Explain the Modern Data Analyst Role

0:00Hello and welcome back to the next skill in the CompTIA Data Plus prep course. In the previous

0:06skill, we built the roadmap for the exam, how it's structured, how it's scored, and how to study

0:12smart using science-backed techniques. In skill two, we'll zoom in on the person this certificate

0:19is really built around, the modern data analyst. So what does that really mean in 2026? What about

0:26AI? We've got a lot to unpack in this skill where we explain the modern data analytics roles. We're

0:34living in what many call the golden age of analytics. Organizations are literally swimming

0:40in data, storage is cheap, and competing power is everywhere. And it's also rentable in the cloud,

0:48which means leadership knows there's real money hiding inside of those numbers. But raw tables

0:54don't create decisions. Raw data by itself does not have value. And that's precisely where the

1:01translation layer comes in, turning messy data into usable insight. That's the analyst's job.

1:07So you might be asking, well, why does this matter for the CompTIA Data Plus prep? Here's why. Data

1:14Plus' five domains aren't random exam buckets. And what makes this really easy to approach

1:21educationally is that all those domains are more or less the steps for data analysis. And that's what

1:29we're going to leverage in this skill, because these five domains are basically a job description

1:35broken into parts. So now going back to this raw data that needs to be converted. First, you need

1:42a development environment. And that's the first domain, understand data environments. Step two is

1:48where you acquire and prepare data. So now we have step one, which is a development environment.

1:54Step two, we have the raw data, and now we're going to prepare it. And step three, which is

2:00analyze it. And you could also do some quick EDA here as well. Now you're starting to turn that raw

2:06data into some kind of value. Then number four, you're going to visualize this. This is super

2:12important, because not everybody understands Python, but they sure do understand visualization.

2:19And that's why it's a universal language, if done right. And then finally, we need to lock this down

2:25with governance, which is now an ever growing field, thanks to AI. And now we have something

2:31called AI governance. It's not just data governance. Now it's AI governance. These are intertwined

2:37because in AI governance, you need to understand how that data is being used to train. So data

2:44governance and AI governance overlap. Super interesting. So what does this all mean? By the

2:51end of this skill, you'll be able to explain what a modern data analyst does end to end, why that

2:58role is growing super fast, even with AI, and how data plus domains map directly to real analyst work.

3:07How does that translate into what we're going to actually cover? Let's break it down. In this skill,

3:13we're building a clear picture of modern data analysts from two angles. First, why the role

3:20exists at this scale right now. And this is really about the golden age of analytics,

3:27forcing analytics to the central part of a business. And then B, we'll explore what analysts

3:34actually do day by day. This is the actual workflow, which includes the techniques and

3:40the tools. We'll walk through the temple mental model. This is a visualization of analytics on

3:48top for three pillars, which includes data storage and compute. And here in the roof,

3:55we can call this analytics. And here we have governance. Then we'll walk the analytics process

4:02end to end and connected to reporting and dashboards. And finally, we'll introduce the

4:08techniques such as descriptive, inferential, predictive, and prescriptive. And this is also

4:15where EDA fits into the picture. Finally, we'll connect modern AI flavored work. So that is the AI

4:23machine learning and deep learning, which includes generative AI, NLP, which stands for natural

4:30language processing. And we'll also cover RPA, which is not AI, but it is part of automation

4:38and people get them confused all the time. So we're going to make sure that these are clearly

4:43delineated and that we understand that they're different for this skill. The second skill,

4:48we're mostly focusing on domain 1.0 data concepts and environments, 3.0 data analysis with a little

4:57bit of overlap into the second, fourth, and fifth domains. But generally we're working with domain

5:03one and three. All right. So that's it for this video. And next up, we have the golden age of

5:09analytics and why analytics are in demand. See you there.

0:00Welcome back. So what is the golden age of analytics? The core idea here is simple.

0:06Organizations now treat data as a strategic asset and not just exhaust from operations,

0:14which was the case for quite some time. As a strategic asset, these organizations are also

0:20investing in dashboards and analytics programs and data teams because leaders expect answers like

0:28where should we cut costs or which customers are likely to churn, but most importantly,

0:34they need so with evidence. And that's why modern analysts are in high demand. So companies start

0:41with this raw material or that raw data, but they need people who can translate it into decisions.

0:48So you have that raw data that is now becoming valuable because leaders want to make evidence

0:54based decisions. And to do that, they need people to pull the right fields, clean the messy data,

1:01do the analysis, build visuals, and explain the story all in plain language. For data plus this

1:08context matters because CompTIA questions are usually framed like workplace decisions. That

1:16means given this situation, what would a competent analyst do? And if you understand the role,

1:23the exam stops feeling like random trivia and starts feeling like job scenarios. And now we can

1:29talk about the analytics temple. I guess we could say that this is a Greek style temple.

