business review, profit, margin, ad efficiency, TACOS, inventory. It's comparing the week to your trailing four-week norm, and explaining margin, fee, anomaly. And it also ends with the proper actions.

So let's actually do it. So I'll just add it to Claude and use it. I'll just copy it, and I'll just paste it inside this chat. So I'll just say Delto UK seller account, and you can actually also check here the action plan. So it will go in the repository with all of the skills, find the skill files, use the skill creator to create that skill.

All right, so the skill is cloned. And now, in order to actually run it, we need to save this skill. Okay, and now let's run it. So, weekly business review, Delto UK seller account, and let's run it. Here, for your view, there's a little bit more of the description, we don't need to go through all of it, I will leave it to all of you. You can also go ahead and check all of the other skills. This will be running right now, I'll just close it.

#### 9:11 Why last week's profit is not reliable yet

And just like this we actually got the answer back very quickly. Here's the warning that profit is not reliable yet, because of the settlement timing artifact. So basically some of the Amazon settlements, they have a delay in time. So when it comes to checking the data, usually for like last week, last two weeks, it overly will be always not accurate fully. So you should always check a little bit higher period of times, but still it got us this data.

It got us the weekly insight. We can see our sales, profit, margin, ad spend, units, as percentage of sales. So we can see here, and ads are flat every week, only fees move. This is the signature of the settlement matching, not the cost change.

And you can actually just see here that this is still not matching, but you can actually see the trend overall, how this looks over time. We have May, June 7, June 14, 21, so we have different weeks. We have the watch out, so we get the stockout alert, it happens, and we see that this product went from 100% of sales last week to zero this week.

And we got some more recommendations, of course, each recommendation, and the chart and numbers will be different, because you're running different seller accounts. Here we have also, like, okay, what should we do this week? We see our top movers. And this is just one of the examples of the skills that you can actually reuse from our repository, or of course you can build your own skills and also reuse them, and it basically works for all of the data that we have available for you.

So in this lesson, we really went through and pretty much started prompting with the AI, getting the answers, combining different data sources. We even used the skill. In the next lesson, I will show you how you can build a dashboard with multiple metrics from different data sources, so you can actually have a helicopter view on your business.

And if you are actually using some apps already, some different tools with the dashboards, after the next one, you will be able to actually build it yourself and configure it the way you want, just by talking with the AI. And we'll also save it as a skill. So, let's jump to the next one.

---

## Lesson 06: Have Claude build your command center

Video: https://www.youtube.com/watch?v=YCPnZU7mpqg (10:00)

### What it covers

Three tabs, fifteen tiles and ten Amazon data sources on one page, and Claude wrote all of it from a single prompt in about twenty minutes.

You do not build this dashboard. You describe it, and it gets built out of your live account while you watch. No coding, and the whole thing is read-only.

Money: revenue, net profit, margin, TACOS and ROI against last month, a P&L waterfall, where the revenue actually goes, and a ticker of the last two days of orders. Products and ads: best and worst SKUs by net profit, spend split by campaign type, and a watch-list of search terms running hot on ACOS. Growth and ops: account health with the policy violations that need attention, reimbursements owed broken out by currency, sales by buyer city, the search-to-purchase funnel, returns by reason, and recent buyer feedback.

Then it gets saved as a Skill, so rebuilding it next month is one sentence.

### Chapters

- 0:00 What a seller command center actually contains
- 1:50 The prompt, and how it was built by iterating
- 4:08 Telling it what to leave out
- 4:47 Asking for QA and recommendations that are not generic
- 5:33 The money tab: P&L waterfall and live orders
- 6:14 Products and ads: best and worst SKUs, hot ACOS terms
- 6:48 Growth and ops: account health, reimbursements, returns
- 9:04 Rebuilding it next month in one sentence

