2. About the Speakers
Víctor Martín
Digital Innovation at
Avanade
Twitter: @tori_parla
LinkedIn:
https://es.linkedin.com/in
/vmartindecabo
Hind Azegrouz
Advanced Analytics
Manager Avanade
Linkedin:
https://www.linkedin.co
m/in/hindazegrouz/
Diego Pecharromán
CRM Manager at Avanade
Twitter: @d_pecharroman
Linkedin:
https://www.linkedin.co
m/in/dpecharroman/
3. AML - CRM - Asistente virtual
Hay mucha información en todos los CRMs que es muy valiosa, pero no se explota
correctamente. Con Azure Machine Learning se puede analizar y predecir clientes insatisfechos,
proponer nuevos productos, segmentación de clientes…
Adicionalmente con Bot Framework podemos leer la información de CRM para ayudar a los
usuarios de CRM.
4. • Companies spend significant sums to acquire customers. Once
hooked, marketers protect those investments by attempting to keep
patrons happy, engaged, and most of all, loyal.
• Reducing customer attrition, or "churn", often involves offering
incentives such as discounts to individuals identified as likely to
defect. The tricky part comes in figuring out exactly who should be
targeted.
Churn Prediction
5. Training data and classification model
Before building the training models, in general a close examination of
the data is required:
• Data cleanup: removing missing data
• Data balancing: We notice that the total training set is formed of 96%
of happy customers vs 4% of unsatisfied ones. we proceed to
generating more occurrences of scarce data using Synthetic Minority
Oversampling Technique node.
• Permutation Feature Importance node, was used to examine the
relevance of the different features: We decided to keep all features, as
all carry some information.
A positive score means variable is
improving classification accuracy
6. The algorithm
• In order to test the AzureML service, we implemented a total of nine classification models offered by AzureMl studio
for binary classification problems
• When possible, the best model using each algorithm is obtained automatically by iterating over a loop of parameters
7. Evaluation method
• Evaluation of the different classification models was performed using the training data set,
70% to train, 30% to test,
• We report AUC values that measures the area under the curve plotted with true positives on
the y axis and false positives on the x axis. This metric is useful because it provides a single
number that lets you compare models of different types.
8. Web Service Call
• A web service can be generated automatically using Azure ML studio
• Set the Input and Output Port of the Web Service
• Build a client application to call the new Web Service, using the
parameters generated
Input and Output Ports
10. ¿Qué hay detrás de un asistente virtual?
• Es la evolución tecnológica más grande que el mundo haya visto jamás
La Inteligencia Artificial permite a las máquinas interactuar de forma natural con las personas, los datos y el entorno. Estos sistemas
crean interacciones más intuitivas y amplían las capacidades de lo que cualquier humano o máquina puede hacer por su cuenta.
Muchas tecnologías han evolucionado y se han combinado para crear tres habilidades
fundamentales en software hoy en día.
El nuevo impulso esencial para la AI es la capacidad de
aprender, adaptarse y mejorar autónomamente - a
costes cada vez más bajos.
APRENDER
Mejorar el rendimiento
(calidad, consistencia y
precisión) basado en
experiencias del mundo real.
SENTIR
Percibir el mundo
mediante la adquisición
y procesamiento de
imágenes, sonidos y
habla.
COMPRENDER
Analizar y comprender
la información
recopilada añadiendo
significado e ideas.
ACTUAR
Tomar acciones en el
mundo físico
basadas en la
comprensión y el
entendimiento.
11. Microsoft Bot Framework
Bot Framework supports a
wide range of deployment
channel options out of the
box to support your users in
the media channels they
prefer
Bot Framework also gives
the ability to create custom
deployment channels,
meaning chatbot can be
consumed almost anywhere
Microsoft Bot Framework
enables your users to
interact with applications,
data, and processes in a
natural, intuitive manner
through utilization of
Microsoft Cognitive Services
and 3rd party AI APIs and
services.
Bot Framework enables
integration across AI
vendors, including Google,
AWS, IBM, and more
Bot Framework applications can be written with multiple languages,
and can be deployed in Azure, on-prem, or another cloud provider.
Flexibility enabled.
Bot Framwork is built to easily integrate with 3rd party backend
systems and APIs. Microsoft AI seeks to integrate, not isolate
Bot Framwork can integrate
with all major authentication
providers.
Azure Active Directory
provides integration with
existing on-premise AD
12. Ecosistema de Microsoft AI
Cortana Intelligence
Suite
Cognitive Services
Suite
Microsoft Graph
Microsoft Bot
Framework
Power Apps
Digital Customer
Digital Worker
14. Dynamics 365 Community
The Dynamics 365 Community is a site where you can find community contributions, ask questions and
interact with Microsoft Dynamics peers and experts. The community has over 200K members and is
growing.
New UI/UX: https://community.dynamics365.com
16. Q&A
Víctor Martín
Digital Innovation at
Avanade
Twitter: @tori_parla
LinkedIn:
https://es.linkedin.com/in
/vmartindecabo
Hind Azegrouz
Advanced Analytics
Manager Avanade
Linkedin:
https://www.linkedin.co
m/in/hindazegrouz/
Diego Pecharromán
CRM Manager at Avanade
Twitter: @d_pecharroman
Linkedin:
https://www.linkedin.co
m/in/dpecharroman/
Editor's Notes
CRM puede sacar mucho aprovecho apoyandose en otras herramientaas que pueden añadir valor y aportar mas insight al negocio.
Hoy vemos como CRM puede interactuar con herramientas de azure, en particular: azure machine learning y Azure bot framework