MicroStrategy World: AI Best Practices and Real-World Examples
Blog
  • Topics
  • Industries
  • Partners

Explore

Topics

Industries

Partners

MicroStrategy World: AI Best Practices and Real-World Examples

Christine Livingston, Chief Strategy for AI at Perficient spoke on AI best practices and gave some real world examples on how it would work.

Where are analytics heading?

Notice that while you spend a lot of time early on with descriptive reports, the higher up the maturity curve you go, the more you need tools that are a key part of the artificial intelligence continuum.

What is AI?

  1. Consumer products like Alexa, Cortana, etc.  They aren’t intelligent systems. They are natural language processing and some automation
  2. Embedded AI: Google uses it for gmail auto replies. It’s 40% of messages.  But that doesn’t work well at work.
  3. Platforms: This is where they are focused on democratizing AI. They want to give key functions and features to all companies.

 

Core Tenets of Artificial Intelligence

Machine Learning: train a machine to do a job

Natural language processing: seeks to understand intent

Predictive analytics: progression of the analytics continuum and quite often paired with machine learning

Cognitive: put machine learning, NLP, and predictive together and add some reasoning capabilities.  This is closer to true AI. Intelligent Virtual Agents are part of this

Signal Services: It’s speech recognition, tone analysis, sentiment analysis

State of AI Deployment

  1. 4% have interest
  2. 21% are experimenting
  3. 25% in long term planning
  4. 35% on the radar only
  5. 14% no interest

 

Quote: Information is a source of learning. But unless it is organized, processed, , and available to the right people in a format for decision making, it is a burden and not a benefit — William Pollard circa 1870

 

Best Practices

  1. Outline a roadmap
    1. establish governance
    2. Figure out what you want and place the priorities
    3. This is both model governance and people governance
  2. Don’t Boil the ocean
    1. Start Small
    2. Prove value
  3. Gargage In, Garbage out
    1. Identify your dataL figure out where it lives. Figure out if it’s ready for ingestion
    2. Cleanse it
  4. Iterate and learn
    1. Allow sufficient stabilization time
  5. One size does not fit all
    1. Evaluate capabilities by use case
    2. Look at best in breed capabilities
      1. best image recognition, best NLP, etc.
  6. Optimize your training efforts
    1. Standardize tooling by capability
      1. Highly paid employees should optimize their time and should use a common set of standards to make their time better spent
    2. Define a common framework

Multiple Cognitive Engines Trap

You can back yourself into a corner by using more than one cognitive engine.  You only have so much bandwidth to train the engine. Don’t try to train too many cognitive engines.

AI Strategy Made Easy

Case Studies

Tri-Health for patient readmissions

Needed to understand and predict what brings patients back to the hospital. The first model brought only 41% accuracy….worse than flipping a coin.  They then used statistical analysis to get to 73% which was better. They finally did a POC which also included unstructured data.  They got to 93% accuracy

Baycare: patient population identification and optimization

Baycare was looking to identify care management approaches to their patient population. This pushed the need to segment and prioritize patients.  They want to correctly identify the patients in fit the disease state.  They took the data and were able to pull the care management team from studying records to taking the patients and creating a better care plan.  Results: 41% improvement in identification accuracy (over humans) This also drove a 75% decrease in record review time.

Best practice: Governance and strategy. Get the team prepared for AI. Get executive leadership involved as owners

Virtual Agent

An insurance company wanted to provide information to care providers looking to verify benefits.  This was a huge volume of calls. They wanted a virtual agent to handle a lot of these calls. They created an interactive query system and took away the dreaded IVR.  This approach decreased live agent requests by 63%.  Average call time also decreased from 8 minutes to 3 minutes.

Challenges: This was a very technical problem. They had to integrate to IVR and ensure that they had authentication and authorization from the callers complete.

Bottom Line: When you follow best practice and identify the right use cases, AI can provide large and verifiable value.

Quote: Learning and innovation are designed to go hand in hand.  The arrogance of success is to think that what you did yesterday will be sufficient for tomorrow — William Pollard

 

 

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.

Subscribe to the Weekly Blog Digest:

Sign Up