4 ways ai can help your small business
TRANSCRIPT
4 Ways to Enhance Your
Business With AI
Keita BroadwaterAugust 27, 201718th TACC-NC Business Workshop
Objectives
• Understand Artificial Intelligence (AI) and Machine Learning (ML)• Learn Pre-requisites for Machine Learning• Understand Practical Applications for your Business
About MeOperations Leader with Data Science Skillset15+ years in high tech
Currently: • CFO of PV Tech, Inc• Data Science Consultant
• BS, Physics – Florida A&M Univ.• MS/PhD, Mechanical Engineering – Univ. of Md• MBA – Cornell Univ.
Loves: Travel, Astronomy, Running, IoT, & Jazz
Email: [email protected]: @keitabr
What is AI & Machine Learning ?
The science of making machines replicate human intelligence.
Artificial Intelligence
A pillar of AI, where algorithms allow machines to learn from data.
Machine Learning
One powerful subset of Machine Learning. An Extended Neural Network. A Neural Network is an algorithm made to mimic human brain functions.
Deep Learning
Heavy SW Development Soft Software Development/PaaS VendorsHow is it Done?
What AI is not:• Not automation per se• Not a Set of Rules
- Ground-up development - Using coding languages (such as
scala, python) and databases (SQL, no-SQL)
- Lighter SW development- Takes advantage of existing packages and platforms (IBM Watson/Bluemix, AWS)
- Minimal or No coding- User friendly products that
require various levels of configuration and input data
Examples
1) AlphaGo – 1st computer program to defeat a Go world champion, Lee Sedol• Innovative, Surprising Moves & Strategies• Deep Neural Network
2) Real-time Object Recognition– Using Tensorflow: Google’s open-source ML library• Innovative, Surprising Moves & Strategies• Neural Network• Trained with images of large variation in
scale, pose, and lighting
Personal Example
Lorem Ipsum is simply dummy
text.
03
Home Security System
Father: 91%
Mother: 9%
Son: 20%
Roomba: 1%
INTRUDER: 2%
Problem: Late Night Intruder
Solution: System that uses footsteps to: • identify people • detect intruders
Machine Learning Tool: Support Vector Machine (SVM); Classification
Data: Sensor Readings
02
03
01
Pre-Requisites for Machine Learning Adoption
Data Collection
01
Security & Privacy
02Online
Presence
03
Data Collection• Size of home grown data• Use someone else’s data
• Garbage-in Garbage-out• “Clean” data
Online Presence• Website
• Social Media• Interactivity with Customers
Consideration for Security and Privacy• If using person-identifiable data
• Anonymize• Terms-of-Use Agreements
• Notifications
Meet Katy: Real Estate Agent
AI
01
• Average 15 Clients/month• Growing Practice• Part of Agency• Sales and Revenue can be Uncertain
Online Presence• Website• Posts once a month on Facebook• Posts Weekly on Weixin
Her Data• List of Clients
• Basic Demographic, Financial Details
• Access to MLS Data• Housing Inventory, Sales History
• Katy’s Transactional History
Some Duties• Time spent generating leads• Writing Monthly Newsletter to Clients• Time Spent Planning of Home Viewings
Content Generation: Overview
Value: Productivity; Save Time
Use cases include:• Financial Updates• Sales Reports• Marketing Briefs• Personalized Reports
Core Technology Involved: NLG: Natural Language Generation• Analyze Data• Interpret Data• Identify the Most Significant Parts• Generate Written Reports in Human Language
Data Used To Train Models: Vendor Data
Turn Spreadsheet Data….. …Into Written Reports
Content Generation: NLGNatural Language Generation is about:- Producing machine written text that
• Has High Quality• Is Understandable • Is Easy to Read
Advantages• Text is the preferred Medium• Important Information Changes• Source data is organized• Variation in the output is required• Automation is an advantage
Alternatives• Fixed Templates• Templates with Variables• Graphics
Text Planning
•Convert specific pieces of data into discrete phrases
Sentence Planning
•Combining the phrases using rules of grammar
Linguistic Realization
•Ensuring that complete text makes sense to the human eye
train_arrival = 20:00:00
“the arrival time is 8PM”
Content Generation: KatyRecent Home SalesHere is the inventory report for July 2017.
Visit: https://katyliu.com/homes.htm to take a quick look at recent home sales.
Real Estate TrendsThe real estate trends we've seen for the last four years are continuing. Both San Mateo County and Santa Clara County show average sales prices are staying higher than list prices.
Price trends are little changed. The peak in price per sq. ft. for the 3rd quarter of 2017 is based on closed escrows for July sales (not all sales have closed escrow), and is likely to change a little.
The California Association of Realtors in a July 24 release stated "In C.A.R.'s newest market indicator of future price appreciation ... indicates that price growth will continue to accelerate, potentially back into double digit territory, as it reached its highest level since 2013." (homes are selling faster than homes are offered for sale)
Advantages for Katy:• Save time prepping monthly
reports• Send Personalized Reports
Depending on Preferences of Client
Content Generation: Implementation
Process:1) Data preparation and upload
2) Editorial Training of the AI• Create the basic outline of desired reports• Connect data to specific parts of text• Prepare Templates and Conditional Formatting
3) Setup automation of data upload and report generation
Company Pricing Website
AX Semantics Starting at $256/user/mth https://cloud.ax-semantics.com/
Arria NLG Unknown, Free Demo http://www.articulatorlite.com/
Yseop Unknown, Free Demo https://savvy.yseop.com/
Chatbots: Overview
Value: Handle customers facing tasks 24/7• Pre-sales• Customer Support• Sales
Technology: NLG: Natural Language GenerationNLP: Natural Language Processing
Data Used to Train Model:Vendor Data
Customers….. …Interact with Machine Agent …That carries out a task
Place Order
Create Service Ticket
Set Appointment
What is a Chatbot?