### Materials

- Free ready-made Skills: https://www.datadoe.com/hub/ai-agents-and-skills

### Prompts used in this lesson

Prompt 1:

```
<context>
I'm an Amazon seller. My account is "[YOUR ACCOUNT NAME]", connected through the DataDoe MCP. I want you to build me one live dashboard from my real data, not a mockup, not placeholder numbers. Pull everything below directly through the DataDoe MCP tools for that account.
</context>

<task>
Build a single, self-contained HTML dashboard called a "Seller Command Center" with three clickable tabs: Money, Products & Ads, and Growth & Ops. Use [LAST FULL MONTH] as the reporting period for anything money, ads or product related, unless a section below says otherwise.
</task>

<tab name="Money">
- Top KPI row: revenue, net profit, margin %, TACOS, and ROI. Next to each KPI, show the percentage change versus [THE MONTH BEFORE] as a small delta, if that comparison is available.
- A P&L waterfall: revenue -> COGS -> Amazon fees -> ad spend -> net profit.
- A "where my money actually goes" breakdown, as percentages of revenue: COGS share, Amazon's combined fee share (FBA + referral/selling), ad spend share, and net profit share.
- A daily sales and profit trend line across the month.
- A live ticker of my last 2 days of orders: product name, quantity, price, and buyer city. Normalize city casing so "LONDON" and "London" don't show as two separate rows.
</tab>

<tab name="Products & Ads">
- Top 8 SKUs by net profit (product name + units sold).
- Worst 5 SKUs by net profit (the loss-makers). For the single worst SKU, also break out its COGS, Amazon fees, and ad spend so I can see exactly why it's losing money.
- Ad spend and ad sales split by campaign type (Sponsored Products / Sponsored Brands / Sponsored Display). If one type is the large majority of spend, state that plainly rather than presenting it as an even three-way split.
- Top 10 campaigns by spend, with spend and attributed sales each.
- A keyword watch-list: search terms with meaningful spend, sorted by ACOS, flagging anything over 30% ACOS.
</tab>

<tab name="Growth & Ops">
- Current seller account health: account health rating, order defect rate, late shipment rate, and policy violation count. This is a live snapshot, so use the most recent available date, not a monthly average.
- Reimbursements owed to me, by reason, for the last 90 days. Report totals BY CURRENCY separately, never sum different currencies into one figure.
- Sales by buyer city, ranked, with city casing normalized.
- Search-to-purchase funnel: impressions -> clicks -> cart adds -> purchases. Label it "search-attributed traffic," not all traffic.
- Returns for the last 90 days, grouped by return reason, with count and refunded amount.
- My 10 most recent pieces of buyer feedback (1-3 star): date, rating, comment. If two or more comments describe the same specific problem, call that pattern out explicitly rather than listing them as unrelated.
</tab>

<design>
This needs to look like a real, modern SaaS product, the kind of dashboard a company like Stripe, Linear or Vercel would ship. Not a plain data table, and not AI-generated-looking slop.
- Clear visual hierarchy: big bold numbers for the headline KPIs, small muted uppercase labels, generous whitespace, no cramped tiles.
- Use actual charts where the data calls for it: a real line chart for the daily trend, a real waterfall or stacked bar for the P&L breakdown, a real bar or donut for the ad spend split. Text stats are fine for single figures, but trends and breakdowns should be visualized.
- One consistent color system throughout: a single accent color for neutral and positive data, red only for losses and flagged warnings. Don't scatter random colors.
- Consistent type scale and spacing across all three tabs.
- Dark, modern theme by default.
- If a stacked bar or donut has a segment under about 3% of the total, still give it a small minimum visible width so it doesn't disappear, keep its real percentage in the label, and add a small caption noting that tiny segments are drawn wider than their true share.
- Any acronym or piece of seller jargon (TACOS, ACOS, ROI, COGS, ODR and similar) should get a plain-English explanation on hover the first time it appears. Be consistent about it.
- When you describe why something is losing money, state it as precisely as the numbers show it. Don't round a real finding up into a more dramatic one.
</design>

<output_format>
One self-contained HTML page, three real clickable tabs, not three stacked scroll sections. Every number must be a real figure you pulled, clearly labeled with its period. Use red or warning styling for loss-makers and anything flagged. No lorem ipsum and no "TBD" tiles. If a tile truly has no data, show a plain "no data this period" state instead of inventing one.

Before you consider this done, render the HTML yourself and check it: no tile content clipped or overflowing, no text cut off, tables and charts fitting their containers at a normal laptop width, and all three tabs switching correctly. Fix anything that doesn't fit before showing me. If you have no way to render it, say so explicitly rather than claiming you checked, and review the HTML and CSS directly for the same failure modes.

Any total or subtotal derived from other numbers on the page must be computed from the same underlying data, never hand-typed as a separate figure.
</output_format>

<final_step>
After the dashboard is built, read it back in full and act as my analyst: tell me the 3 things I should fix first, ranked by financial impact. Be specific. Name the SKU, the number, and the reason. No generic advice.

Then save this exact build, the tabs, panels, sources and periods above, as a Skill, so I can rebuild the same dashboard again later with one sentence.
</final_step>
```

### Transcript

#### 0:00 What a seller command center actually contains

All right, so in this lesson we will be building a dashboard using DataDoe MCP and pulling data from multiple sources, so you can actually have a whole overview of your business. And I'll share this prompt with you, so you'll be able to save it as a skill, change it, modify it so it fits your needs.

