Slide 1

Slide 1 text

Scaling Machine Learning at Holiday Extras REBECCA VICKERY | DATA SCIENTIST @vickdata

Slide 2

Slide 2 text

Travel planning is time consuming Airport parking Airport hotels Airport lounges Travel insurance Holiday money Port products Car hire Airport transfers 582 minutes Over 46 days* Travel Planning *Facebook commissioned consumer research company GfK

Slide 3

Slide 3 text

Optimising consumer decision making Airport parking Airport hotels Airport lounges Travel insurance Holiday money Port products Car hire Airport transfers Less Hassle. More Holiday Trip recommendations

Slide 4

Slide 4 text

The machine learning process Historic Data (Input) (Input Algorithm Learns Mapping Predictions (Output)

Slide 5

Slide 5 text

Automated bidding Ad targeting Channel optimisation 1 Ad spend 2 Commercial 3 Customer Experience 4 Marketing Lots of other processes to optimise Automated pricing Allocation Revenue optimisation Automated call handling Personalised experiences Intelligent messaging Optimising send frequency

Slide 6

Slide 6 text

Machine learning needs to scale

Slide 7

Slide 7 text

Deploying machine learning is hard Scaling is even harder

Slide 8

Slide 8 text

Tools - Data Scientists Open source Lack Software Development expertise Mainly Python c Flaticon

Slide 9

Slide 9 text

Tools - Software Engineers Different tools Lack ML/Data expertise Mainly Javascript c Flaticon

Slide 10

Slide 10 text

Data science process The wrong kind of independence c Flaticon

Slide 11

Slide 11 text

People Small data science team Science + software experts are rare c Flaticon

Slide 12

Slide 12 text

Two types of deployment

Slide 13

Slide 13 text

Bespoke Solutions “Ideas are worth nothing unless executed”, Derek Sivers c Daniel Moyo

Slide 14

Slide 14 text

Unused Models Many models never make it to production “Ideas are worth nothing unless executed”, Derek Sivers

Slide 15

Slide 15 text

Time to model deployment Model development = days to weeks Model deployment = weeks to never! “Ideas are worth nothing unless executed”, Derek Sivers

Slide 16

Slide 16 text

The Google Way

Slide 17

Slide 17 text

c Flaticon __init__.py task.py setup.py model.py Model Package

Slide 18

Slide 18 text

Repeatable, Reusable Process __init__.py task.py setup.py model.py Model Package

Slide 19

Slide 19 text

No content

Slide 20

Slide 20 text

Not Quite!

Slide 21

Slide 21 text

Collaborative Project

Slide 22

Slide 22 text

ML Proxy (bespoke ML microservice)

Slide 23

Slide 23 text

Monitoring - Model Performance

Slide 24

Slide 24 text

Monitoring - AI Platform Performance

Slide 25

Slide 25 text

AI Platform A technical solution but also a strategic solution c Google

Slide 26

Slide 26 text

The right kind of independence c flaticon Data Scientists can use preferred tools

Slide 27

Slide 27 text

The right kind of independence c flaticon Repeatable process for deployment of most models

Slide 28

Slide 28 text

Faster time to production c flaticon Fully Managed service

Slide 29

Slide 29 text

Faster time to production c flaticon Fast response from Google Cloud Support

Slide 30

Slide 30 text

Extensibility c flaticon.com Multiple models and versions

Slide 31

Slide 31 text

Customisable c flaticon.com Regularly release support for newer versions of tools

Slide 32

Slide 32 text

Time to model deployment Model development = days to weeks Model deployment = hours to days “Ideas are worth nothing unless executed”, Derek Sivers

Slide 33

Slide 33 text

Less Hassle. More Holiday Trip recommendations

Slide 34

Slide 34 text

Thank you @vickdata