Transcript Natural Language

Virtual Health Assistants in Specialty Pharmacy
Thomas Morrow MD
Author of
Tomorrow’s Medicine
Managed Care
The Maasai Warrior in Kenya has better access to
data and communication than President Reagan!
Disclaimer
Dr. Morrow:
• is a full time employee of of a large biotech company but no company
products will be discussed during this presentation. The opinions
expressed during this discussion are his alone and do not reflect the
opinions of his full time employer.
• has created this talk from his research concerning VHA for an article
published earlier this year in Managed Care and again in Forbes
Natural Language Processing + Conversational Interface:
Keys to Artificial Intelligence
Artificial Intelligence
The ability of a computer or other machine to perform actions
thought to require intelligence.
Natural Language
Human written or spoken language as opposed to a computer
language.
Top Companies:
–
–
–
–
Next IT: focusing on the patient activation
Nuance/VirtuOz: focusing on medical records
Creative Virtual: no obvious medical focus
IBM Watson: No commercially available products, Oncology/Cardiac
physician decision support
Medical Decisions Occur in a Minute by Minute Basis
• Patients spend a few hours per year with their
physician for a “specialty pharmacy” disorder.
• They spend 5000 hours per year making literally
thousands of decisions that affect their health
• They need day to day decision support
• Natural language- driven virtual assistants can fill this
need
Without a Relationship, There is no Influence
We know that advice and guidance is much more
influential coming from someone with whom you
have a trusting relationship
The Virtual Health Assistant:
Technology that extends the patient/provider relationship
•
•
•
•
•
A New Definition of High Touch
Fulfillment Needs
Patient Education
Disease Treatment Management Programs
A New Level of Data Collection: Virtual Head-to-Head Trials
Reach and Influence Patients the other 5000 hrs.
Virtual health assistants go far
beyond reminders, and have
been proven to make an
emotional, social, and
visual connection with patients
Automated system choices available to an organization
Characteristic
Input
Traditional IVR
Voice and DTMF tones input via
keypad
Traditional Web Chatbot
Text
Next IT’s Multimodal Virtual Assistant
Text, Talk, Tap
Language Processing
Decision Tree with limited FAQ based
interactions
Searchable FAQ
Stochastic NLP engine with an intent based language model
Logic Model
Linear with branches
Typically 4x4 or 5x4 model
Silo singular answer based
Human Emulation – The model is a combination of decision tree, FAQ
and intent model which also incorporates context like a human does.
Channels
Phone
Web
Phone, web, SMS Text, mobile, kiosk, social media
Output
Voice
Text and sometimes voice
Voice, Text, Navigation often simultaneously based on channel
Contextual Awareness
Minimal: Based upon account, user
profile
Minimal Based upon account, user
profile
or General Answers
Page Awareness
Conversational Awareness
User Profile
Every question is taken in the context of the entire conversation as well
as other data sources
Proactive Engagement
None
None
Multiple options, dynamic, personalized
language model size
4x4 to 5x4
200-300 in a basic FAQ mode
10s of thousands of intents in a single model
Where placed in
organization
IVR’s can be set up to support specific
tasks in an organization
Usually isolated to a section of a web
site, not the entire site.
Across the entire web site/portal and across multiple channels as well.
Breadth/ Depth
minimal on both since a human
listening to the phone can only
remember a small number of options
Typically limited to either broad OR
deep but not both
Able to cover a very broad domain of knowledge while also having a
great deal of depth where applicable.
End Point
Simple task
General Answer to simple questions
Truly conversational
Cost to Build
How built
Future enhancements
$$
Voice Recognition
Limits reached
$$
Search Based
Limits almost reached
$$$
Chat Recognition
Virtually infinite
Overall Long Term Value
$
$$
$$$$$
Monthly Operational Cost
to Organization
$$$$$
(based on need to divert calls to Live
Agent)
$$$$$
(based on need to divert calls to Live
Agent)
$
(Virtually all calls can be handled by Virtual Agent limiting the number of
live agents needed)