This is just the beginning of what's truly possible with tools like Claude Code, that allow you to build dashboards and tools to better run your Amazon business. So let's dive into it. This will be read only, as we'll cover the write operations in the next lessons. So let's start.

So, basically, how it works for all of you, if you're trying Claude Code for the first time: you can just describe what you want, and Claude will build it. And by build it I mean it will literally write code for you, and you don't need any coding experience for this. You literally just need to have a tool like DataDoe, which has the MCP or API, so you can easily get all of your Amazon data and just tell it what you actually want to achieve.

So we'll build a nice dashboard with three different tabs. Those will be like different categories, with multiple data sources, a beautiful design, and recommendations and actions, so you will know what to do to scale your Amazon business. And I'll also show you how you can keep it as a skill, so you can actually reuse it and refresh it.

So we'll have three tabs: everything about money, products and ads, and growth and ops. So this will be a start of having your own command center that has access to all of your data, that you can keep improving and adding more and more over time to it. So let me show you how to do it. So for this one I've actually

#### 1:50 The prompt, and how it was built by iterating

Prepared this long prompt that we will use just in a moment. I first want to tell you something about that. So, basically, how I created this: I started talking with the AI. I told it that I want to build a dashboard with different data sources, different tabs, using DataDoe. I want to check the data mostly for June, but basically I also want to have it as a skill, so I can reuse it and I can run it every month to update it automatically.

And basically AI gave me the prompt. I tested it. Then I improved that, I changed a couple of things, I got a new prompt, I reran it. And this is how I got to the perfection. And this is still a very basic build that you can just take, reuse, and keep improving over time.

So, basically, as a context, I said: I am the Amazon seller, my account is Delto UK, I'm using DataDoe MCP, and I want to build a live dashboard with my real data. So it's not a mockup, and there won't be any placeholder numbers, those will be all true numbers pulled from Amazon.

And the task is to build a single self-contained HTML dashboard called the seller command center, with three tabs: money, products and ads, and growth and ops. And also, I want to use June 2026 for my reporting period. In the money table I will have different KPIs, like revenue, net profit, margin, TACOS, ROI for June. Next to each KPI I want to see the percentage versus May, small delta, if the comparison is available, because you may not have this data, for example, so it's also important to mention this.

I want to have a full profit and loss waterfall, which is revenue, cost of goods, Amazon fees, ad spend, net profit, also like the breakdown where my money goes, a daily sales and profit trend, a live ticker of my last two days of orders.

Here, the next one would be products and ads. So I want to see my top eight SKUs by net profit in June, worst five SKUs by net profit in June, ad spend and sales split by campaign type, top 10 campaigns by spend in June. I'm not reading those details, you will have access to this prompt.

Then growth and ops: I want to see the seller account health, reimbursements owed to me, sales by buyer city, search-to-purchase funnel, returns, and my top recent pieces of buyer feedback. Here also you can say what you don't want. For example, so for

#### 4:08 Telling it what to leave out

Example, in this dashboard, I don't want information about inventory, stock levels, restock recommendations, also share search panel. And I'm showing you this just as an example, you may want to have some exclusions, you may don't want to, it's up to you purely.

And here, for the design, basically I'm not going to read it all. I just started having the conversation with the AI, that I want to have a modern dashboard with the breakdowns, with different KPIs, with tables, and this is what I got back. So, as output, I want to have this HTML page. I want to have those tabs where I can click through. I don't want to have any lorem ipsum if something is not available. So I

#### 4:47 Asking for QA and recommendations that are not generic

Also asked it to do a quality assurance on this one, so like verify if it renders properly, if something is wrong, fix it. And as a final step, I also want recommendations from the AI, basically, and those cannot be generic advice. I told it that I'm the Amazon seller, I want the best Amazon practices, and also DataDoe MCP has it in it already.

