SentiSquare develops an AI technology which processes company-customer communication. Through automatization of routine processes and providing a clear understanding of how customers feel about specific issues, it helps companies to laser-focus their resources to deliver a superior customer experience. The solution powered by cross-lingual semantics enables the comparison of conversations across languages without the need for machine translation. Companies can thus analyse and respond to both global and local issues their customers face.
2. Udanda vel molore
nonsenitae. Et exerori
tinctat que niatum,
sitis sapienis.
Each year globally,
companies lose $300bn
due to poor CX
How much is
your firm
losing?
3. “SentiSquare AI
is a great saver
of people & time”
—Vojtěch Vycudilík
Cxpert @ T-Mobile
Udanda vel molore
nonsenitae. Et exerori
tinctat que niatum,
sitis sapienis.
4. AI-as-a-
Service
Since 2014
• 5 years on the market
(CZ,SK,PL,DE)
• Investors
Luboš Černý
Former CEO of eBanka
Michael Fehn
Former CEO of E.ON CZ
• StartupYard, E.ON :agile,
& Poland Prize alumni
5. doc. Josef Steinberger,
Ph.D.
Michal Konkol,
Ph.D.
Tomáš Brychcín,
Ph.D.
Most Precise NLP engine on the market:
1st & 2nd place at SemEval San Diego 2016
Dev Team
Tomáš Hercig,
Ph.D.
2 100+ citations in NLP
7. THE OLD WAY
Udanda vel molore
nonsenitae. Et exerori
tinctat que niatum,
sitis sapienis.
LEGACY
FILTER
MANAGER
?
FLOODS OF CUSTOMER-GENERATED TEXT
REPETITIVE
WORK
THE NEW
WAY
8. Udanda vel molore
nonsenitae. Et exerori
tinctat que niatum,
sitis sapienis.
LEGACY
FILTER
MANAGER
?
AI-
POWERED
ANALYTICS
FLOODS OF CUSTOMER-GENERATED TEXT
CREATING
VALUE+REPETITIVE
WORK
OR
9. Customer Care ops clean-up
1 INTELLIGENT ROUTING
• Query category & customer
identification
• Reduces handling time
2 AGENT JET-PACK
• Auto-fill suggestions based
on relevant past replies
• Increases agent productivity
3 NOTE-TAKING
• Automated summary generation into
CRM
• Saves time, boosts morale
4 PERFORMANCE &
COMPLIANCE
• Good/Bad interaction
recognition
• Upgrades agent training
5 CHURN ALERT
• Identifies churn candidates
• Reduces churn
10. TASK: Routing messages
It takes time to dispatch emails
(2 min 39 sec per email; 12 FTE per 500k emails)
SOLUTION: NLP-driven automation
Automatic dispatching to 280 queues
Accuracy: 1st attempt >70%; 2nd attempt >80%
11. Finding messages that matter
1,250,000
60 %
Every year, client center asks 1.250.000
customers for SMS feedback.
60 % clients text back
( >10,000 SMS / month)
?
Customers are glad to receive a reply –
but which ones should get it?
Dissatisfied
customers
Lucrative
customers
12. Churn alert
CHALLENGE: Difficult to identify churn candidates
OUR SOLUTION: Assessing churn risk based on data from past
interactions, picking out customers for extra care
BENEFITS: Reduced churn
13. Revealing what customers talk about and what they want improved
OUR SOLUTION: Identification of topics, keywords and sentiment — clearly visualized trends over time
Deep Insight
14.
15. Read and classify one verbatim
Quarterly chart ready in
How would you use 120 MD?
Deep Insight: Delivering Value
Before
2 min 24 sec …………… ….
3 weeks
Tagging verbatims
Immediately …. …..
2 days
Increasing NPS
CLIENT: (retailer); TASK: Analyze 40k feedback pieces each quarter
16. • Our AI learns the language of any specific market
(e.g. German Automotive)
• Our AI learns the language of any specific channel
(e.g. SMS vs emails vs call transcripts)
• Solutions deployed in German, English, Czech,
French, Hungarian, Chinese, Japanese & more
• Language-agnostic tech: Words are reduced to
numbers and vectors – Our Natural Language
processing works in any language
Language
Independence
17. • Words occurring in similar contexts have similar
meanings
• Meaning = point in a high-dimensional space
• Unsupervised machine learning: No human input
Distributional Semantics
18. • XYZ is made of barley and hops.
• There is a pint of XYZ on the table.
• XYZ can make you drunk.
• Let’s go grab a XYZ!!
What is XYZ?