Cognitive Computing Improves RCA

Jon Boisoneau, VP Products

When I saw IBM’s cognitive computing system, Watson, beat the Jeopardy champions in 2011, I started wondering about the possibilities this exciting new technology could bring to root cause analysis (RCA).  I’ve been involved in RCA software development for over a decade and know that the semantics of language can be a hurdle when looking for systemic causes and solutions to problems across an organization.  Sometimes people describe the same problem in different ways.  That difference is a barrier when looking for incident patterns and trying to learn from past mistakes.  Could Watson help bridge the gap?  When IBM opened the Watson API to 3rd party developers earlier this year, we decided to find out.  

Causelink powered by IBM WatsonThe timing couldn’t have been better for our software team.  We had just released a major update to Causelink Enterprise and were ready for a new challenge.  We organized a 3-week development sprint with the following goals:  1) Learn the capabilities of the Watson toolset (along with their newly acquired Alchemy API).  2) Build a prototype that adds value to our RCA software, Causelink Enterprise.  3) Obtain feedback from a select group of customers.  

Based upon the responses in our Causelink/IBM Watson focus groups, the prototype was a success.  Although additional refinements are being made, we can now leverage Watson's powerful capabilities to analyze data inside Causelink Enterprise and uncover patterns in the data that were otherwise hidden, or would have taken many hours of expert analysis to uncover.

Here is what IBM Watson can do inside of Causelink Enterprise:

1. Create potential causes from problem descriptions

In this feature you can use voice to text to describe the problem you want to solve.  Alternatively, copy/paste the description from an existing document.  When the description is complete, click "Extract Causes" and Watson will parse your description to find potential causes.  Once extracted, simply drag and drop causes onto the cause and effect chart.  While these causes aren’t always perfect, they are easily edited.  We estimate that this feature saves 1-3 hours per investigation.  A future version will allow users to define these Watson-generated records as causes, evidence, solutions, notes, or actions and to edit them earlier in the process.   

2. Analyze potential solutions

After adding solutions to an RCA record, click "Analyze Solutions" and a Watson-powered evaluation tool appears.  There are several options available that help you visualize solution data.  This tool is most helpful when you are evaluating six or more solutions because you see all the data at once, plotted on a graph displaying your most important criteria.  Report criteria includes cost, term, effectiveness, ease of implementation, and return on investment.    This helps identify the most valuable solutions identified and available to the team which will result in the highest ROI for the RCA.

3. Run Reports

The “Related RCAs” report parses all of your RCA data and looks for similarities, based upon keywords and Watson-derived concepts.  This report helps you see the similarities in problems being addressed by other departments or divisions in the organization.  Most exciting, since Watson excels at finding patterns in unstructured data, the minimum data requirement for this report is a problem description.  This is a key feature that will lead to proactive solutions because it points out the systemic causes resident across the organization that may not be apparent when people are working at an individual RCA level.

We also created a new report (codename “Cause Solver”), which Watson builds by finding all causes that do not have a solution, then recommending solution concepts based upon similar cause/solution pairs in the data.  Admittedly, at present, this report’s value is a function of the data set. However, the proof of concept is there and we will improve upon the report over time by suggesting potential causes and solutions in-line where appropriate.  This feature will enable even novice RCA analysts to benefit from the past wisdom of subject matter experts who may no longer be available to assist with the RCA.

So where do we go from here?  Feedback from end users has helped us understand the changes necessary to bring this innovative feature to market.  We are also investigating the ability for Watson to span multiple languages in one report, and the ability to project future risks based upon past investigation data.  Our development team will be working on a Beta version of the product later this year.  If you would like to be involved in Beta testing, contact us and we’ll be happy to include you.

Alchemy Logo IBM Logo