And after the build is done, I also want to save it as a skill, so I can rebuild it on demand. So this is pretty much the whole prompt. What I will do is I will just copy it, and I will paste it here, and let it run.

Now this may take around 20 minutes, 25 minutes, so I will just show you how to build the whole thing. So after you build it, you

#### 5:33 The money tab: P&L waterfall and live orders

Can end up with having something like this. It may differ, it may look a little bit different, but most likely it will be very similar. So there's light and dark mode, and you can already see the money table with revenue, net profit, margin, TACOS, ROI.

We see the waterfall here, so that's revenue, cost of goods, Amazon fees, ad spend, net profit, we see where the revenue actually goes, and you can actually see it over there. You can see the daily revenue and net profit for June. You can even improve it to say, I wanted to have it in one place.

We see the live order ticker. Here we see the product, the quantity, the value, and where it was ordered from. Now, the product and ads.

#### 6:14 Products and ads: best and worst SKUs, hot ACOS terms

So we actually can see top eight SKUs by net profit, so the product, units, net profit, worst five SKUs by net profit. We can already see this one that is losing a lot of money, and there's already recommendations from AI on this one.

We can see the ad spend by campaign type. So I'm mostly running sponsored products campaigns. Here are 10 campaigns by the ad spend. Here you can see keyword watch list, so for example, for those search terms we have very high ACOS, so most likely we should actually do something with them.

#### 6:48 Growth and ops: account health, reimbursements, returns

And growth and ops. So we have the account health, this is pretty much a very healthy account. Order defect rate, late shipment rate, listings policy violations. We see six of them that need attention, so we can actually go in the AI and start the conversation about that, to get more information about them.

Reimbursements, we can see here. So, customer return, those are in pounds, those are in euros, those in PLN. We can see here the breakdown: so customer return, lost in warehouse, damaged warehouse, and so on.

We can see sales by buyer city. We can already see that most of the orders actually are from London. Here, search-attributed traffic funnel: so we see search impressions, clicks, cart adds, purchases, and the returns. We can also see them here, how many returns, for example, we got with, like, no reason given, or unwanted item, mis-order, and so on.

And here we can also see our buyer feedback. So we see the review, opinion, and everything. We see that AI actually got, like, a repeated pattern for the same specific problem, so we can already see how we can improve our Amazon business with it. So this is pretty much what you can just build, or start building. This is one of the examples.

I'll show you something that I'm working on also, this is actually for the write access. This is a tool which basically manages your PPC. You can see the estimated monthly waste. You can tell it what to do, and you can actually run it, and it will, for example, pause everything over 40% ACOS, total ad spend, blended ACOS flagged campaigns.

We can see here the campaigns that we can actually pause, or we can cut the bids just by clicking here. We can see the Amazon AI recommendations, we can apply those suggestions already. We can see search terms which we can block, and the placements, and we can also shift budget just with the click.

So for this one, I don't have the prompt yet to give you, I'm still working on it. But I just wanted to show you, in this particular lesson, what are the possibilities, what you can build with it. Even though in this lesson I just wanted to show you a read access, this one is already for the write access, because you can start clicking in the software and apply the changes to Amazon, in this case running your PPC from there. So that's basically what you can really do, and

#### 9:04 Rebuilding it next month in one sentence

What you can really build with DataDoe and a tool like Claude Code. Now, I just opened here another conversation, I just wanted to show you, and I just asked it: hey, how to run this skill. And because we already have this saved as a skill after running this prompt, we can then just ask in plain language, like, build my seller command center for July, or rebuild the command center for Delto UK, June.

We can change the seller name even, we can say refresh the dashboard, and this is basically what you can achieve just by creating the skills with those prompts, and what you can build on top of DataDoe MCP.

This would be it for this lesson, and in the next one I will show you how you can put skills or commands on a schedule and run them daily, weekly, on autopilot, so you won't have to always type things manually.

---

## Lesson 07: Put the morning check on a schedule

Video: https://www.youtube.com/watch?v=3CXLvqnPE2Q (13:25)

### What it covers

Amazon account health, stock and ACOS checked every morning by Claude, with one rule: stay quiet unless something actually needs you.

Most sellers open Seller Central every morning to check nothing broke overnight. This hands that job over, using Claude's own scheduled routines.

