AI and Machine Learning are everywhere and people often confuse Analytics with these two…
or even with Engineering or Tagging.
Descriptive Analytics is what you will often do and what companies usually need.
But even so, you don’t simply need analysis but business and domain knowledge.
Today this task is relegated to specialists (e.g. SEO, PPC, Product, Growth)… but does it even need to be?
Table of Contents
Descriptive Analytics… What?
There are different types of Analytics, as summed up by the usual picture I like to share:
Describing something does NOT identify the root causes of why it happens.
Analyzing trends and past events is what Descriptive Analytics is.
We have no information about what could happen in the future but…
your domain knowledge should help you!
This is quite different from Statistics which often helps you answer those questions.
I have many people asking me stuff like A/B testing or even from unrelated fields.
Everything has its place but today we’ll just talk about a specific piece of Analytics.
And if you are interested in testing, my friend Giulia has a great free resource.
The Reality Of Your Average Website
Most websites require simple operations, even enterprise projects.
If you have properly understood the definition of Analytics, you know that it’s about simplicity.
Many get this wrong and confuse it with:
- learning Python/coding
- SQL
- programming
- SaaS
It’s none of them!
It’s a way of simplifying reality and asking good questions.
In many cases, you can survive with Looker Studio, BigQuery and some common sense.
Basic data wrangling will get you quite far, really.
We could summarize the analytical maturity of a company with the picture below:
Many companies can stop at the “Beginner” level and still get a massive profit out of it.
In my article about Content Auditing I show you how to audit content and how you can make it convenient and useful for any company.
This is a prime example of Descriptive Analytics at play and what you need for most cases.
A Brief Case Study
I had the chance to work with an enterprise and most of the work I did was quite “basic”.
Now, technical people usually frown on this type of work because it’s not cool or challenging. From a business perspective, it’s the opposite.
For example, defining the requirements and how to structure a dashboard are elementary tasks, right?
Well, it turns out many projects fail because these are not even done properly!
The actual hard skill is domain knowledge and being able to understand what is needed.
Going back to the example:
- Identifying the correct North Star, e.g. closed won opportunities > revenue
- Figuring out how the metrics interact. E.g. New vs Returning Users and how these lead to the creation of MQLs
- Understanding the flow, how are people actually using the dashboard
- How can we tie all of this to a sustainable and scalable process
See? No models or anything complex but the value added is immense.
And here it’s time to mention one of my best use cases:
This problem is often complicated but it’s not that. hard to find a method to quantify it…
These are real-life examples of “simple” solutions that require quite some engineering and governance work.
But again, we are not doing any Predictive Analytics.
The same goes for another use case, Content Decay, which is based on an extremely simple yet powerful concept.
If you consider how much time is saved and how you can help your audience, it’s much easier to build something with data.
If you want to analyze website, you need SEO (GSC too) and GA4 knowledge, both of them I deeply cover in my course:
If you want to learn more about this topic, I’ve prepared an Analytics for SEO course that covers both theory and practice:
“I don’t have enough data”
How often did you hear this sentence in meetings?
I hear it all the time and it’s annoying like a few things in life!
Remember that the goal of Descriptive Analytics is to have enough to make good and quick decisions.
Think about the practice of sampling, i.e. taking a piece of your data.
This topic is all but trivial and I would need an entire article on it…
just joking, for Web Analytics, it’s more than fine to be inaccurate.
As a rule of thumb:
- GSC data is heavily anonymized but can’t be tampered with (excl. when tools massively scrape you).
- GA4/alternative data can be wrong due to improper consent mode setting and even bot traffic. Apart from that, its data is sufficient to make informed decisions.
- 3rd-party tools like Semrush, Ahrefs and Similarweb, should be taken with a pinch of salt.
The Value You Can Add
Analytics should be seen as a way to multiply the capacity of a business.
Analyzing trends is powerful enough to get more traffic or sales with your website.
A very underrated example is provided by Microsoft Clarity.
It’s a simple tool you can use to analyze how people are using your website and improve your UX.
Does it require advanced knowledge? No.
Yet, it adds more value than countless hours on bounce rates and tricky metrics.
The advice that helped me the most is to think about outcomes and how you can add value.
When not sure about something, ask right away!
I am the first guy in the room to challenge solutions I don’t understand.
It’s for the good of the project, even if you come across as annoying.
Is this solution going to add business value? How?
And we get back to the issue that people think about solutions first and problems second (which is wrong!).
A lot of these topics are covered in my course to be released on May 12.
Leading VS Lagging Metrics
If you’ve been reading me in the last few months, you know that leading metrics are used to affect outcomes (lagging indicators).
It’s correct to focus on financial metrics (e.g. revenue) but you shouldn’t ignore traffic and all the rest.
That’s because revenue is your outcome and is affected by other factors.
For example, a shift in the conversion rate can affect how much you make…
but if you only focus on the output, you will not find this important link.
This is why we rely on metric trees:
Think about a systems of pulls and levers, your levers are what affect the final outcome (lagging indicators).
From a purely operational perspective, this is all the forecasting you need.
Identifying correctly what affects your outcome is already a form of “prediction”.
So you can shift from metric trees to questions:
Once again, we are not using models or anything super complex, just common sense and business knowledge.