1:35And you can see that the roof here is going to be analytics and it's being supported by three

1:41pillars. We have data storage and compute or computing power, but no building is complete

1:48without a strong foundation. And that means that the whole thing sits on a foundational slab

1:55labeled governance. So this picture explains precisely why analytics is exploding right now

2:02instead of decades ago. And what's super interesting about this is that right before AI,

2:08that's when everyone is really rushing to use all of the data resources. And that's why data

2:15analytics is also very in demand. Before that it was in demand, but for larger companies. Now

2:21everybody's jumping on the bandwagon because data is the new gold. And since AI is here, well,

2:28what does AI run on? Data. Data has exploded. Transactions, customer interactions, sensor logs,

2:36web events, photos, videos, audio, and storage used to be painfully expensive. Now it's cheap

2:43enough to keep far more historical data for far longer. And that means that analysts can now spot

2:50trends and long tail behavior, but also computing power has skyrocketed and also cloud, which is a

2:58new focus of the CompTIA Data Plus. And that means that analytics aren't just limited to supercomputers.

3:04You can rent serious horsepower on demand through the cloud. And in fact, that's what we're doing

3:10with Google Colab. I have access to tons of GPUs through Google Colab. I do not have supercomputers

3:17here at my house. I pretty much rent all of my compute through the cloud. I do have machines

3:23locally for learning and building local models and things like that, but the heavy horsepower,

3:29I rely on the cloud. And don't forget this foundation of governance. That's what keeps

3:35all that power from becoming chaos because we have access control, privacy, quality standards,

3:43and rules that are responsible for use. Now let's do a real analyst workflow on a churn scenario.

3:51If we decide to zoom into the day-to-day, the data analytics process has five stages. And what I was

3:58super excited about is that, well, these five stages are also the five stages of data analytics

4:05at your job. So that end-to-end analytics process are mirrored in the five domains of Data Plus,

4:12which is convenient and probably by design. All right. So we start with acquisition,

4:18which is domain two, because we don't need to focus on domain one, which is our development

4:23environment. Acquisition means pulling from internal and external sources, but also consider

4:30reality, cleaning and manipulation. Then reality hits because you also have to contend with

4:36cleaning and data manipulation. You have missing values, invalid codes, duplicates, inconsistent

4:43formats, and all of this could consume a huge share of project time. And this is where the

4:50analyst earns their money, really, because you can make data usable and trustworthy. And that's

4:56why cleaning and manipulation takes so much time. And sure, because it offers the most value. And

5:02again, I could just say that this is where the analyst really earns their money because they're

5:07making the data usable and trustworthy. Then we move into analyzing. And this is often where we

5:14start with EDA or exploratory data analysis. You can run simple summaries and visuals to help you

5:21better understand distributions, maybe spot some outliers or form a general hypothesis. Then we

5:29move into visualization and reporting and communication because we have to turn those

5:35results into something humans can act on. And crucially, the process is iterative. That means

5:42that once we're done here, we'll go all the way back if needed to the first step, which means

5:49getting new data and then cleaning it, analyzing it, visualizing it, and then communicating those

5:54results via a report or dashboard, some form of communication. All right, so enough theory.

6:01In the next video, we're going to dive into Google Colab and try this out. We're going to perform an

6:07end-to-end analytics process, much like you would find at a job. See you there.

0:00Welcome back.

0:01In this video, we're going to talk about Google Colab, which is exactly what you see here.

0:06The one that you're going to open looks like this.

0:09It has a bunch of information or code here, but I do want to explain what we're doing

0:13so that this is not confusing.

0:15All this allows us to do is to use Google in the cloud and we have Python loaded here.

0:21And if I go to runtime and then I go to change runtime type, I could choose different GPUs.

0:27I have a CPU here, but I can also use a T4 or V5E-1 or any of these like H100, but you

0:35have to pay for these.

0:36They do let you use these for free, which is pretty cool.

0:40And you can also use R and Julia, not just Python.

0:43So in short, this is probably the easiest way to get Python up and running on your machine.

0:49So I can prove it to you.

0:50I can just say print.

0:51This is what programmers would use when they're testing out a development environment to make

0:56sure that they have Python loaded.

0:58And it takes a second to connect to the cloud, but there it is.

1:01So Python's version.

1:03But to do this, we're going to need the exclamation mark so we can talk to the backend.

1:07And that lets us know that this is Python 3.12.12.

1:11But don't worry.