Two get built. A Monday briefing with last week's sales, units, ad spend, TACOS and net profit, which marks profit as preliminary when Amazon fees have not settled yet. And a daily check that stays silent unless a SKU is running low, a campaign's ACOS jumps past its normal range, account health moves, or returns spike on one product.

The daily one earns its place on camera: six SKUs at or near zero stock with live sales, seven campaigns through 35% ACOS, and product safety complaints that sat flat at 9 for two weeks then hit 17 in two days.

There is no alarm watching your account in real time. Every heads-up is a scheduled check that ran and found something.

### Chapters

- 0:00 Two routines: a Monday briefing and a daily check
- 2:26 The rule that stops it inventing numbers
- 4:10 Writing the Monday briefing prompt
- 5:34 How Claude turns a prompt into a schedule
- 6:58 Reading the first briefing, and why profit is preliminary
- 9:19 The daily check that stays quiet unless something breaks
- 11:06 Account health: complaints that tripled in two days
- 12:13 Local routines against ones that run without your laptop

### Prompts used in this lesson

Prompt 1:

```
Every Monday at 8am, give me a business briefing for the past week, pulling from my sales, ad, inventory and account data for my [YOUR ACCOUNT NAME] seller account:

1. Headline: total sales, units, ad spend, TACOS and net profit for the week. If Amazon's fees haven't fully settled yet, say so and mark net profit as preliminary rather than final.
2. My 3 best and 3 worst SKUs by net profit this week - for the worst one, say whether it's COGS, fees or ad spend dragging it down.
3. Any PPC campaign whose ACOS jumped week over week, or is now over 35%.
4. Any SKU at real risk of stocking out in the next 2 weeks, and how much to ship in.
5. Anything Amazon owes me in reimbursements, and any new pattern in returns or buyer feedback worth knowing about.

Put "What needs my attention" at the top, ranked by financial impact, then the full numbers below. Pull every figure from my real account data - never estimate. If a section has nothing to report, say so in one line instead of skipping it silently.
```

Prompt 2:

```
For Amazon [YOUR ACCOUNT NAME] Seller.
Every morning do a quick health check across my account and only message me if something needs attention - otherwise stay quiet:
Use DataDoe MCP.

- Any SKU running low on stock against its recent sales pace.
- Any PPC campaign whose ACOS jumped out of its normal range, or crossed 35%.
- Any account health change - a new policy violation, a rising defect rate, or a health score drop.
- Any sudden spike in returns for a single product.

If any of these trigger, tell me exactly what changed and the number behind it - not just "something's off." Pull everything from real account data, and if nothing's wrong, don't send anything.
```

Prompt 3:

```
Set this up to run every morning automatically
```

### Transcript

#### 0:00 Two routines: a Monday briefing and a daily check

All right, so let me show you how you can run your reports automatically by using Claude's native functions called routines. We'll pretty much set up two routines: one will be running weekly and the other one daily. So what we will cover here — I'll show you how you can create the every Monday 8 a.m. summary of your business, and also every morning a quick check of what needs your attention so you can always react. I will show you what's possible just to really get you started, and you can go ahead and create more advanced prompts, use some of the skills from our library, or create totally your own.

Quick reminder about the skills that we have in DataDoe: if you go to datadoe.com/hub, you can go into Skills and Agents. We have a growing library of skills, and we are adding more and more over time as we talk to more users and check their needs. So you can definitely pick some of those. Most of them are basically read-only skills that will report to you, give you recommendations, suggestions. But there are also some which basically do something on Amazon — for example, PPC bid optimizer or PPC negative keyword applier. So definitely go to Skills and Agents in the hub and check them on your own, and I'm sure you will find something very useful for your business.

Now, how do you set this up? I will show you a demo in just a moment. Basically, you only have to set this up once and it will run on its own, and you can also tell it how frequently you want it to run — once the run is finished, it will report back to you. Some of the good practices here are, I would say, very similar to overall AI practices when it comes to prompting or creating skills: you need to be very specific, and you need to work on the format. Usually a good routine requires a couple of iterations in plain chat, so you can get exactly what you need. You can start the conversation, then tell it that you want to create the routine, that you want certain data points, and you can also ask the AI to give you the prompt so you can actually turn it into the routine. Always cite the data — so basically always check that the data source exists.

