Dr.-Ing. Farshad FirouziSr. Technical Manager
Mobile: +49 171 4434886
Email: [email protected]
© Dr.-Ing. Farshad Firouzi
Artificial Intelligence Driven Omni-channel Customer Journey: From Awareness, Purchase, Service, to Loyalty
2
Background
Technical Project Manager, Solution Architect, Artificial Intelligence Engineer, and Data Scientist
• Post-Doctoral Degree in Computer Engineering (Full Scholarship)
– (Post-Doctoral: Machine Learning and Alternative Scaling)
– (PhD: Machine Learning on Chip for Predictive Maintenance and Reliability Analysis)
• Publication: 40+ scientific papers in peer-reviewed journals and conferences
• Research (Machine Learning, Reliability, IoT, Big Data, eHealth, Smart City)
– Associate Editor of three IEEE/ACM Journals
– Conference/workshop Program-chair• ICCAD, California, USA
• Complexis, Portugal
• IoTBDS, Portugal
• IoT for eHealth, Washington D.C., USA
– Program Committee Member• ATS, Japan
• Industrial Work Experience (Iran, USA, Belgium, Germany)
– Technical Manager/Leader: Smart Parking
– Technical Manager/Leader: Predictive Maintenance
– Technical Manager/Leader: eHealth Platform
– Technical Manager/Leader: Spike-based ANN System for Pattern Recognition
– Technical Manager/Leader: Statistical Timing and Reliability Analysis Framework
© Dr.-Ing. Farshad Firouzi
3
Are you ready for AI-driven IoT-based CUSTOMER JOURNEY!?
© Dr.-Ing. Farshad Firouzi
4
CustomersServices
Physical Store
Advertisement SocialWeb
Chat
MobileEmail
AI, ML, and Big Data give you a 360 degree view over your business and customers
Awareness Consideration Purchase Service Loyalty
Several touch points (email, web, social, etc.) during the journey © Dr.-Ing. Farshad Firouzi
5
UtilitiesUtilities
TransportationTransportation
BankingBanking
ManufacturingManufacturing
HealthcareHealthcare
HospitalityHospitality RetailRetail
InsuranceInsurance
AI, ML, and Big Data for Customer Analytics and Digital Transformation
© Dr.-Ing. Farshad Firouzi
6
A Real Life Scenario
ML-Driven 360 Customer Analytics: Omni channel says that Nina has propensity to fashion catalogs
ML-Driven Targeted Ad.: we tailor the catalog for her
We notice that Nina clicks on a boot
We find the best channel to approach her by ML
Using big data, we find her address We send an SMS invite her to the
nearby shop for visit
Beacon & Video analytics: We detect when Nina has entered the store
We send relevant information to her mobile app Store associate already know her identity and interest. We sell the boot
We make her loyal to our brand Product recommendation Churn prediction Satisfaction score Customer Lifetime Value (CLV) Appropriate incentive/offer
© Dr.-Ing. Farshad Firouzi
7
Benefits of Customer Analytics
Increase revenue
Decrease customer acquisition cost
Product enhancementReduce customer churn
Increase customer acquisition
© Dr.-Ing. Farshad Firouzi
8
Business StatisticsCustomer Analytics for Digital Transformation
50% 50% increase in loyalty program enrolment
40%
10%
40% improvement in call hold time
10% growth in data users
105% increase in offer sales105%
25% 25% drop in customer churn
87% 87% improvement in usage
64% 64% of customer think experience they have is more important than the price they pay!
