Beyond analytics: Prescriptive analytics for the future of your business by Álvaro Barbero at Big Data Spain 2015

Beyond analyticsPrescriptive analytics for the future of your business

Álvaro Barbero

www.iic.uam.es

Beyond analytics | Prescriptive analytics for the future of your businesswww.iic.uam.es

Beyond analytics

A note on Big Data Hype


Gartner’s Hype Cycle of emerging technologies

2

0

1

2



2

0

1

3



2

0

1

4



2

0

1

5


Big data Hypes leves

Big

dat

a H

ypes

leve

s

2012 2013 2014 2015



Big

dat

a H

ypes

leve

s

2012 2013 2014 2015 2016 2017



Big

dat

a H

ypes

leve

s

2012 2013 2014 2015 2016 2017



Big

dat

a H

ypes

leve

s

2012 2013 2014 2015 2016 2017

GO SELL BIG DATA PROJECTS

NOW


What´s better to know?

What’s better to know?

What is happening



What will happen



How do I take advantage




Levels of analytics

Levels of analytics

Descriptive analyticsWhat is happening

Predictive analyticsWhat will happen

Prescriptive analyticsWhat is the best strategy


Descriptive analytics


What is happening

- KPIs- Dashboards- Visualizations

Describe the data to get useful info




What is happening




What is happening

Estimate data we don’t

haveBeyond analytics | Prescriptive analytics for the future of your businesswww.iic.uam.es

Predictive analytics


What will happenClusteringRegression

Anomaly detectionClassification




What will happen

Real Predicción




What will happen


Levels of analytics

84%

13% 3%

Descriptive

Predictive

Prescriptive

Levels of analytics


Prescriptive analytics


What is the beststrategy





START





START

??

??

?????

?

??





START

??

??

?????

?

??

Descriptive Analytics





START

??

??

?????

?

??

Predictive Analytics





START


How it works

How it works


How it works

How it works

Numerical optimization




=+




Simple approach

- Predict expected benefit of each business action, independently of other actions

- Choose the action or actions that maximize benefit, within a budget or other constraints

- Easy to apply on top of standard analytic pipelines

A

B

C

B

C

A

B

C

22

93

78




Advanced approach

- Iteratively predict and simulate different future scenarios, considering all possible business actions and parameters, to obtain the different expected outcomes- Use outcomes of previous simulations to drive the search toward better solutions- Expert knowledge in optimization is required

Operations research Genetic algorithms Ant colony optimization

Application examples


Application examples

Customer service automation





Level 1General support

Level 2Specialized Support



Level 1General support

Active Learning

Level 2Specialized Support



Experience with a

national bank

Over 300 issue classes

% a

ccur

acy

% machine classified

Automated

Manual

Smart pricing



Smart pricing

Predictionof future room demands.

Simulationof expected sales given a pricing strategy.

Optimizationof the pricing strategy to maximize revenue.


Smart pricing

Predictions must be accurate and consider all possible factors

… and if they are not, we need to know confidence bounds or some way to assess risk

Risk


Smart pricing

Client traits

Marketing actionsHotel features

Environmental conditions

- Demography- City of origin- Family, business, …- Stays at other hotels

- Special offers- Promotions- Client interactions

- Hotel placement- Kind of location

- Competition- Weather

- Services- Rooms- Room rates

% Dto.


Smart pricing

When smart pricing goes wild


Smart pricing

Bussiness rules and constraints must be coded into the model

Prevents wild decisions

Retail logistics optimization





Cash demands forecasting

? ?

Transportation schedule optimization

? ?

?



Predicted ATM and branch withdrawals

Predicted deposits Acceptable stock-out risk

Transportation rates Euribor rates Transportation andregulation constraints



Expected gains

reduction in costs 14%


Take-home message

Take-home message


Don´t ask

Don’t ask what you can do for your data

Sculpture by Alicia Martín


Ask

Ask what your data are going to do for you

(with prescriptive analytics)

C/ Francisco Tomás y Valiente, nº 11EPS, Edificio B, 5ª planta

UAM Cantoblanco. 28049 MadridTel.: (+34) 91 497 2323

www.iic.uam.es blog.iic.uam.es

Elementos gráficos de apoyo obtenidos en

&

Beyond analytics: Prescriptive analytics for the future of your business by Álvaro Barbero at Big Data Spain 2015

Business