In any system, there is an expectation of how things should perform. When this is measured and performance changes, it’s natural for humans to ask the inquisitive question of “Why did that happen?”
Root Cause Analysis, in a nutshell, is the discipline of answering this question in a structured way.
The approach to this could well vary depending on the scenario – for example:
My car doesn’t start – the mechanic performs root cause analysis and tries to identify the faulty component(s) for replacement
My sales are lower this week vs. Last week – an analyst performs root cause analysis to identify the reasons behind the drop, and take business action to bring it in line with expectations
Approaches to Root Cause Analysis
There are a myriad of ways to approach Root Cause Analysis – we’ve assessed some of the most popular:
5 Whys – A human-based approach, on asking questions based on the answer of the previous question. Used in the Toyota Production System. Developed by Sakichi Toyoda.
Cause and effect fishbone diagram/ Fault tree analysis – again, human-based – someone has to comb through and perform this
Event analysis – can be machine based, if appropriately designed – what other events coincided with/preceded the event for which Root Cause Analysis is performed? This works best when the event being investigated is highly unusual
Change analysis – can be machine based, if appropriately designed – recognising that many things could have caused a change, and showing those that have the most relevant impact
Defining The Problem for Business Analytics
It’s important to ask the right question – without the correct framing, it is difficult to arrive at the right answer!
Depending on the scenario for root cause analysis, the question asked may be different. Using our previous examples:
“My car doesn’t start” – There is a clear baseline (my car was working before), and a clear desired end-state (my car should start) – the question is “Why is my car not starting?”
“My sales are lower this week vs. Last week” – the question is often “Why is sales lower than before” – the desired end state is less well defined!
All of this assumes that the root causes are present and measurable/identifiable – this is not always the case!
In business analytics, there are some common questions:
What is unusual about a particular subset of data?
Example: I have X, why?
For the same metric being tracked, why has performance changed from one day/week/month to the next?
To answer questions in an evidence-based way, some data must be available for analysis. This can be done on an ad-hoc basis, as questions arise, or rely on data that is already measured and stored relating to a business function. Some data common tracked for these purposes are:
Not (easily/consistently) Measurable Factors (Latent Variables) – e.g. competitor strategies, a customer’s likelihood to churn, wider geo-political risk/factors
Whilst there is likely a non-zero impact due to the second group, they are not well defined and should be used in a contextual way – focusing on measurable factors also has the benefit that because they are tracked, they are usually highly actionable!
Once the possible factors have been split up, it’s then the responsibility of the analyst (or investigative system) to look through all possible factors, and assess whether they are relevant to the change in performance.
Identifying Root Causes
A range of techniques help to perform this assessment:
Find outliers – (these may not be relevant, as they contribute little to the overall impact)
Correlated – what happened at the same time/immediately preceding this change? This is particularly useful when systems are disparate/siloed, and there are no common linkages, with the exception of time
Impact-based analysis – when compared against an earlier date range, which factors have had the greatest impact in terms of explaining the change in performance? For example, sales is up by $4M, and the category “Health & Beauty” saw an increase in $3M, this is likely to be highly relevant!
Beyond this level of analysis, one might go deeper and examine the metrics that drove that change – e.g. the increase is sales is associated with a 20% drop in average order value (due to discounting), but more than offset by a 50% jump in number of orders! Capturing these relationships goes beyond the straightforward answers and allow users to go one level deeper, but needs the right supporting framework.
There’s a number of ways to implement Root Cause Analysis into your business – it often proceeds in this order:
Analysts answering ad-hoc queries with easily available tools (Excel, SQL), taking significant amounts of time to answer a question
Dashboards that allow users to dig into and investigate pre-determined views of businesses
Dedicated tools that integrate with granular datasets and enables business users and analysts alike to ask and answer questions
Regardless of which way it’s put in place, here are some important considerations:
With your implementation, are you able to ask questions at any time? Often, business problems arise outside of core working hours
There’s a human component in the root cause analysis journey – they bring value in domain knowledge and context. Should they be spending the majority of their time on number-crunching?
Time-to-insight is important – knowing the broad reasons behind a change in performance is usually sufficient to trigger some corrective actions – this is where automated systems can really help, since they’re not reliant on a human to get started
If you have a complex business, with many entities, clients, campaigns, etc – how do you know what question to ask? Will the system do this on your behalf, or will you be again dependent on the human-element to look for problems?
How can you share the insights?
Importantly – consider how you can act off these insights!
I hope that’s been a useful introduction to the concepts and techniques behind root cause analysis, and the most important considerations for implementing a system, whichever you choose! A few parting thoughts:
Make sure you ask the right question, based on the (measurable) data you have
Pick the right tool that reflects the level of human involvement you want
Rapidly identify measurable factors and receive a quick guide to the root cause – always reserve the option to go one level deeper
Once you have your insight – share it with the relevant people and act on it!
Do you need Root Cause Analysis?
Discover in seconds which factors lead to unexpected changes, the impact of each factor driving the change and the metric forecast.