1:12If you don't know Python, you definitely are going to learn some Python because I'm going

1:17to make it really intuitive.

1:19But more importantly, I'm going to show you that you don't have to use Google Colab.

1:23You can use whatever tool that you're more familiar with.

1:25For example, let's say you work with spreadsheets and that's easier for you.

1:29Let me show you how you can use those.

1:31All right.

1:32So now that you have Google Colab, all you really need is a Google account.

1:35You can see that I'm signed in here in the top right.

1:38And once you're logged in, all you have to do is click on this Google Docs link and it'll

1:43open up and you should be able to see everything that I see and you should be able to run everything

1:49that I run.

1:50However, you will not be able to make changes because this is not yours.

1:54But if you want to make a copy of this, all you do is go to file, save a copy and drive

1:59and then it'll copy this notebook for you.

2:02I accidentally clicked on text, which I'm glad I did.

2:05But if you do that, you can click on delete here.

2:08And if you want to run these code cells, you can do that with this play button or you can

2:12use shift enter.

2:14But if you hover right here in this center area, you'll get either code so that you could

2:18write code or text so you could write text.

2:21I'm going to delete these two and I'm going to show you how to run this first cell.

2:26So shift enter.

2:27And this is talking to the cloud, so it's going to run and then you'll see this check

2:31mark.

2:32It completed in zero seconds.

2:33It was very, very fast.

2:35And before I show you how you can use this in a spreadsheet, we have to load these sources

2:40because in real work, you almost never start with a perfect table.

2:44I wish you start with files from different systems.

2:48So here we're going to use our CRM and our demo file.

2:53So this CRM right here is an internal CRM that has churn and some interesting behavior.

3:00And we can call this the internal truth.

3:02And then here we have a kind of a demo.

3:05I call this demo and this is just enrichment.

3:08This is external data, but there is data mismatch and duplicates and missing data because I

3:13want it to be a little bit realistic.

3:16So let's click on this folder here and you're going to see that if I click on refresh, you

3:21just have sample data and all of this.

3:23Not really a big deal.

3:24But once I run this, we're loading the data.

3:27So this is the very first part.

3:29So now that you've loaded this data and you click on refresh, now you have CRM customers

3:34CSV, which is a comma separated value.

3:38And now you have demographics dot CSV.

3:40That's the demo, which is the enrichment data, which is dot CSV as well.

3:45So you can just click on these and then download these.

3:47So once you've downloaded these, you can go to Excel or even Google Sheets, go to file

3:53and then import.

3:55And once you do that, then you would go to upload here and then browse or drag a file.

4:00And then once you uploaded that, now you would have that same CSV for the CRM, which is,

4:06I closed this one.

4:07We don't need that, which is here.

4:08OK, so this is the CRM data.

4:10So you don't have to use Google Colab.

4:13You can use spreadsheets.

4:14So if this is more familiar for you, you can see that we have customer ID from 1001 to

4:201009.

4:21If we go back to this data here, you can see that we have 1001 to 1010.

4:26And we're going to explain that in just a second.

4:28Why do we have 1010, but there's no 1010 here?

4:32This is going to come up and this is part of the lesson.

4:35OK.

4:36So now that you have these two data sets loaded into a table, you can convert to a table like

4:41this.

4:42And I'm going to close this.

4:43And now you can do things in this table.

4:45So I can do math if you have the correct data types.

4:48And again, that's part of cleaning.

4:50So let's go back over here and get started.

4:53All right.

4:54So all I've done is just imported pandas, which is a digital version or let's say a

4:58Python version.

4:59Right.

5:00It's a Python package for tables and data manipulation.

5:04And then we have NumPy, which because pandas is built on top of NumPy.

5:09And then you have matplotlib for visualizations.

5:12Normally, Colab has these installed, but I'd like to show this as a best practice in case

5:17you're using non-Colab like Jupyter Notebook, which is a local version of what you see here

5:23or a non-cloud version.

5:24All right.

5:25So don't worry about all of this stuff.

5:27We're just loading the data.

5:29That's all we're really doing.

5:30And I'm giving you some options so that you can use spreadsheets or any other tool by

5:36downloading the CSV and performing those actions.

5:40Because this is a vendor neutral certification, which means, well, if you understand the process

5:46on multiple tools, that's all you really need to do.

5:49So you can follow along with me and it won't be too difficult to do that.

5:52We're not writing any code.

5:54You can use Google Colab or you can use the spreadsheet.

5:57I'll leave it up to you.

5:58All right.

5:59So you've loaded this data or you have it in a spreadsheet.

6:02So now I'm going to return the first five rows with this code right here.

6:07And now you can see that we have 1001 to 1005.

