CHRIS HAUSLER Building Data Products with Machine Learning @ Zendesk 18 JULY 2019
Data Product >> Building Models
What is a zen desk? Some Context
Hi, I’m Chris
Be the company your customers want you to be
Automation Recommendation Prediction Remove Inform decisions Spot trends repetitive work humans make humans can’t see Satisfaction Prediction Answer Bot Content Cues Tickets Satisfaction Prediction Question about delivery 88/100 500 relevant tickets Product question 78/100 Help Reset Password Locked Out 65/100 Reset my password Do any of these articles answer your question? password locked help International shipments Product doesn’t work 45/100 Yes, close my request Shipping information Yes, close my request Create New Article Cancel my policy 22/100 European Size Conversions Yes, close my request 12/100 Terrible service
DATA PRODUCT IS STILL PRODUCT 1 INVEST IN DATA INFRASTRUCTURE 2 LEARN TO LEARN 3 SCALING IS HARD 4 UX FTW 5
Data Product is still Product
ML is a hammer, not everything is a nail
Extreme customer-centricity for better experiences Start with the Embrace your data Be agile and iterative customer
TAKE-AWAY Be clear how to measure success Work with your Product Always come back to Manager the customer value
Of course! Invest in data infrastructure
WE HAD NO CENTRAL DATA STORE DATA CENTRES MORE DATA CENTRES . . . POD 1 POD 2 POD 3 Application A D W A D W A D W Servers Database Clusters PRIMARY SHARDS SECONDARIES Zendesk accounts live here
WE MADE A DATALAKE Db1 binlog Db2 Db1 events Maxwell Db[n] events P0 Db2 events P1 Kafka topic github.com/zendesk/maxwell
AND WE BUILT A THING
TAKE-AWAY Tie infrastructure investment to customer value
and don’t be afraid to pivot! Learn to Learn
ANSWER BOT Subject Re: Get my ticket data out of Zendesk Body Hi! We’d really like to dump our ticket data out of Zendesk so we can import it into an external reporting product and identify high risk customers. Can you help us out? Thanks a bunch George Support & Analytics Manager AwesomeCorp Pty Ltd Melbourne
WE STARTED WITH CLASSIC ML
BUT WE NEEDED MORE Global Deep Learning Model Solves the “cold start” problem and enables anyone to leverage AI immediately and respond quickly to new problems
TAKE-AWAYS: MAKE LEARNING PART OF YOUR CULTURE Create a safe space Run a Journal Club Get research as far ahead of engineering as far as possible
Scaling is hard
BUILDING MORE THINGS AWS BATCH Compute Environments Job Queues Trigger Job Model Serving Model Build Job Service Training Data (S3) Model Binary (S3) SNS + SQS
MAKE ONE MODEL DO MORE One Global Deep Learning Model
SO MANY MODELS Ticket: Language Code: Hoe reset ik mijn Language Detection nl wachtwoord? Tensorflow Serving Portuguese English Spanish French Dutch German (pt) (ne) (es) (fr) (nl) (de) Encoded ticket
TAKE-AWAYS Getting from one customer to many is hard Scaling needs Tooling Global models are great
UX FTW
AUTOMATICALLY RESOLVE CUSTOMER ISSUES WITH ANSWER BOT 1 2 3 A customer has Answers are The ticket is solved or passed to an agent a question suggested
Wording Matters “Solve my request” vs “Yes, close my request”
TAKE-AWAYS It doesn’t matter how good your model is if no one engages with it Make interactions clear so you can trust the feedback ML should never get in the way
Data Product >> Building Models
We’re Hiring! Thank you We’re Hiring!
Recommend
More recommend