6% 6% of organizations already started to invest on customer experience using AI/ML/Big data
© Dr.-Ing. Farshad Firouzi
9
What is Big Data
Structured, unstructured,
semi-structured
Terabytes of data
Batch, real-time, stream processing
Velocity Volume
Variety
three V’s
Structured
CSV, Columnar Storage (Parquet,ORC). Strict data model structure
Unstructured
Audio, video, images. Meaningless without adding
some structure
Semi-Structured
JSON, XML, sensor data, social media, device data, web logs. Flexible data model structure
Relational databases (RDBMS) work with structured data. Non-relational databases (NoSQL) work with semi-structured (streaming) data © Dr.-Ing. Farshad Firouzi
10
AI, ML, and Big Data Deliver Omni-channel Insights
Web Call center
Social network
Mobile
AdvertisingCRM/ERP/Transaction Data
Public data
Big Data, AI, ML, Analytics
85% of generated data by 2020 are unstructured! AI, ML, and Big Data techniques can rapidly
correlate, aggregate, and analyze your data and gain actionable insights
AI, ML, and Big Data techniques can quickly combine and enrich your existing data sets with 3rd
party data
© Dr.-Ing. Farshad Firouzi
11
Why Do You Need Big Data Solution
Old Technology was based on a Problem Driven Methodology Save some specific data Archive and never visit the rest again SQL Databases (e.g., SQL Server)
Schema on Write (Extract, Transform, Load (ETL)): Structured is applied to the data only when it’s Write!
New Technology is based on a Data Driven Methodology Store all the data! Extract value from data No-SQL Databases (e.g., Hadoop)
Schema on Read (Extract, Load, Transform (ELT)) : Structured is applied to the data only when it’s Read!
© Dr.-Ing. Farshad Firouzi
12
Hierarchy of Analytics
Proactive
What will happen?
What Should we do?
What happened?
The data
Do it for me
Perspective
Predictive
Diagnostic
Descriptive
Hie
rarc
hy
of
An
alyt
ics
© Dr.-Ing. Farshad Firouzi
13
AI, ML, Big Data Across All the Customer Journey Steps
Customer journey begins the moment they become aware of your brand
Get Data
Visualize Results
Retention• High-threat customers• Churn customers?• Timing• Find best future
customer
Service• Product performance• Channels and agent• Satisfaction• Next actionUpsell
• Product/service recommendation
• Timing• ChannelAcquisition
• Potential buyers• Optimal channels• Price optimization
© Dr.-Ing. Farshad Firouzi
14
The Evolution from Single-channel to Omni-channel
Customers experience a single type of touch-point
Retailers have a single type of touch-point
Customers see multiple touch-points acting independently
Retailers' channel knowledge and operators exist in technical & functional silos
Customers see multiple touch-points as part of the same brand
Retailers have a single view of the customers but operate in functional silos
Customers experience a brand not a channel within a brand
Retailers leverage their single view of the customer in coordinated and strategic ways
Single-channel Multi-channel Cross-channel Omni-channel
© Dr.-Ing. Farshad Firouzi
15
Customer Acquisition
Will you become my customer?
Reach out???
Calculate Score & Recommend best channel
Prospect Propensity Score
Channel
Farshad 10% Mobile
Stefan 30% Email
Reiner 60% Social Media
Kathrin 90% Pop-up ads
Manfred 45% Phone call
Find customer propensity score for each prospect (potential customer)
Recommend the best channel to contact each prospect
Execute data-driven campaigns
© Dr.-Ing. Farshad Firouzi
16
Brand Monitoring (I)
Social Network and Sentiment Analysis
Improve customer satisfaction Identify patterns and trends Male smarter (marketing/support)
decision
Assess
Segment
Discover
• Are we invest on right marketing channels
• What is the “share of voice” and “reachability” of our marketing strategy
• What users say about our brand and campaign? Understand the meaning of their comment using Natural Language Processing and convert it to a score ( Positive or negative?)
• Find meaningful insight about prospective customers
• Discover new ideas, trends, etc.• Topic analysis• Sentiment analysis
• What kind of audiences we have?
• Geographics, demographics
• Influence score• Recommenders
Social Media Assessment
Social Media Discovery
Social Media Segmentation
© Dr.-Ing. Farshad Firouzi
17
Brand Monitoring (II)
Sentiment Analysis: discovere people opinions, emotions, and feeling about a product or service
Good job but I expected a lot more.Totally dissatisfied with the service. Worth customer service ever.
Excellent effort guys. I appreciate your work.