6:11But let's say you want to look at the last five, you can do something called tail.

6:15So the last five here.

6:17And you can see that we have 1010 to 1004.

6:21Now let's do the same with the CRM tail.

6:24OK.

6:25And you see this one goes to customer ID 1009.

6:28It doesn't go to 1010.

6:30So that is just a little foreshadowing of something that we'll need to address together.

6:34All right.

6:35This is pretty much what data acquisition actually feels like.

6:39So where are we in this process?

6:41We basically are in the first stage, which is acquisition.

6:45And this is before you clean anything.

6:47You can confirm what you actually have.

6:49And that's what we've done here with head and tail.

6:51I could also just do CRM like this and get the first 10.

6:56And in this case, it's trying to show me as much as it can.

6:59It will truncate it if this is a big data set.

7:02And so the first thing that I would do is to inspect the columns and the row counts.

7:07So the columns is going to tell you what fields exist.

7:10So here are the columns.

7:11So you have customer ID, plan, tier, churn, monthly, and so on.

7:17And if you see some missing values, then you'll know what kind of values are missing.

7:21Row count tells you if you're about to lose data during a join.

7:26So sometimes you can perform a join where you take two tables and combine them.

7:31And based on that join, sometimes you can lose rows.

7:35And looking at the row count is a good way to make sure that you're staying one step

7:40ahead of that, meaning you're not losing any rows during a join.

7:44All right.

7:45So this is the habit.

7:47You want to always check the columns, number of rows, and use a tail or head or just look

7:54at the data like this.

7:56And you can also do so here in a spreadsheet.

7:58All right.

7:59Now let's move down over here.

8:00So we've already acquired and loaded that.

8:03So let's, well, we did that here inside of Colab, but we need to load those here.

8:08And the way that you would do that, let's say that we didn't use pandas, but instead

8:13we have them here, right?

8:15And if that's the case, you can upload CSVs to Google Colab.

8:19And if you run this, then this is the load and inspect.

8:23And then here, it's going to tell me what the column names are and the row counts for

8:29each data set.

8:30So I have 10 and then eight, oh, oh, warning.

8:32Let's make sure that we keep this in mind.

8:35And whenever we perform joins, we want to make sure that we're not losing rows.

8:38And now here's where we make our money.

8:40This is the stage where analysts actually earn their paycheck and we're making the data

8:46set joinable and trustworthy.

8:49So what does that mean?

8:50So first, let's go ahead and deal with this one right here.

8:53So when you're cleaning and manipulating, that includes fixing things like white space,

8:58which are hard to see, and also data type mismatch.

9:02So before we do this, let me just show you what this is here.

9:05Let's inspect this a little bit more.

9:07Let's say crm.info.

9:09OK, so what we want to do is make sure that the CRM customer ID and the demo or demographics

9:15customer ID are both going to be strings, because right now it's an integer.

9:20So we want to change that to string.

9:22And we also want to fix the white space, remove the white space, because when you're doing

9:26a join, it's very strict.

9:28If you have one extra space, there's going to be no match, meaning if you had one extra

9:33space, we would get a bunch of unknown regions and the join would fail.

9:38So that's why we're doing this.

9:40So let's go back to here.

9:41Now we're going to see that they're integers and we're going to run.

9:45Let's go ahead and just run this here so that you can see the changes.

9:49So now I'm going to run this cell again so that I can see what's changed.

9:54And now I'm going to be paying attention to customer underscore ID here.

9:57And now you can see object and object is a string.

10:01That's just what Panda calls a string.

10:03So just have to memorize that if you're working with Pandas.

10:07And if you're working with a spreadsheet, it's going to be slightly different.

10:10But what we're doing is just changing it from an integer or a number into text or a string,

10:16which Pandas calls an object.

10:17And if I change this to demo, we would see that we also have now the customer ID as an

10:22object.

10:23OK, so I'm just going to delete this because we're going to do this down here again.

10:28All right.

10:29So next you want to drop duplicates.

10:31What you're doing is you're collapsing multiple records to one per customer ID.

10:36And here's something important to consider.

10:39In real work, you don't blindly keep the first row right here.

10:43And I put this here because we're just doing that because this is a toy demo.

10:47You would decide that based on the data that you're working with, maybe the latest update

10:52or the most complete row, et cetera.

10:54In this case, it's a toy data set.

10:56So we're just going to keep the first row like this.

10:59So that's something to consider.

11:00And then here we're going to merge before we do merge the duplicates.

11:04The reason that we're doing this, too, is that duplicates can inflate counts and distort

11:09rates.

11:10So this is important.

11:11Now we're going to do a merge, which is a join.