#### 2:26 The rule that stops it inventing numbers

In DataDoe, those numbers cannot be created just by the AI, because it can hallucinate and give you something that you don't want as a recommendation for your business. My recommendation is to keep it read-only, especially at the beginning. You can also create skills that will do something on Amazon — you can actually create skills that monitor a couple of products, checking the prices, checking your competitors, and then this routine can actually reprice your listings so you can win the buy box. And of course, you can do a lot of PPC tools with them as well. Like here, there are no limits — this is very powerful, it runs on its own.

But just my recommendation: start with the read-only skills, get comfortable with them. Then, when it comes to write access, I will show you how it works in the next lessons. And once you're very comfortable and you've actually used write access quite a few times, this is where you can start turning this into a skill and also into a routine. So let me show you a demo.

So basically, if you want to get to the routines, you can go to normal Claude or Claude Code — I use Claude Code because it works just the best for my use cases, and it can really give you a lot of details, and it can use a lot of processing power and tokens. Basically, for the routines, there's no need for token optimization, because they can literally run when you are asleep, so you don't really get into those token windows when you are using Claude — also a very nice way to optimize your token usage. So all you have to do is go to Routines. I already have one routine here — you can actually start by basically explaining what you want to achieve right here, but you don't have to. You can actually also go and open a new chat and start a plain conversation here, saying that you want to create the routine. So let me show you the first one. So, for my weekly

#### 4:10 Writing the Monday briefing prompt

briefing, I will say: "Every Monday at 8:00 a.m., give me a business briefing for the past week, pulling from my sales, ads, inventory, and account data for my Delto UK seller account." Of course, in your case, it will be your seller account. One thing to mention here is that if you connected multiple Amazon accounts to DataDoe, it will also work — so you can literally do weekly briefings for multiple accounts just in one execution, and you can even compare the numbers between those different accounts. Something that was very hard to achieve before tools like DataDoe.

So what I want — I want the headline: total sales, units, ad spend, TACOS, and net profit for the week. And if Amazon fees haven't fully settled yet, say so, and mark net profit as preliminary rather than final. I want my three best and three worst SKUs by net profit this week — for the worst one, say whether it's cost of goods, fees, or ad spend dragging it down — and the PPC campaigns whose ACOS jumped week over week or is now over 35%, and any SKU at risk of stocking out in the next two weeks, how much to ship, anything Amazon owes me in reimbursements, and any new pattern in returns or buyer feedback worth knowing about.

Also put what needs my attention at the top, ranked by financial impact, and the full numbers below. Pull every figure from real seller data — never estimate. This is very important. If a section has nothing to report, say so in one line instead of skipping it silently. So what I will do is I will just apply this. And we got

#### 5:34 How Claude turns a prompt into a schedule

this history. So Claude knows that it needs to set up the recurring weekly schedule task, and it's actually loading its own skill — it's a skill called Schedule, every Claude has it. Then it's checking the DataDoe MCP. I also have SP-API tools, but it will not use them here — I already have them in my session since we've been connecting to the SP-API MCP in this course already. So it's checking DataDoe and checking all of the data points here. You can see that it actually used DataDoe for it, and different export sources — so it got the account and it got all of the tables it needed.

Then it actually set this up to run every Monday, pulling the live data for both Seller Central and ads, which are active in this connection in DataDoe, and then it will give us the headline, the best and worst SKUs, PPC, stockout risk, reimbursements, returns, feedback. So it actually did that, and I triggered it manually — I recommend you do the same, just to verify this, to see if you're happy with the output or if you want to keep improving it.

I said run it now here, and it started working. So we can already see it used DataDoe, got a lot of data points here. We can even see more, and here is this quick start — what needs my attention, I will not go through it here. Most of it will be blurred; I don't want to share my client's data. But what we can see is

#### 6:58 Reading the first briefing, and why profit is preliminary

that, for example, this week's profit is preliminary because we don't have all of the Amazon fees yet, so it might be better to run this every month. We got our worst SKUs, and we already see that we have a cost of goods problem, not fees or ads, because cost of goods ate 97% of the sales. Of course, this is done on purpose — I actually uploaded this value, so we have a nice use case for our course.

We see here that for this one, ACOS blew from 5.2% to 30% this week, so the spend nearly tripled while its sales halved — so this one is actually a problem. We see three SKUs at zero stock with live sales velocity right now. We see returns for "not compatible" more than doubled this week. And we see also the reimbursements.