NEGATIVE NEUTRAL POSITIVE
© Dr.-Ing. Farshad Firouzi
18
Brand Monitoring (III)
Social Network and Sentiment Analysis
60%30%
10%
Female Male Unknown
Share of voice (Author) by gender
60
10
30
10
15
5
30
60
20
35
20
9
B A DE N - WÜRT T E M B E RG
B A V A RIA
B E RL IN
H A M B URG
N O RT H RH IN E - WE S T P H A L IA
H E S S E
Female Male
0
5000
10000
15000
20000
25000
Positive Negative Neutral
Female Male
Distribution of gender across geographics
Sentiment by gender The trend (#mention) of the brand over time in different channels
0
5000
10000
15000
20000
25000
Facebook Instagram Pinterest Telegram Web
© Dr.-Ing. Farshad Firouzi
19
Brand Monitoring (IV)
Social Network and Sentiment Analysis
can exhaust featuring bigger html bit cnn
Sport cool youtube don’t comment
cnnmoney overview davidcward Farshadtwitterscene July hotel finance AOK million march
Action public answer iot pic days AI droneQualcomm
Automatically thanks round public ces dataOkay
Context of discussion Influence of social media authors
Sara
Alex
Tara
Niki
20.808%
12.203%
11.309%
9.503%
Share of voice
Share of voice
Share of voice
Share of voice
© Dr.-Ing. Farshad Firouzi
20
Image Processing & Video Analytics (I)
Real-life scenarios
Female: 99% Age (25-30): 95% Happy: 97%
Female: 99% Age (25-30): 95% Happy: 97%
Sarah O'Connor: 98% Last visit: 7.8.2018 Gold Customer Interest: Gucci, Armani Birthday: 21.03.1984 Single Address: Dusseldorf
Sarah O'Connor: 98% Last visit: 7.8.2018 Gold Customer Interest: Gucci, Armani Birthday: 21.03.1984 Single Address: Dusseldorf
Can be combined with other source of data e.g., CRM, Social networks, etc.
Loyalty program Customer satisfaction Upselling Tailored marketing
Customer analytics & product enhancement Pattern analytics for e.g., targeted advertising Demographic (age, gender, etc.) Analysis Location/product analytics: heatmap, #users,
duration of stay, hot products, interaction of users, happy or not?
Non-Registered Users Registered Customers
© Dr.-Ing. Farshad Firouzi
21
Image Processing & Video Analytics (II)
Distributed Architecture
Video Analytics At Edge (Local Store)Video Analytics At Edge (Local Store)
Camera Camera Camera
Analytics on Cloud (Headquarters)
Video Analytics At Edge (Local Store)Video Analytics At Edge (Local Store)
Camera Camera Camera
Send
Watch
-list
Send
Watch
-listReg
iste
r n
ew
use
rs
Reg
iste
r n
ew
use
rs
1
2
3
4
5
6
1. Farshad enters to a local office/store which is equipped with camera
2. A new face is detected by AI-driven video analytics in a local store. Sales associate can register the face with some info (name, address, etc.) in the system
3. Each face is represented by an embedding. The information is then sent to the cloud (headquarters of the business).
4. In the cloud, we store all the data. Please note that we do not same any real-image of our customer. Each customer just represented by an embedding (sequence of numbers). In the cloud we can perform analytics, category management, customer segmentation, etc. Moreover, the cloud can generate a “Watch-list”. A list of faces that should be detected by all branches (local stores) of a same brand.
5. The cloud sends the updated “Watch-list” to all the local stores
6. Now all the stores can detect Farshad. Thus, we can implement several interesting applications such as “loyalty program” and “360 degree customer analytics”
Dr. Firouzi© Dr.-Ing. Farshad Firouzi
22
Oh, hi!
AI-driven Beacon (I)
OH!You’re
nearby!