11:13So here we're going to emphasize the left join.

11:17I put that here in the comments, but you can see here how is a left join.

11:21So our CRM is the analysis population and demo is the optional enrichment, the demographic.

11:29And one thing to notice here is that we're not using demo as the base.

11:33And as a result, if we go back up here, you can see that we have 10, 10 here and demo.

11:39And since we're not using 10, 10 or demo as the base, that means that it's going to disappear

11:43here when we perform that join.

11:45Let's go over here.

11:46OK, so that explains that join there.

11:48And then finally, we have simple missing handling.

11:52And for that, we're just going to handle missing values with fill and a unknown.

11:57And that's how we do that right here.

11:59We're just going to handle the missing values.

12:01And when we see a known, we're not saying that the region is truly unknown.

12:06What we're saying is that we're labeling missing data so that we can track it.

12:11That's the whole idea here.

12:12All right.

12:13So now I'm going to grab this little line here and remove it.

12:16And then I'm going to run this cell and paste this here.

12:19And this is where you check everything, right?

12:20We can do this one by one as well.

12:22So here we have the rows after dupe.

12:25Let's do this one here.

12:26OK, so nine.

12:27That's good.

12:28And then finally, we're going to check for the percentage of missing demographics.

12:34And here we have 33.3 missing.

12:36And then finally, we can do this, which is merged head.

12:40And now you can see that you have both of those tables merged.

12:43So what's important to consider here is that every analyst should run sanity checks after

12:48a merge.

12:49And that's what these were right here.

12:50So you want to check your work, right?

12:53If you do a join in these spreadsheets, you want to do the same thing because this is

12:58vendor neutral.

12:59We just need to do the sanity checks after you do the join.

13:03And now we can finally analyze.

13:06But we start simple.

13:07We're going to use descriptive statistics first and a little bit of EDA, but not too

13:13much.

13:14So we want to analyze the churn rate by plant tier.

13:17And we're going to say churn by tier.

13:20And that's how we get churn rate by plant tier.

13:24And then here's a very quick EDA.

13:25And what we're doing here is just asking questions.

13:28Do the ranges look same?

13:30Do we see weird minimums or maximums, right?

13:34Or is the data skewed in some particular way?

13:37So again, here for the EDA, we're just asking questions, not making predictions about churn

13:42just yet.

13:43You're just identifying patterns that are worth investigating.

13:46So let's go ahead and run this.

13:48So we have basic and enterprise at 50% churn, and then we have pro at 33.

13:55Here's some very simple statistics.

13:57I think this is dot describe.

13:59Yep.

14:00This is how we do that.

14:01And here we can create a simple visualization, which is, in essence, the same visualization

14:07or the same insight, I should say, as the table, but faster for humans to scan because

14:13it's visualization and not all this code.

14:16So let's go ahead and run this.

14:17All right.

14:18Let's check this out.

14:19So here it's the same information, right?

14:2050%, 50%, 33%.

14:23But the cool part here with this, bar charts are great for category comparisons.

14:28Look at that.

14:29You can see that and you can show this to non-technical audience and you can add labels

14:34and titles, whatever you need for interoperability.

14:37Okay.

14:38So here, let's do a quick distribution check on last login days.

14:43So what we're doing here is pretty classic EDA.

14:46How is the engagement distributed?

14:49All right.

14:50So histograms will show you a shape like, and that means like a cluster or let me just

14:56show you.

14:57All right.

14:58So if you had a normal distribution, you would see something that goes up like this, right?

15:02Let me change the pen color.

15:04A normal distribution.

15:05So again, here's a different color.

15:07A normal distribution might be clumped up like this in the middle, but you can see that

15:13this distribution is skewed like this and this is called a right skew.

15:17So now we're looking at the distribution and again, we're just looking at the shape and

15:21this supports hypothesis building.

15:23You can say our most inactive customers churning and other questions.

15:28And you can think of this as a investigation, like a quick investigations.

15:33That's sort of like what I like to think of EDA.

15:35And if we go down here, finally we have some kind of communication, right?

15:41So we're not going to get too deep into this, but if we run this, here's actionable takeaway.

15:45The highest churn is in the basic tier.

15:47Well, it is, but it is also in the highest, in the, where is it?

15:52Let's go here.

15:53It's also an enterprise tier.

15:55They're both 50% and next step segment that tier by inactivity and then some kind of next

16:01steps.

16:02All right.

16:03So this is pretty cool.

16:04We acquired and loaded that CRM CSV, it was multiple steps, right?

16:08And then we clean and manipulate.

16:10We fix IDs dedup, meaning deduplicated.

16:14We merged and we label missingness as unknown.

