Issue #29 – The Four Levels of Analytics
What it means to ‘do’ analytics properly, evolving from descriptive to prescriptive in a logical manner
Read time: 12 minutes
Today it seems like every data domain is severely misunderstood….
But I think the domain that takes the cake is analytics, a word that is so buzzy that nobody quite knows what it means.
But Dylan, everybody is still doing analytics, so what do you mean?
Well, while most companies do some analytics work, I also see them:
Treat analytics teams like ad-hoc order takers
Never clearly define business questions or use cases
Conduct manual analyses primarily using offline Excel data
Don’t have a view of what types of analysis are happening across teams
Haven’t thought about analytics strategically and are treading water in their capabilities
Look, you can do analytics on Excel or even get value from ad hoc analyses.
But in the end if you want to excel at data, you need to take this domain seriously. And the best way to start doing that is by digging into this article on the four levels of analytics!
The Real Role of Analytics in Business
Before we define each level of analytics, let's talk about what it enables for businesses today.
Analytics isn't just about creating dashboards or running reports. It's about creating a systematic approach to understanding your business through data. With that knowledge, your people (or your AI) can make better decisions, improve business outcomes, and hit strategic goals/ targets.
I’ve written articles on the business strategy, stakeholder interviews, and building data products. These create the foundations for figuring out how analytics can add value to your business.
At the same time, your analytics needs to be propped up by high quality data, an architecture that enables analytical data products, and an operating model & org structure to help the business leverage data to make better decisions.
Before even sitting down and creating an analytical data product, think about these things:
What is the business strategy?
What analytical use cases enable that strategy?
What value do those use cases create?
What stakeholders benefit? Will they use these tools?
How do we build those products in a logical, efficient way?
Do we have the data fundamentals to operationalise them?
To understand your business through data, you need to think holistically.
So as we define each level of analytics, think about the relationship between them all and how they build on top of one another.
With that said, let’s dive into the four levels!
Descriptive Analytics
The first level of analytics is descriptive. Here is where teams analyse and visualise historical data to provide a clear picture of what happened in the past.
Realistically, descriptive analytics is the entry-level stage. Using historical data and apparent trends to make decisions has been done for decades. Reporting is a perfect example of this—companies love to produce reports and 95% of them are a simple historical perspective of what has happened in the business, whether it relates to sales volume/ revenue, customer traffic on the website, inventory/ supply chain, etc.
Here are some common forms of Descriptive Analytics:
Monthly/ Yearly Reporting – Standardized reports that track core business metrics and KPIs over set periods to monitor performance and progress against targets.
Website Analytics Reports – Tracking reports that show how users interact with your website, including sources, user behaviour, conversion rates, and engagement.
CRM Reporting – Customer-focused reports from your CRM tool to understand sales pipelines, customer interactions, and relationship health across the customer lifecycle.
Financial Statements – Standard financial reports like balance sheets, income statements, or cash flow that provide a view of an organisation's financial position and performance.
Low-Fidelity Dashboards – Simple, straightforward visualisations of key metrics that provide at-a-glance insights without complex interactivity or additional analytics.
Companies think they do these things well, but most don’t. Instead, teams pull these things manually; there are multiple versions of the truth; the data is poor; the metrics are inconsistent across stakeholders, and so much more.
So despite being entry-level and seemingly easy, here are some things companies should consider when doing descriptive analytics:
Keep it Simple – Use mainstream, understandable tools (hello Excel). Don’t overcomplicate calculations. Make sure everybody is on the same page.
Automate Data Collection & Cleaning – Document the data lineage and transformations. Ensure the data feed/ pipelines from the source are well-built and maintained. Reduce manual configurations and adjustments.
Align KPIs/ Metrics – Agree on metrics beforehand. Get alignment across all stakeholders who use the tool. Define how those KPIs are calculated and the outcomes they feed into.
Be Realistic with Takeaways – Don’t infer things that aren’t represented. Know what you get out of the tool and stick with that. Historical analysis has its limitations.
Descriptive analytics can be a powerful tool for any organisation, but only if stakeholders have the time to glean insights from within the data. Before yelling ML or AI, make sure your company can do this efficiently and effectively.
Diagnostic Analytics
The second level starts to get into the fun stuff, understanding why things happened! Beyond simple calculations of historical metrics, here analysts apply additional models on top of the descriptive, historical data to better understand the reasons behind the numbers.