… Which wake up an application on your mobile
device andLets you to calculate your
location and PROXIMITY To The Beacon
… Which wake up an application on your mobile
device andLets you to calculate your
location and PROXIMITY To The Beacon
Near
Immediate
Far
Bluetooth Beacon transmit small packets of data
23
AI-driven Beacon (II)
Data Ingestion
21
4
3
BeaconBLE transmission
BeaconBLE transmission
Mobile App Sends contextual data (User,
Device, Application & Location) to cloud
Display tailored context-aware profile-based message to users (received from cloud)
Mobile App Sends contextual data (User,
Device, Application & Location) to cloud
Display tailored context-aware profile-based message to users (received from cloud)
Cloud1. Trace Apps/Users2. Combine beacon data with CRM, marketing, and
other sources of data3. Create User Specific Experiences (tailored proximity
and profile-based marketing/info/message)4. Geofence analytics (how many visitors, gender of
visitors, time spent by users, pick time)5. Perform location/traffic pattern analytics6. Perform Demographic Analysis7. Category management & heatmap (which products
get more attention)
Cloud1. Trace Apps/Users2. Combine beacon data with CRM, marketing, and
other sources of data3. Create User Specific Experiences (tailored proximity
and profile-based marketing/info/message)4. Geofence analytics (how many visitors, gender of
visitors, time spent by users, pick time)5. Perform location/traffic pattern analytics6. Perform Demographic Analysis7. Category management & heatmap (which products
get more attention)
Engagement
Social data
CRM
Other data
© Dr.-Ing. Farshad Firouzi
24
Product/Service Recommendation
What else are you interested in?
Cross-selling &Collaborative Filtering
Content-based Filtering
Social Interest Based
You and your friend like angry bird in Facebook
Kill bill is like… Reservoir dogs is like…
Upselling &Item Hierarchy
© Dr.-Ing. Farshad Firouzi
25
How Categorize Customers (Customer Segmentation)?
We need to spend our budget in a wise way!
Frequency3x month
Recency5 days ago
Monetary ValueEUR 120
RFM Model
High potential valueHigh Current Value
Keep These Customers
High potential valueLow Current Value
Grow These Customers
Low potential valueLow Current Value
Should you spend money here?
Low potential valueHigh Current Value
Grow These Customers
Potential ValueC
urr
ent
Val
ue
Jim Novo© Dr.-Ing. Farshad Firouzi
26
How Much is Your Future Business Worth?
Focus your marketing focuses on most valuable customers!
Customer Customer Value Class
Farshad Gold
Stefan Silver
Reiner Bronze
Kathrin Platinum
Manfred Silver
Ingestion, Cleaning, &
Fusion
Noise Removal &Feature
Engineering
Data Set
Demographics (e.g., Age, Gender, Income) Transactions
Build a model that predict the customer group of a new customer
Classification (e.g., Random
Forest)
© Dr.-Ing. Farshad Firouzi
27
Discover Response Patterns
Group customers based on their response patterns
Analyze groups to understand patterns Identify how customers respond to your offer Build data-driven marketing Target each group with a specific campaigns
17 Customers
Clustering
4 groups Channels
© Dr.-Ing. Farshad Firouzi
28
Customer Retention and Customer Loyalty
Retention Strategy: A plan or process designed to help retain customers after the first sale!
Retention Effort
Meaningful Memorable Personal
Customer Retention: Strategic actions and
efforts that promote loyalty Customer loyalty: An emotional bond
between customers and a business Loyalty is merely a symptom or a result, and
retention efforts are the cause Retention is all about proactive efforts.
© Dr.-Ing. Farshad Firouzi
29
Customer Lifetime Value (CLV)
Invest your marketing budget on most important customers!
Customer CLV
Farshad $2,132
Stefan S1,200
Reiner $3,750
Kathrin $10,000
Manfred $950
Ingestion, Cleaning, &
Fusion
Noise Removal &Feature
Engineering
Data Set
Demographics (e.g., Age, Gender, Income) Transactions
Regression
80% | 80% of your business comes from 20% of your customers 80% | It costs 10x less time to sell to an existing customer than finding a new
customer
© Dr.-Ing. Farshad Firouzi
30
Are you Happy with me? (I)
Customer Service Goals Minimum questions Early resolution Keep track of context
Customer Service Goals Self-service and chat bots Need for intelligence Predictive analysis can help
WWW
2
1
3
4
5
© Dr.-Ing. Farshad Firouzi
31
Are you Happy with me? (II)
Solve the problem in the first interaction!
