How Augmented Analytics helps transform data to intelligence: A technical analysis
by Ramesh Panuganty, Founder & CEO
How Augmented Analytics helps transform data to intelligence: A technical analysis
by Ramesh Panuganty, Founder & CEO
Augmented analytics is the automated way by which business users receive intelligence from data without explicitly asking a question. This expands users’ abilities to interact with data by search-driven as well as search-less technology, powered by machine learning.
From philosopher Carl Jung’s perspective, “To ask the right question is already half the solution to a problem.” But in oceans of enterprise data, business users may not always know what they don’t know. Humans can’t predict or ask about hidden trends and anomalies that are only revealed with AI-driven analysis. Augmented analytics discovers contextual intelligence without waiting for questions.
Augmented analytics: An introduction
Augmented analytics revolves around the lifecycle of data to intelligence – all the way from understanding the data and enabling users to converse with data (search-driven), to identifying signals that influence the context of data (search-less), to identifying insights (ML model-driven findings), and to presenting them in a way business users can understand immediately. In this case, users can be any decision maker at an organization, from a procurement manager to a store clerk operating inventory. Augmented analytics benefits everyone, regardless of skill level or business function.
To put augmented analytics into context, let’s examine this example of credit card statements: Imagine when your credit card statement arrives, it delivers these additional insights along with the usual details:
- Your spend on restaurants that month is anomalous as compared to your past 30 months of data
- The reason for anomalously low spending in the restaurant category is a monthly coupon that you used for every meal
- Your savings resulted in 8% lower monthly spend, which is the equivalent of 3 meals
I emphasize that these data points are NOT based on what everyone else is doing but just based on what you are doing. Today, I don’t know of any credit card companies who can recommend such detailed insights.
Now let me explain this with an enterprise example. A retail company’s procurement officer asks, “How many lawn chair orders this year?” In the answer, he gets the specific order quantity for lawn chairs in his territory (say, US West Coast), along with the intelligence that:
- 32% of lawn chairs were sold in South California
- 21% of total lawn chairs were sold accompanied by a table
- The coupon BOGO drove up sales by 15% in the last 2 weeks
This would be a good example of an augmented analytical system that helps enterprises transform complex raw data into actionable intelligence.
This way, with accelerated “time to insight,” business users can make quick actions and confident decisions.
Reduced time to insight is the greatest differentiator that augmented analytics can offer to enterprises. In a highly competitive business landscape, customer preferences are influenced by a variety of factors. Reduced time to insight helps enterprises identify and track these preferences faster and fulfill them with targeted personalized offerings. Augmented analytics capabilities help increase productivity, speed up decision making, and help businesses reach their goals more quickly.
The role of machine learning in augmented analytics – a technical analysis
Democratization of both BI and AI are outcomes of augmented analytics, and machine learning is at the center of every augmented analytics benefit. Let’s visualize the workflow of both the front-end path (as seen by a user) and the back-end path (may or may not be seen by the user).
First, the data catalog is generated by understanding the vocabulary of the industry and organization-specific semantics, the use cases of data, and the continuous learning of how users refer to different sets of data.
Next, the natural language search engine uses machine learning to learn vocabulary and grammar to provide personalized results to users.
Then, the insight detector uses machine learning to identify correlations across attributes, causation of correlations, and prediction models of metrics across attributes. This learning needs to happen regularly on the back end with or without users’ knowledge. These correlations may be used for different purposes to improve the identification of data intelligence for every user context.
Lastly, the narration engine uses machine learning to improve upon text narrations and generate text in the corresponding language. There are only a few solution vendors who offer natural language generation.
Augmented Analytics vs. Business Intelligence
BI is passé. BI addresses only the ‘what’ of a business scenario and does not help improve iterative decisions – where each decision is refined and informed by updated data.
Augmented analytics not only answers the ‘what,’ but more importantly also answers ‘why’ and ‘how.’ It does not require users to start with a hypothesis, because insights are discovered automatically. However, augmented analytics can only answer ‘why’ and ‘how’ if implemented properly and end-to-end in the analytics workflow. Nowadays, every BI vendor claims that they are an augmented analytics vendor, while all they do is find anomalies. Be careful not to fall into this trap, because using the wrong technology could result in missing out on valuable intelligence.
Where BI stops at finding the ‘what,’ augmented analytics goes beyond and discovers the ‘why’ and ‘how’ – translating insights into actionable strategies in real time.
6 aspects to better understand augmented analytics
Now that we’ve covered the differences between BI and Augmented Analytics, let's take a closer look at some aspects of Augmented Analytics.