I view Diagnostic Analytics as the most important level. This is where real insights start to form. If descriptive analytics tells you that sales dropped 20% last quarter, diagnostic analytics helps you understand the driving factors behind that drop.
This is crucial because most business decisions require context—it's not enough to know something happened, you need to know why to take meaningful action.
There are so many different types of Diagnostic Analytics:
Root Cause Analysis – Trends and patterns to identify the underlying causes of business outcomes or performance changes (i.e., why did this happen?)
Customer Segmentation – Segmenting customers into meaningful groups by behaviour patterns, performance, demographics, etc. (i.e., why do customers behave this way?)
Marketing Attribution – Relative impact of marketing channels and activities (i.e., why did the marketing drive sales in this campaign?)
Performance Variance Analysis – Examining the outcome of forecasted performance metrics (i.e., why did results vary from the forecast?)
Process Analytics – Understanding Investigation of operational workflows to identify bottlenecks, inefficiencies, and improvement opportunities (i.e., why is one process less efficient than the other?)
These analyses can be found in reports, dashboards, or self-service tools that analysts or business stakeholders have access to. For example, a marketing segmentation analytical tool may have both marketing attribution and customer segmentation analysis to provide a more holistic picture of how things work.
Unfortunately, most companies struggle here. Doing this effectively requires the right business context/ input, a clean combination of data sources (e.g., sound data engineering attached to business process models), testing and validation, and demonstrated outputs via accessible technology. These combine many different data, technology and business domains all working collaboratively. And it also takes time, which companies don’t have.
In the end, they settle for descriptive analytics and think they are doing diagnostic.
So I will cap of this section with a clear progression about how Descriptive feeds into Diagnostic:
The descriptive analytics provides the Takeaway – “This is what happened.”
The diagnostic analytics provides the Insight – “This is why it happened.”
The business or data analyst derives the Implication – “This is what needs to be done based on that information.”
If your company can go from step 1 to step 3, you are doing analytics better than 97% of organisations out there (note: this stat was totally made up but I stand by it and believe it may even be a conservative estimate)
Predictive Analytics
With the third level we start getting into the fun stuff. With machine learning algorithms and advanced analytics, Predictive Analytics use historical data and other variables to forecast future outcomes, trends, or behaviours.
Using relationships and correlations between historical data and identified variables, predictive models aim to determine what is most likely to happen in the future. These variables might be determined by business stakeholders (who know the context) or by conducting statistical tests and exploring potential correlations within datasets.
When it comes to Predictive Analytics, a lot of people immediately think of machine learning. There is an inherent danger in making ML a catch-all category for this. Instead, here are five statistical Machine Learning approaches/ methods that you might consider for predictive analytics (and beyond):
Regression – The classic and most used method creates a relationship between a dependent variable and one or more independent variables. Analysts can use this to predict relationships and help forecast an outcome or future values. Linear and logistic are the most popular forms of regression.
ML Classification – Decision trees or random forests learn from historical data and create decision pathways to classify outcomes. Support Vector Machines (SVM) and Naïve Bayes classifier are also examples of classification, using decision boundaries or probability theory to classify outcomes.
Clustering – Popular with segmentation or basket analysis, clustering methods like k-means, association, and hierarchical clustering categorise similar data points together through variable relationships to develop distinct traits by group. This makes it easier to predict behaviour or outcomes specific to that cluster
Advanced ML – Here we are grouping together different methods like neural networks, deep learning, gradient boosting, etc. into one category. These approaches create complex pattern recognition to build predictive models. These methods can also go beyond prediction into prescriptive as well
Text Analytics – Natural Language Processing (NLP) or sentiment analysis are very common methods of ML for analysing unstructured text data to evaluate, summarise and create new outputs. Similar to Advanced ML, these methods are starting to be used for other types of AI-enabled analytics, like prescriptive (e.g., LLMs)
These methods are the foundation for common forms of Predictive Analytics like demand forecasting, churn prediction, risk modelling, sales forecasts, etc. Companies hear these things and want to jump to them immediately but don’t understand the fundamentals that go into them.
Because while you can make predictions, those predictions might not be right
Without clean data and strong descriptive or diagnostic foundational analytics, predictions become questionable at best and misleading at worst. Predictive analytics isn't about perfect forecasting—it's about using data intelligently to make better-informed decisions about the future.
Remember, the challenge isn't just technical—it's getting the business to trust and act on the predictions. Many organisations invest heavily in predictive models that end up unused because stakeholders don't understand or trust them.
