BINGO! DINO DNA! HEALTH IDENTITY FROM THE WRIST BRIAN WILT HEAD OF - PowerPoint PPT Presentation

BINGO! DINO DNA! HEALTH IDENTITY FROM THE WRIST BRIAN WILT HEAD OF DATA SCIENCE AND ANALYTICS @BRIANWILT

NOV 16 2015

“THEY TOLD ME TO SLEEP NORMAL”

BRIDGING SCIENCE AND ENGINEERING

UP 3

ROLE OF DATA+ML AT JAWBONE • SCIENCE • PERSONALIZATION • MOTIVATION

SCIENCE: UNDERSTANDING THINGS NEVER BEFORE SEEN

SOUTH NAPA EARTHQUAKE 100% NAPA, SAN FRANCISCO, SONOMA OAKLAND 75% NAPA, SONOMA, 50% BERKELEY 0-15 mi 0-25 mi 25% 25-50 mi 50-75 mi 75-100 mi 0% 9p 10p 11p 12a 1a 2a 3a 4a 5a 6a 7a 8a 9a

HEALTH WITH SHARED CONTEXT IN REAL TIME

PERSONALIZATION: AN EXPERIENCE THAT KNOWS YOU

RISE OF THE DATA NATIVES MONICA ROGATI YouTube was abuzz with viral videos of small children — yet to speak, read or write — “pinching” magazine articles with their fingers as they would an iPad. These children were heralded as […] “digital natives” . [Now is a] new revolution, this time of “data natives” who expect their world to be “smart” and seamlessly adapt to them and their taste and habits .

MOTIVATION: A SMART COACH HELPING YOU LIVE HEALTHIER

7.2 HRS SLEEP WEIGHT AFFECTS SLEEP 7 6.8 FEMALE 6.6 HEALTHY WEIGHT 6.4 MALE 6.2 19 23 27 31 35 39 43 47 (BMI)

UP USERS WITH MAJOR WEIGHT LOSS logged 75% more meals per week logged 60% more workouts per week beat their step goal 43% more often averaged 17% more steps per day logged 16% more weigh-ins per week

IDENTITY 24 7 DATA A COMPLETE PICTURE OF YOU ACTIVITIES BIOMETRICS FULLY CONTEXTUALIZED

HOW DO WE MAKE THIS HAPPEN?

DATA PRODUCT ALL-STARS

DATA PRODUCT ALL-STARS PEOPLE YOU 
 NEWS FEED PAGERANK RELATED ITEMS MAY KNOW

DATA PRODUCT ALL-STARS PEOPLE YOU 
 NEWS FEED PAGERANK RELATED ITEMS MAY KNOW CTR

CHRONIC DISEASE CARE IS 86% OF US HEALTHCARE COSTS CARDIOVASCULAR DIABETES 67% OF GYM MEMBERSHIPS 💕 🍱 $193 BN $176 BN GO UNUSED OBESITY CANCER 🍠 🚭 $147 BN $157 BN

HOW WILL WE MAKE HEALTH DATA PRODUCTS?

DATA SCIENCE FOR THE LITTLE GUY

DATA SCIENCE FOR THE LITTLE GUY PROBLEM LARGE COMPANY SMALL COMPANY Where do I put my data science team? PEOPLE Only one or two people LOL Vertical/horizontal organization? PRODUCT Established Nascent DATA A mess A mess ML Deep on solving a problem, predictive Understanding and interpreting

DATA SCIENCE FOR HARDWARE Raw + Rich Compressed Limited SIGNALS Limited Sensor fusion Historical + Population CONTEXT Single Single Aggregate USERS Seconds Minutes Slow LATENCY Limited Powerful HAL 9000 COMPUTE Months Weeks Hours DEPLOYMENT

ACTIVITY CLASSIFICATION VERSION 0 MOST COMMON WORKOUT 54m activity 58% ACCURACY +5,972 steps | +50%of goal VERSION 1 LAST WORKOUT 65% ACCURACY Awesome! Looks like you racked up some steps. What were you up to? VERSION 2 BASKETBALL SERVER-SIDE MODEL TENNIS 85% ACCURACY ZUMBA VERSION 3 Edit BAND-BASED MODEL 99% ACCURACY

THE FIRMWARE HAS TO WORK THE FIRST TIME

TESLA HAS TO DELIVER THE CAR FIRST

DATA SCIENCE FOR WEARABLES

WE CAN GET EXPONENTIALLY MORE FROM THE WRIST UP UP 3 2011 2015 Accelerometer Accelerometer Temperature GSR Heart Rate

WE CAN GET EXPONENTIALLY MORE FROM THE WRIST UP UP 3 2011 2015 GERMAN Accelerometer Accelerometer PEDOMETER Temperature 1590 GSR Heart Rate