16:18Here we're going to clean and manipulate like this, what we just talked about.

16:22And then finally analyze.

16:23We compute churn rate and run some kind of a quick EDA.

16:27And then over here, we're going to check out the distribution.

16:31And this is also just the histogram.

16:33That's what it's called.

16:34And then some kind of communication at the end.

16:37This is just to show you how this works.

16:39And since I didn't go into it very detailed, let's do that in the next video where we talk

16:44about communication and insights.

16:47And to do that, we'll look at tables and visuals and reports and dashboards.

16:52See you there.

0:00A big part of being a modern data analyst is communication or the skill of communicating

0:06insights.

0:07You could say that, okay, we're choosing tables, visuals, reports, and dashboards and

0:12call it a day.

0:13But when we're looking at tables, we know that they are precise, but they're hard for

0:17humans to scan for patterns.

0:20They're precise, but there is a higher cognitive effort, meaning that a chart often makes the

0:25same point almost effortlessly and instantly.

0:29You can spot trends, outliers, and comparisons.

0:34Then we package visuals into deliverables.

0:37A report is a point in time snapshot because these charts and maps not only give those

0:43patterns visible at a glance, but a lower cognitive load.

0:47So a report is visuals plus narrative that explains what happened and why it matters.

0:53A dashboard, on the other hand, is for ongoing monitoring, meaning visuals that are updated

0:59regularly and support filtering and drill down.

1:03And be prepared for a scenario where you're given an audience and a goal and which deliverables

1:10and visualization choices might be best.

1:13But we really want to look at these deliverables and examine reports and dashboards with a

1:18little more detail.

1:20So now let's go back to Colab and try this out.

1:23Just remember, this is about communication.

1:26Analysts don't just compute answers, they package answers.

1:30That's a good way to think of it.

1:32Now in this Colab notebook, we're going to take the same churn insight and deliver it

1:37as a table, a chart, a short report, and a dashboard like view.

1:43Again, we're not going to be able to go into super detail here because we're just doing

1:48an overview of how all of these work together in the role of a modern data analyst.

1:53All right.

1:54So we've run our imports and now we need to load or recreate the merged data set because

2:00we already merged it.

2:01And so this should be pretty familiar.

2:03So I think that we can go over here to this folder.

2:06You'll see that there are no files here.

2:08So if you need those files again, you're going to need to go over here and download them

2:12from this place because I made them downloadable here.

2:16I turned them into CSVs just for this exercise.

2:19So definitely go to demo one.

2:21You should still have these downloaded if that's what you're doing.

2:23I'll close this and now we can see our table and the way that we're doing this is DF head.

2:30So let's go through this together.

2:31And a quick note, this is domain four style of thinking.

2:35You're choosing the best deliverable for the audience and goal.

2:38And again, we have two data sets, two different worlds.

2:41We have the CRM, which is the internal system.

2:44And then we have the demographics, which is this third party external enrichment.

2:50This could be region, age, income, and so on.

2:53And all of this mirrors a real business workflow.

2:57You want to join business behavior with some kind of context.

3:01And as you see here, we're also taking care of the white space because otherwise if the

3:06customer ID is numeric, like an integer, the join could fail silently because the keys

3:12don't match.

3:13And that's another important reason why we're doing that.

3:17Here we're dropping duplicates and here we're doing that join.

3:20And we're using fill NA for those missing records.

3:23And we're saying that we don't know what those missing values are.

3:26We're not saying anything specifically like unknown age.

3:30We don't know what age it is.

3:31It's just more like cracking the missing value, right?

3:34Okay.

3:35So now we're ready for that merge.

3:36Okay.

3:37So here's our table.

3:38So we have a pivot table.

3:39And what we want to do here is keep all the CRM customers and pull in the demographics

3:44information when available.

3:46And that's why we're also going to do, let's go back up here.

3:49I believe the join is here.

3:50There's the join, a left join.

3:52This means that the left is the CRM, which is the truth set, meaning the customers we

3:58have churn labels for.

4:01Demographics is the optional data, the enrichment.

4:05Okay.

4:06So let's go ahead and run this first table here.

4:08We could also say this is the table version, meaning that we have precise compact and we're

4:14aggregating raw rows into decision ready metrics.

4:18All right.

4:19So looking at this first pivot, this looks pretty familiar.

4:22And here we're also sorting the values and this makes the story, I guess, more readable.

4:28You're sorting and renaming.

4:30And this is something you could consider this reporting polish, right?

4:33To make it look good.

4:34And what we're trying to convey here is which tier churns the most.

4:38So let's do the second one here.

4:40And here we want to shape it like a dashboard matrix.

4:43Okay.