So while the third level of analytics is appealing and can transform an organisation’s decision-making, it must be built on solid foundations and implemented with clear business use cases in mind.
Prescriptive Analytics
Welcome to Prescriptive Analytics, the fourth and final level (some talk about a fifth, but we won’t go there for now). Despite what many people think, prescriptive analytics is not AI, just like predictive analytics is not ML.
Instead of being about how things are done (e.g., via AI tools or ML algorithms), it is about the outcome of what the data product is providing. For this stage, analytics is about predicting what might happen and recommending actions to take. Essentially, this is about automating decisions while optimising the outcome, two human-centric activities.
As we get into examples of prescriptive analytics, we start to think about the art of the possible. But it is only possible if you have completed the previous stages. So building off some of the predictive tools we mention above, here are some real-life examples of Prescriptive Analytics:
Automated Investment Decisioning Engines – Algorithms that identify risks and opportunities to automate investment decisions
Recommendation Engines (e.g., Netflix) – Systems that suggest specific actions based on analysed patterns and predicted outcomes
Digital Twins – Virtual replicas of physical systems or processes. Teams can simulate scenarios to find optimal outcomes (e.g., like number of staff needed to work a shift or design of a supply chain process to reduce waste), automating the decision-making process
Banking Fraud Detection – Algorithms detect anomalies in transaction data to identify and flag potential fraudulent activity in real-time
Chatbots & Virtual Assistants – Conversational agents that assist users with tasks, answer queries, provide customer support, and recommend actions. (Note that not all chatbots out there are at this level, most are garbage)
Similar to predictive analytics, prescriptive analytics might suggest counterintuitive actions that goes against stakeholder intuition. The adoption and data/ AI literacy to get stakeholders to trust and act on these recommendations is crucial and requires significant change management.
One thing to note is that while AI is prevalent here, AI tools could actually be used across every level of analytics. Credit to Jon Cooke to pointing that out on one of my previous LinkedIn posts because it is important to decouple the tool from the outcome when discussing analytics.
So, while prescriptive analytics represents the pinnacle of the four analytical levels, not every decision needs this level of sophistication. Sometimes simpler analytics are more appropriate, effective and much easier to implement.
Building Your Analytics Capability Logically
My biggest pet peeve about Data Products and Analytics in general is how disconnected people view them.
Business teams, executives and even sometimes data teams (including analysts) decouple the different levels by use cases. For example, thinking that forecasting models should be built in isolation. Or that AI-enabled optimisation engines don’t need to first understand the ‘why’ that comes through diagnostic analytics.
Here's the thing – each level is not an independent stage but building blocks that support each other.
If you ever played Pokémon, it’s like trying to catch a Charizard in the wild rather than training your Charmander so that it eventually evolves into a Chameleon and then a Charizard (and for reference, it is impossible to find a Charizard in the wild).
In data, your descriptive analytics will provide the historical data and exploratory insights to help you evolve your thinking to the why. The diagnostic analytics of the why (coupled with the descriptive analytics) can help you create predictive models; otherwise, how can you forecast what will come if you don’t know why it happened in the past? Finally, prescriptive analytics requires you to bring it all together. You need to understand the why for your prediction engine to make the right decision, otherwise the prescriptive nature of the tool will lead you in the wrong direction.
So instead of jumping straight to a forecasting model, build in a logical manner. The usefulness of your outcome depends on it!
Next week, we will keep the analytical theme and bring it back to the value discussion! From the great mind of Mark De Jong (and in a Data Ecosystem first), we will have a co-written article about deciding on the right KPIs and standardising those within your organisation. This topic is hugely underrated because I find most organiations struggle with this task, which ends up undermining the efforts of their entire data team. Remember, your data products and analytics are only helpful if aligned with the right KPIs.
Until then, have a great Sunday and thanks for reading!
Thanks for the read! Comment below and share the newsletter/ issue if you think it is relevant! Feel free to also follow me on LinkedIn (very active) or Medium (increasingly active). See you amazing folks next week!
Where would you fit the operational analytics in this scale? Descriptive?
Operational analytics as defined in Joe Reis's book:
Operational analytics focuses on the fine-grained details of operations, promoting actions that a user of the reports can act upon immediately. Operational analytics could be a live view of inventory or real-time dashboarding of website or application health. In this case, data is consumed in real time, either directly from a source system or from a streaming data pipeline. The types of insights in operational analytics differ from traditional BI since operational analytics is focused on the present and doesn’t necessarily concern historical trends.