1. Too many insights are useless
No business user can consume dozens or even more than a few insights at a time. Your augmented intelligence platform should have a way to de-duplicate similar insights, rank them on a scale of business applicability and show only the most important insights. If you receive 30 insights per hour, you can’t act on them in a meaningful way. The most valuable insight would be the one that’s actionable and results in the most dollar impact for you.
Thus, the augmented platform should not only find the answer, but also show the insight with the highest business impact.
2. Anomalies are not everything, and need to be curated well
Intelligence is not just about anomalies. Anomalies are completely useless unless every other possible pattern in the data is found. Also, anomalies must be filtered by seasonality and trends to avoid misleading the audience. For example, in the context of yearly sales, an anomaly stating that Apple achieved its highest sales in November and December is not useful, because Apple always releases their newest models in late September every year. A good anomaly always subtracts the impacts of seasonality and trends. On the other hand, a useful anomaly would be one that shows that sales peaked in May due to a campaign for long-term customers.
3. Raw anomalies are useless
Sometimes known information masquerades as anomalies, adding very little value to insights and actions. Imagine you’re the Northeast regional manager for Starbucks. When you’re inquiring about last month’s sales, the system finds an anomaly in sales of the product ‘caramel macchiato.’ You might already be aware that it is the most-sold product in every other Starbucks as well – it’s not useful information! What would have been useful is excluding well-known anomalies and identifying a different class of anomaly, such as how ‘Java Chip Frappuccino’ sales have been going up in Vermont for the past 3 months. Business data must be analyzed over a time series that identifies consistent patterns.
4. Automated driver identification
The objective of a driver identification process is to find attributes or variables that predict the metric of interest. For a given metric, the augmented analytics tool needs to identify associated drivers using the entire dataset and a combination of classical statistical and advanced machine learning models.
If the number of attributes for a given model is large, the analytics platform must first identify the attributes that are correlated with the metric. The specific correlation method to use would depend upon whether the attribute and the metric are continuous-continuous, continuous-categorical, or categorical-categorical. These significantly correlated attributes are then fed into a machine learning algorithm to predict the metric of interest. If the accuracy of prediction is high, then we should use explainable machine learning methods to rank the importance of each attribute in predicting the metric. The top-ranked attributes are identified as the drivers for the metric.
The identified drivers will be used in the subsequent ‘why’ analyses of the metric.
5. Insights are not useful unless you know why they occur
The augmented intelligence platform should not only answer ‘what’ happened, but also recommend ‘why’ and ‘how to act’ in the given context.
Advanced analysis of why a business insight happened and how the user should act on it makes the insight actionable and result-oriented. It is extremely difficult to derive actionable learnings from traditional analysis and visualizations due to increasing data complexities and data volumes. Augmented platforms should provide seamless and complete understanding of What is happening, Why it’s happening, and How to act on the outcomes.
6. Insights might not always arrive
Depending on the data and context, there may not be any insights at all. You need to be very well aware of this scenario, but the analytics platform should be able to tell you why it could not find any insights. These are not false recommendations based on other user patterns, like Netflix movie recommendations that are not based on your watch history. You should expect high quality, concise insights – therefore “no insights” is a likely scenario.
Why should you choose augmented analytics?
If your BI tool tells you that product A performing poorly and leaves you to analyze why the product is lagging, then your BI tool is not fit to equip you with advanced intelligence. Here’s why choosing an augmented analytics platform is essential:
- Thorough and end-to-end analytics capabilities powered by machine learning
- Search-driven and search-less insight discovery and generation
- Detailed ‘what,’ ‘why,’ and 'how' of insights for better understanding and faster decisions
- Actionable insights personalized for you and your business context
- Curated intelligence to help focus on the most important aspects and ignore the known, less usable ones
- Democratization of data and insights across enterprise and employees for iterative decisioning
- Reduced time-to-insight, resulting in increased productivity and agile operations
To learn more about the benefits of augmented analytics, see What is Augmented Analytics? The Ultimate Guide to the Next Generation of BI.
How MachEye delivers the best in augmented analytics
MachEye is the only augmented analytics platform that enables every business user to understand “What” happened in the past, “Why” business metrics are changing, and “How” to take an action based on insights.
In addition to answering natural language search queries, MachEye presents anomalies, analogies, clusters, trends, growth rate and other advanced insights related to your query. Furthermore, we anticipate your needs and provide recommendations and actionable insights automatically. Our search-less insight discovery presents critical business headlines as they appear in data, so business users never miss important opportunities.
MachEye was recently recognized by both Gartner and Forrester as a leading Augmented Analytics and BI Platform. We deliver augmented analytics by offering intelligent search, actionable insights, and interactive audio-visual stories on our cloud-native platform. Get started with the free trial today!