4:44This is more of a report place in time.

4:46So let's go ahead and do this one here.

4:48It's categories on the side, and then you have these segments across the top.

4:52And so here the index is now the plan tier and the columns are the different regions.

4:57Okay.

4:58And this creates a grid that's going to be visually easier to support comparisons.

5:02And that's why we're doing this.

5:04And you could also say this is where the raw data becomes a monitoring view.

5:09And if you're working in a spreadsheet, well, pivot tables are the backbone of Excel and

5:13dashboards and BI summary.

5:16So this should be very easy and other tools as well.

5:18Okay.

5:19Now going down to this pivot chart here.

5:21And like we said before, these tables are exact, but charts are fast for human brains.

5:27So let's go ahead and run this.

5:28All right.

5:29So it's really easy to see that we have the highest return rates are both basic and enterprise,

5:34nothing new.

5:35And now I want to compare a report versus a dashboard.

5:38This is really important.

5:40And I'm going to leave these here for you.

5:41A report is a point in time and a dashboard is ongoing.

5:46That's the takeaway.

5:47So when we're talking about a report, we really mean narrative bullets and some kind of chart

5:53or table or a snapshot.

5:55So you could say that churn is concentrated around the less engaged customers, but you

6:02actually haven't proven it in this notebook.

6:04So it really depends.

6:06This is toy data, so we can't take it too far.

6:10That would be more of a hypothesis that you'd need to validate with some kind of a quick

6:14check, like a correlation or grouping by last login days or something like that.

6:19So what we're doing here is labeling what's confirmed or comparing that to what's suspected,

6:24right?

6:25So here we're labeling what's confirmed versus what we suspect.

6:30Now let's take a look at the dashboard.

6:31Well, let me run this cell here.

6:33OK, so here's a dashboard.

6:35It's saying here that the highest churn rate is basic, but this is just a demo, right?

6:40You can see that they're tied.

6:41So we get the highest churn and churn appears concentrated among less engaged customers,

6:46right?

6:47And the next step might be target retention offers for high risk customers in the top

6:51churn tier.

6:52OK, cool.

6:53Now let's look at this dashboard.

6:55A dashboard is not a story.

6:57That's the first thing to say.

6:59It's more of an interface, so it's meant for repeated use.

7:03That's the difference between a report and a dashboard.

7:05And that's why we're using a reusable function.

7:09It's more of an interactive approach.

7:11You could say an interactive mindset.

7:14And here we're going to plot churn by tier for a particular region.

7:19And you could say this is simulating kind of like a filter.

7:22It's the same metric, but we're packaging it a little bit different.

7:26It's the same as a pivot table idea, but it's just applied after filtering.

7:31That's one way to think of it.

7:32We're handling empty filters gracefully.

7:35So that's just a little detail.

7:36And then we're going to run two regions live.

7:39So let's go ahead and run this and let me show you what I mean.

7:42And now let's run this.

7:43OK, so now you have basic and pro and then you have basic and enterprise as a comparison.

7:49You can also use this UI that I have right here.

7:52Let me run this and show you how that works.

7:55And you can just click on South.

7:57And this is really great for non-technical audiences.

8:00Here's the unknown, West and Midwest.

8:04So we'll go back to South here.

8:05I'm showing you that you could use this user interface

8:08dropdown for the dashboard.

8:10All right, so that's it for this video.

8:12And I'll see you in the next video on core analytics techniques

8:16and where EDA fits in.

8:18See you there.

0:00Under the hood, analytics uses different classes of techniques,

0:04depending on the question that could be descriptive, inferential,

0:08predictive, and prescriptive.

0:10And that's what we're going to cover in this video.

0:12We're going to connect directly to the data plus domain 3.0 analysis,

0:18especially choosing the right method for the situation. And we'll start.

0:21So let's start with descriptive here.

0:24We're answering what happened and we can use summaries and distributions,

0:29but this is where the EDA usually starts.

0:33When we're looking at inferential statistics,

0:35we're talking about confidence intervals,

0:37which we'll get into a little bit later, including tests and regression.

0:41But for now, inferential answers,

0:43what can we conclude about a population from a sample? Okay.

0:48And when we're talking about predictive analytics,

0:50we're talking about forecasting risk and likelihood.

0:54And then finally here with prescriptive and you can think of a prescription

0:59here. We're recommending actions under constraints,

1:02take two cookies and call me in the morning. That's my kind of doctor,

1:07but that's just really an easy way to remember this. And again,

1:10we'll have a lot more opportunities on how to choose the right method for the

1:14question.