Here we have the headline, so we see information for this week: total sales, units sold, ad spend, TACOS, cost of goods, fees posted, net profit as calculated — and this is marked preliminary. And here we see our best and worst SKUs by profit. We see PPC campaigns with rising or high ACOS, and we see the campaign ACOS this week, ACOS last week, and the delta, stockout risk. So we can actually see what we should ship and when. We see the reimbursements, returns and feedback — here are reimbursements, here are returns, and this is the feedback.

So this is one of the routines that you can actually create, run, test and optimize, so you get all of the information you care about in the format you want, because you can also have it as a widget, or as an HTML dashboard that is being automatically refreshed — it doesn't have to be plain text here. This can also be a routine that is connected with multiple MCPs, so you can, for example, send an email to your suppliers or a message on Slack or Microsoft Teams to inform your teammates what they should work on. And there are more and more use cases that you can actually do with it.

So that was this first schedule prompt that is running weekly. Now, if you go to Routines, you can already see that we have this Delto UK weekly briefing that is pretty much live, and it repeats every Monday at 8:00 a.m., and it has all of the instructions over there. You can click Run Now, and you can modify it fully so it fits all of your needs. Now, for a daily check, this is the second prompt

#### 9:19 The daily check that stays quiet unless something breaks

that I've prepared for you. So: for Amazon Delto UK seller, every morning do a quick health check across my account and only message me if something needs attention — otherwise, stay quiet. Use DataDoe MCP. And basically, here are the conditions: any SKU running low on stock against its recent sales pace; any PPC campaign whose ACOS jumped out of its normal range or crosses 35%; an account health change, a new policy violation, a rising defect rate or health score drop; and a sudden spike in returns for a single product.

And if any of these trigger, tell me exactly what changed and the number behind it, not just that something changed. And of course, pull everything from real account data. And if nothing's wrong, don't send anything.

And basically, we've created this one as well. And here, for example, that was the already existing conversation, and I just pasted it here, and it didn't create the routine. So if something like this happens, you actually have to say it specifically — that we actually got the recommendation here: set this up to run every morning automatically. So if we just go ahead, it will run automatically.

For now, we just got the actual output for this, so we can also check it. We can see that it connected to DataDoe, it found the account, it confirmed it has all of the data sources, and then it got the data. It even saw the exports are large, so it created a special script to load it, and then gave us the answer.

So we actually see the stock: six actively selling SKUs at or near zero — those are the SKUs that you actually need to get inventory for, so basically for those SKUs you need to send them to Amazon. Here we got the seven campaigns whose ACOS blew through 35% — those are the campaigns that require your attention right away. You can pause them, optimize them, change them. Pretty much this is something you should start working on right now — otherwise, you will burn a lot

#### 11:06 Account health: complaints that tripled in two days

of money. We got the account health, and we can see product safety complaints tripled in two days. We can see that we got the seller product safety customer complaints, and it was all good, but then it jumped to 17 — so basically it was flat at 9 for two weeks, then it jumped to 17. Everything else is actually great. We got a sudden spike in returns on two SKUs, and I will just fill it in.

So, set this up to run every morning automatically. And what it will do is it will also translate our prompt into the actual routine, so it will actually give more context to it as well. Here I will say 9:00 a.m. — recommended. And here you can also set up how it can notify you when something triggers, and you can basically get a push notification from the app, Slack, or email.

And you can also do something on your own — I will just say push notification. Okay, so we got it set up. We will receive it at 9:00 a.m. every day through push notification, and this is set up as Delto UK Morning Health Check. And also there are some notes: so if nothing is there to worry about, it will stay silent, and it will only notify us if

#### 12:13 Local routines against ones that run without your laptop

something needs our attention. And this one is the local routine, so it will only run while the app is open, meaning the Claude app on your computer. But you can also set this up so it runs in the cloud automatically and does it without the requirement of having your computer open.

And here there's a recommendation that we should also hit Run Now once, just to approve certain permissions for the DataDoe MCP, if we haven't approved them for the whole organization or for your whole account on Claude. And you can already see here we got Delto UK Morning Health Check, and we got the instructions, we got the setup, and you can run it now also to check.

So that was basically it when it comes to creating the routines, so you can actually run things on autopilot and don't really work directly with the AI, only by prompting with it. And you can set up a lot of different routines, and trust me, once you do those two, you'll end up building more and more. In the next lesson, I will show you the last piece of this whole equation, which is write access — so you'll be able to run your Amazon business fully with the help of AI.