Augmented BI – Combining Machine Learning (Root Cause Analysis) with Human Intuition
This is the 2nd and last part of blog series on Augmented BI.
Part 2: The root cause of the problem – “Find Out Why, Fast”
“Why did that happen?” is a disarmingly simple question to ask, but often fiendishly difficult to answer. On most BI tools, it takes hours, if not days or even weeks of analysis across hundreds of potential reasons. In this post, Alvin Chan, AVORA’s Head of Product, lays out how to get to the root cause in minutes using machine learning.
In our last post we looked at anomaly detection and how this forms the basis of the Smart Alerts feature in AVORA. Alerts are a fundamental component in helping flag up data variances in real-time. But what really matters is how speedily the user can respond to and interpret that alert. Is there a way machine learning could assist here?
It’s the cause of that alert that is the fundamental issue and its likely impact on the business. It’s here where Root Cause Analysis comes into its own. Sifting through the likely indicators that relate to that change and prioritising the relevance of each as a contributing factor.
Let’s’ say a digital media agency is experiencing an improved conversion rate due to a more compelling advertising message. How do you isolate the part of the business chain responsible for that? Root Cause Analysis provides a truly deep dive into the data. It doesn’t simply associate a spike in conversions with compelling messaging. It looks at the underlying drivers behind the increased conversions, recognising the contribution of business drivers such as:
Impressions delivered (captured as Cost-per-Mille, or Impressions per £/$ spent)
Clicks (presented as as Click-Through Rate)
Orders placed (captured as Conversion Rate)
Waterfall chart with top 10 contributors
This allows us to deliver a top-level view for what’s changed, in minutes. Users compare performance day-on-day, week-on-week, month-on-month, or any other comparative period. Clearly seeing how each underlying driver impacts the business end goal.
Behind this, the Root Cause algorithm has leveraged AVORA’s bespoke machine learning. It identifies and surfaces the most relevant reasons behind any change of a KPI. It provides this in the form of a curated list of dimensions. This is at the heart of the augmented analytics approach, designed to focus the user’s attention on exactly the right parts of the business.
Table view with the most relevant list of dimensions in Root Cause Analysis
Understanding of these granular reasons empowers users to affect change quickly within their business unit. Providing a view on exactly how much impact it has had on the commercial goal. As an example, it might have been a certain marketing channel, over a set of products, that delivered lower conversions this week, caused by a higher click-through rate. Equipped with this knowledge, our user can quickly act to recover lost revenue going forward in real-time.
AVORA’s bespoke machine learning…identifies and surfaces the most relevant reasons behind any change of a KPI.
By leveraging the business logic and knowledge from users, Root Cause is configurable to any business process and its underlying metrics. This only needs to be entered into the system once for everyone to benefit. It can then provide answers across the entire business, enabling users to focus on what really matters. In the end, a decision on what action to take rests, quite rightly, with the user.
Users are empowered to make decisions to improve performance in a meaningful way, backed by data evidence. Augmented analytics tools are not there to automate away your job, but to help you make better decisions.
AVORA has pioneered the development of Root Cause Analysis and is the only analytics provider offering this level of drill down cause/effect analysis powered by machine learning.
Answers to critical business questions are surfaced in minutes by our algorithms, leveraging machine learning to remove tedious, manual analysis. All so you can get back to running the business!