A machine that can:• Give coherent and meaningful answers• Conduct a conversation with a human
Customer Chatbot: Katy
Advantages for Katy:• Multiple customers can be
managed simultaneously• Customer has access to basic
information on demand• Customer preferences and
sentiment can be analyzed
Case 1: 1. Customer browses
property on website2. Chatbot offers more
information
Case 2:1. Customer is given
chatbot phone number
2. Customer dialogues with chatbot on mobile device when looking at homes
Chatbot: Implementation
Company Pricing Website
Promero Starting at $25/mth, Free Demo https://www.promero.com/sms-webchat-voice-email-bots/
Hutoma 0.0025/API call, Free Demo https://www.hutoma.ai/
Motion Starting at $15/mth, Free Demo https://www.motion.ai
Process• Create Customer Conversation Flow
• Chosen from Template• Link to Data• Link to Platform (facebook messenger,
website, SMS etc)• Link to other functions
• e.g., API to order from a website
Sales Forecasting: Overview
Value: • Greater Operational Efficiency• More Accurate Forecasts• Less time spent in forecasting
Technology: Time-Series Regression
Data Used to Train Models:- User’s Historical Data- Vendor’s Comparative Data
Turn Historical Data….. …Into Forecasts & Predictions
Sales Forecasting : Underlying TechTime Series Regression:
• Linear Regression of a time series• Less Flexible Model• Uses Entire dataset to fit model
• Auto-regressive models (ARIMA)• More Flexible• Uses dataset + takes into account
each previous value when making the fit/forecast
Sales Forecasting : KatyInput: Sales Record of Agent from MLS
Input: Details of Sales Transactions from MLS
Advantages for Katy:• Speed-up & Enhance Katy’s
Forecasting Process• Katy can adjust marketing plan• Katy can adjust expenses
Output: 12 Month Sales Forecast
Sales Forecasting : Implementation
Process:• Have CRM Setup -Or- Have Data in Spreadsheet
• Clean up the data, if necessary• Upload spreadsheet data
Company Pricing Website
IBM Watson Analytics Free Tier if Data Size is Low https://www.ibm.com/watson-analytics
Salesforce Einstein Multiple Packages/Add Ons https://www.salesforce.com/products/sales-cloud/features/sales-cloud-einstein/
Sales Temperature $25/location/month http://www.salestemperature.com/
Contract Review: Overview
Value: Streamline Contract Reviews; Enhance Due Diligence Process
• Accurately Highlight Risks and Problem Areas
• Reduce Legal Costs
AI Technologies: NLP: Natural Language ProcessingRegressive ML ModelsClassification ML Models
Data Used to Train Models:Vendor Data
Upload Contract into System… …Algorithm Analyzes and:
• Identifies Key Clauses• Identifies Areas of Risk• Highlights these for Review• Provide Visualization of
information in Contract
Contract Review: Case StudyMaster Services Agreements (MSAs): Contacts between service providers and customers
Contract Negotiation:• Duration: 3-9 months
• Inspection Revision Approvals
• Contracts Size:• 25-75 Pages• 10 – 20 Sections
• Key Bottleneck of Process• Methodical Reading and Manual Parsing • Prepping Comments for Review Meetings
• Risks• Overlooking key terms and conditions
• What are the key tasks and duties• Missing a key stakeholder’s input
Document Parsing: Implementation
Company Pricing Website
Beagle ~$100/mth, Free 7 day trial http://beagle.ai/
Kira Unknown; Demo Offered https://kirasystems.com/
Legal Robot Beta Product https://www.legalrobot.com/
Process• Upload Contracts• Assist in Training AI
• Identify Areas/Clauses of Concern• Connect system to stakeholders
• Legal Team(s)• Business Owners
ConclusionsAI & Machine Learning is Practical for Business at all Levels: Small, Medium and Large Enterprise
Benefits: • Save Time• Reach Customers• Reduce Manual Errors• More insights for Strategy and Planning
Requirement: Having a DataSet• Size of Data: Enough Cases so that Algorithm Can Learn
• Often Vendors Use their own Data to build their models• Clean Data:
• Organized in a tabular form• Erroneous Data taken out• Missing Data
• Make zero• Average of surrounding data points
Key Machine Learning Tech(implementations of these can be found on IBM’s BlueMix platform: https://console.bluemix.net/catalog/?category=watsonhttps://console.bluemix.net/catalog/services/ibm-graph?env_id=ibm:yp:eu-gb
• Classification – Classification is a tool that predicts what group or segment something belongs to (e.g., given a user data point, a tool can predict whether the person is male or female; a stay-cationer or world traveler, etc.). This classification can be used on data, text, images, etc.
• Image Recognition – Tools that can recognize an image by class or individually
• Natural Language Processing (NLP) – Being able to put regular human speech in a format that is understandable to a machine or software
• Sentiment Analysis – Taking a piece of text and identifying emotional intent, attitude or sentiment towards something
• Recommendation Systems – Systems that can recommend based on like-users’ preferences
• Clustering – Breaking an unorganized set of data into sub-groups or segments