And then, those metrics can go straight to your dashboards or be used to affect your actions:
This is an example I just made up to show you how to think about dashboards.
Say, it’s a B2B SaaS, you can have conversions, new users and user stickiness (DAU/MAU) as your KPIs and then MRR (or anything else) on the far right.
Take this as a demonstrative example, of course!
A Necessary Parenthesis
I haven’t really talked about metric trees in this article, that’s for another one.
For the time being, remember that picking your most important metric (aka North Star) doesn’t imply it will always be Revenue or some financial metric.
From a Product perspective, this would be considered wrong.
So remember that financial metrics matter a lot but aren’t always your first and foremost target when analyzing data!
Redefining Value In The Era Of AI
Many of you ask me “Do I need coding to do X or Y”.
This is a legitimate question and I am happy to tell you “No, you don’t really need to be a master at it”.
It’s crucial that you learn how to think and add business value.
The 3 above are a summary of the principles I live by daily.
If you have a good knowledge of Marketing, Web Analytics will be much easier.
Otherwise, you’ll be learning a lot of notions without context.
P.S. Hard skills are still the baseline, of course!
Web Analytics is often relegated to technical skills but it doesn’t have to be!
Marketers with technical skills or Analysts with marketing/business knowledge sit on the left side of the spectrum:
This is what already adds value and will make a breakthrough in the era of AI.
A folk with good marketing knowledge is already invaluable, imagine one who combines tech knowledge too!
And What About Other Analytics?
Yes, I’ve talked well about Descriptive Analytics but it’s not enough, as you may expect.
Predictive solutions (i.e. ML algorithms) are cool and fine if you are sure your data is decent enough.
This is not often the case in SEO but I am quite bullish on PPC and conversion data.
The big deal with ML is that building a model isn’t enough because:
- it needs to be maintained
- it should be deployed
- it needs to be trained on the “right” data
Prescriptive solutions can be quite vague and I don’t have a strict definition.
Preventing damage or understanding what should happen doesn’t necessarily require complex modeling.
Being proactive is already enough to cover most use cases.
The Existing Divide
I am against a rigid distinction of roles, it doesn’t matter if an Analyst does X or Y… what matters is the outcome.
The idea is that you are able to analyze your data in the majority of scenarios and don’t overspend in methods that don’t deliver business value.
In fields like SEO, I’ve seen infinite approaches on how to estimate “ROI” or prepare estimates…
not only most of these methods don’t make sense but they are expensive in terms of efforts.
There are exceptions to this rule, though. So far I’ve seen more value in Machine Learning methods like Clustering but there some useful exceptions where AI is required:
- Translations, after all, language models excel at language
- Converting unstructured data into structured and tidy data
- Labeling pages and classifying them into groups
Even so, you need to write good prompts and understand how these models work.
Some Real Life Situations
Theory is nice but real life hits hard.
The majority of scenarios go like this:
- GA4 is set up incorrectly and/or GTM is not used to track relevant events
- no one knows how to use the data to build use cases
- your boss/client obsesses over new tools
My complaint when I was more of an SEO than an Analyst was that data people didn’t (often) understand the data.
Now as a 100% Analyst, the situation hasn’t changed.
The industry is filled with practical knowledge and tutorials but rarely goes beyond and looks at the value of data.
This is where Seotistics helps you and also why shifting to a different mindset is 90% of the effort.
Once you know what you want and the requirements are clear, I recommend considering data as a product.
You Paid To Use Web Data But You Got Nothing
This sounds pretty much like Marketing, right?
We’ve gone through years of crazy hype over data and now expectations have shifted.
The answer is that companies often want something without having the right structure.
If you can’t support these initiatives, take a step back and focus on something else.
The reason is usually tied to how we interact with data.
Many companies never cared about having a strategy, do you think they will succeed with data?
The Great Disconnect
When I was a student, I recall that many people with a pure background in Computer Science went into Data Science and Analytics.
It wasn’t necessarily a great thing seeing how the industry evolved!
An unhealthy obsession with technology is one of the factors that made people see data as a cost center.
This is why today you see people focusing more on the business (finally!).
The success of your analysis largely depends on how well you know the industry.
If you jump to tools and want to work on cool stuff… then Analytics is not for you!
Many often claim that Descriptive Analytics alone isn’t sufficient and can be quite limited…
I will tell you what, it’s not true in Marketing.
In multiple occasions, I have taken important decisions just by looking at historical data and using my 1st-hand experience.
Do you think running expensive models and spending months of work is better than years in the trenches?
Think again!
Insights -> Action
If what you propose doesn’t have any repercussion on the business or doesn’t lead to any action, no one will care.
To prevent this, shift your mindset to always tie insights to business actions (or even outcomes):
The risk in thinking in purely practical terms is that you miss the big picture.
You may think that a series of disconnected tasks makes perfect sense when you are just hitting blank spots.
The best and simplest thing you can do is tie your results within the framework below:
An audit can be considered an entire process falling inside Content Management, a general practice and system.
The smallest unit is a single activity/task, for example updating a page or even deleting it.
Considering existing processes and how they fit within the organization resources may not be 100% Analytics but it’s common sense you must develop.
All of this is a way to bypass the Data-Business disconnect when paired with strong knowledge of Marketing.