Ingestion, Cleaning, &
Fusion
Noise Removal &Feature
Engineering
Data Set
Sales Transaction Product Amount Order Date Status Shipped On Delivered Returned
Previous Contacts Date Reason Agent Duration Root Cause Resolved Ticket
Classification(e.g., Random
Forest)
Customer Predicted issue
Agent
Farshad Product issue
Agent 121
Stefan Billing issue
Agent 52
Reiner Shipping issue
Agent 21
Kathrin New product
Agent 15
Manfred Returns Agent 7
1. Predict the intent of contact and assign an appropriate agent with relevant skill/knowledge to solve the issue in the first interaction!
2. Processional customer services are expensive, so we also need to optimize their time by assigning critical issues to more professional one
3. Be proactive and start by asking the customer a very simple question. Are you experiencing any problems with our products or services?
© Dr.-Ing. Farshad Firouzi
32
Are you Happy with me? (III)
Find unsatisfied users and predict customer churn!
Only 10% of customers answer to surveys
We need to find unsatisfied users, or they go to our
competitors Preparation & Machine Learning (e.g., Regression)
Data Set
Customer SAT. Score
Farshad 1.1
Stefan 2
Reiner 4.3
Kathrin 5
Manfred 3.5
Demographic History Transaction Social medias Surveys
© Dr.-Ing. Farshad Firouzi
33
Are you Happy with me? (IV)
Ingestion, Cleaning, &
Fusion
Noise Removal &Feature
Engineering
Data Set
Association Rules Mining
First item Second item
% times
Lifetime > 3 years
Age < 27 90%
Single German 75%
Find Customer Attrition Patterns by Data Mining
© Dr.-Ing. Farshad Firouzi
34
Are you Happy with me? (V)
ML-driven incentives recommendation and loyalty program engine
Extended Warranty
Free Software
Coupon
1. Based on attrition and satisfaction score, we can detect which customer is willing to leave us!
2. Marketing and support team to reach customer with an offer that makes them stick with us!
3. We need to find an appropriate offer for each person, since different customers react differently to different offers (longer warranty, coupon)
Machine Learning (Recommendation Engine) will find incentives for each user
Farshad
Stefan
Kathrin
© Dr.-Ing. Farshad Firouzi
35
Path Analytics
34%
23%
11%
2%
Majority of interactions happen is based on multi-event, multi-channel journeys
To better understand users to be able to improve the product/service and provide high-quality uniform user experience at each user touchpoint, it is important to understand user interactions with the brand
Big data and machine learning can help us to visualize and analyze the paths customers take, over time and across channels
Big data and machine learning also enable us to contact the customers on a personal level and create a long-term relationship to build loyalty.
Focus your effort (e.g., Marketing), time, and budget on important paths & channels
© Dr.-Ing. Farshad Firouzi
36
Dr. Firouzi
Home SAT. Score Churn Services
0
2
4
6
CLV
(EU
R)
Time Longer warranty
Coupon
Discount
Free Software
Analytics Firouzi
Customer Satisfaction
63%
Probability to leave
83%
Dr. Farshad Firouzi
Male 1983
Bettina-von-arnim weg 7Karlsruhe
+49 1714434886
Probability to Call
95%
Customer Lifetime Value (CLV) Appropriate Incentives
Customer Journey
July 1, 2017
Dec 1, 2017
Dec 5, 2017
OverallOverall
PaymentPayment
Customer ServiceCustomer Service© Dr.-Ing. Farshad Firouzi
37
Conclusion
► Customer journey: Interaction between an organization and a customer in different channels over the duration of their relationship
► Digital transformation allows you to automate the whole process from customer acquisition to service, retention, and loyalty
► Identify and priorities your business
► Build faster and higher customer conversion rates► Percentage of users who take a desired action
► You can validate your customer's journey. It also enables you to find your customer's pain points
► Increase customer satisfaction, loyalty and word-of-mouth recommendations
► You can be proactive. You can predict the next move of your customer
► Increase customer satisfaction, loyalty and word-of-mouth recommendations
© Dr.-Ing. Farshad Firouzi
38
Thank you.
© Dr.-Ing. Farshad Firouzi