1:15When we continue to explore core analytics techniques throughout this course,

1:20and we're just getting started. All right. So now to talk about AI here,

1:24modern analytics, isn't just charts and averages anymore.

1:28And it's increasingly AI powered. And when I say AI, AI,

1:32isn't just one thing. So here AI is an umbrella term for machine learning,

1:36deep learning, generative AI, but it's not RPA.

1:42Then that's what I wanted to talk about. Artificial intelligence.

1:45It's really an umbrella term that includes everything such as

1:50deep learning, machine learning, generative AI, NLP, and more.

1:54The purpose of this slide is to let you know that RPA is not

1:59the same as AI. RPA stands for robotic process automation.

2:04Let's break this down. So when you're thinking about AI,

2:07just think that AI is an umbrella term.

2:09And there's two main fields, machine learning and deep learning.

2:13These are subsets of AI. Don't worry about that for now.

2:16These are techniques that imitate aspects of human behavior or

2:21decision-making. A simpler way to think of this is software modeled

2:26after the brain.

2:27It uses artificial neurons and all of this really super interesting stuff that

2:32we're going to get into in this course. So by the end of this course,

2:35you'll have a good idea of machine learning and all the subsets and related

2:39technologies. For instance, when we're talking about machine learning,

2:43this is really a subset of AI where algorithms learn patterns from data.

2:48Here, when we're looking at deep learning,

2:50this is a subset of machine learning,

2:52which is a subset of AI that uses neural networks,

2:57especially with images, audio,

3:00and large scale text. And that's the software modeled after the brain.

3:04And that's why we're using neural networks or more accurately,

3:08we should say artificial neural networks.

3:11We have biological neural networks and these models,

3:15AI models use artificial neural networks, and there's so much more.

3:20For example, generative AI,

3:22this is where the chat bots like chat GPT and Grok and Anthropic

3:27come from, which includes NLP. We have here,

3:30which is natural language processing. Again,

3:33we'll get into all of this very soon. Just remember in analyst work,

3:37this can help draft narratives, summarize documents,

3:40suggest SQL code or scaffolding,

3:43or translate findings into plain English,

3:47but you're still going to verify outputs and follow governance.

3:50You don't just rely on AI. It's actually pretty dangerous to do that.

3:54And then finally we have the whole purpose of this slide,

3:57which is to compare AI and RPA, which is again,

4:01robotic process automation. This is where people get confused.

4:06RPA is not learning it's rule-based automation.

4:10These are bots that do repetitive steps like exporting a report,

4:15moving files, refreshing templates, or maybe sending routine outputs.

4:20So just remember, these two are not the same.

4:23RPA is kind of the boring one.

4:26And all these other technologies are the exciting AI ones.

4:29So just remember RPA is kind of automated, a little bit boring,

4:33but that's a good thing because we want it to be automated and rule-based and

4:38predictable.

4:39And when we look at what machine learning looks like every day for an analyst,

4:43you might see something like customer segmentation here.

4:46We're grouping customers to support retention or maybe onboarding offers or

4:51to identify the strongest customers. Anomaly detection here.

4:55You want to look for those spikes, right?

4:57You want to spot unusual spikes and logs errors or activities.

5:01Outliers are a good way to spot those. And finally, we have sales forecasting.

5:06Here we're estimating future demand from historical data or historical

5:11patterns.

5:11And all of this power only works if people trust the results and the

5:16organization stays safe.

5:19That's why governance is the foundation in the temple that we looked at.

5:23Remember that analytics roof, the three pillars.

5:26And then now here we're talking about governance, the foundation.

5:31These include consistent definitions, access control, privacy rules,

5:35quality expectations, and even documentation practices.

5:40So if you don't write documentation, maybe you should,

5:42it's a big part of data governance and AI governance moving forward.

5:47This is what lets analytics scale responsibly. All right.

5:50So here we have 1.0. You want to understand,

5:54this is the first domain. These are systems, data concepts, and AI terms.

5:59All right. So let's map these to the different domains. So for 1.0,

6:04we're really thinking about systems, terms, AI concepts,

6:09data concepts. For 2.0, we're thinking about acquisition and cleaning.

6:14That means acquiring, cleaning, shaping the data, basically preparing it.

6:18And for number three, where we analyze, these are analysis methods,

6:23statistics, machine learning choices, maybe deep learning.

6:26And then once you have something to talk about, well,

6:28maybe you can create some visualization,

6:30some charts or a report or dashboard, depending on the context.

6:35And then finally governance. Here,

6:38we want to ensure access privacy quality and trust for responsible use.

6:43And that's it for this last video,

6:45where we explain the modern data analytics roles. And until next time,

6:50I hope this has been informative. And I'd like to thank you for viewing.