THE IPHONE MOMENT IS COMING

STEPS 4.1T SLEEP 330M MEALS 270M WORKOUTS 100M

HEALTH DATA IS WIDE

HEALTH DATA IS WIDE

BUT WIDE DATA TAKES MORE CLEANING • CLEAN EARLY AND OFTEN • APIS ARE HARD • BETTER TOOLS ARE NEEDED • ML IS NAIVE

CONTEXT IS KING 80 EMPLOYEE HEART RATE DURING LAST JAWBONE ALL HANDS 70 60 50 11:00 AM 11:10 AM 11:20 AM 11:30 AM 11:40 AM 11:50 AM 12:00 PM 12:10 PM 12:20 PM 12:30 PM

SCALE GRACEFULLY

BEDTIME

BEDTIME

16 MIN

SLEEP RECOVERY

BUT WHAT ABOUT DATA WE CAN’T GET

• API INTEGRATIONS • BAYESIAN INFERENCE • PUBLIC WHO/CENSUS DATA • USER DATA • UNSTRUCTURED TEXT • CREATIVITY • HIRING

DATA PRODUCTS SCALE HUMAN JUDGMENT. THEY ALSO SCALE AND REINFORCE HUMAN BIAS. @CLARECORTHELL

TOKYO’S WORKERS GET LESS SLEEP THAN US, ASIAN COUNTERPARTS KANOKO MATSUYAMA (2011) Tokyo o ffi ce workers slept at least 30 minutes less than their counterparts in New York, Paris, Shanghai and Stockholm each night, averaging six hours, or 14 percent less than the recommended minimum, a study said. The study surveyed 180 men and women from each of the five cities.

DATA DEMOCRATIZATION IS TABLE STAKES datazoo=# SELECT gender, COUNT(*), AVG(total_hrs) datazoo-# FROM up_mod.sleep_clean_night datazoo-# WHERE gender IS NOT NULL GROUP BY gender; gender | count | avg --------+-----------+--------- M | 147284397 | 6.77170 F | 165579431 | 7.11274 Time: 1333.536 ms

DATA PRODUCTS PIPE API USER DATA EVENTING STREAM ANALYTICS REDSHIFT DYNAMO

EXPERIENCE FIRST, THEN TECH

NOT JUST ANALYTICS — DEMOCRATIZE DATA PRODUCTS

NOT TRACKING, BUT COACHING

SMART COACH Early to Bed Smart Coach noticed that you’ve been turning in late recently. Remember, your brain needs plenty of pillowtime to sort new information. Get in bed by 10:50pm tonight? Not today I’m in

SMART COACH SMART COACH Early to Bed Early to Bed Smart Coach noticed that you’ve been Smart Coach noticed that you’ve been turning in late recently. Remember, your turning in late recently. Remember, your brain needs plenty of pillowtime to sort brain needs plenty of pillowtime to sort new information. Get in bed by 10:50pm new information. Get in bed by 10:50pm tonight? tonight? Not today Not today I’m in I’m in

SMART COACH GUIDES YOUR SLEEP SMART COACH Early to Bed UP UP now Smart Coach noticed that you’ve been Today’s move summary: 10,518 turning in late recently. Remember, your steps, covering 5.1 mi, burning brain needs plenty of pillowtime to sort 1,518 cal. That’s 105% of your goal. new information. Get in bed by 10:50pm slide to view tonight? Not today I’m in

SMART COACH GUIDES YOUR SLEEP OCT 23 5 DAY SMART COACH Early to Bed STREAK UP UP now Smart Coach noticed that you’ve been Averaging 7h 6m per night Today’s move summary: 10,518 turning in late recently. Remember, your steps, covering 5.1 mi, burning brain needs plenty of pillowtime to sort 1,518 cal. That’s 105% of your goal. new information. Get in bed by 10:50pm slide to view tonight? Not today I’m in

SMART COACH GUIDES YOUR SLEEP OCT 23 5 DAY STREAK UP UP now e been Averaging 7h 6m per night Today’s move summary: 10,518 , your steps, covering 5.1 mi, burning o sort 1,518 cal. That’s 105% of your goal. y 10:50pm slide to view I’m in

UP HELPS USERS 23M SLEEP MORE SMART COACH MORE SLEEP, COMPARED Early to Bed TO THE CONTROL GROUP Smart Coach noticed that you’ve been turning in late recently. Remember, your brain needs plenty of pillowtime to sort new information. Get in bed by 10:50pm tonight? 72% Not today I’m in INCREASED LIKELIHOOD OF BEATING THEIR SLEEP GOAL

ANDRE IGOUDALA FINALS MVP SLEEP PROFILE SLEEP COACHING • HOME: SLEEPS 5H 48M • CONSISTENT BEDTIME • AWAY: SLEEPS 6H 15M • PUSH FOR LONGER SLEEP DURATION • DIFFICULTIES • TOOLS TO HELP FALL ASLEEP - FALLING ASLEEP AFTER GAME • STRATEGIC CAFFEINE USAGE - STAYING ASLEEP • NAPPING RECOMMENDATIONS • NAPS SOMETIMES • DRINKS CAFFEINE BEFORE AND DURING GAMES

SLEEP HELPED ANDRE’S GAME SLEEP IMPROVED ANDRE’S 3P% BY 2.18X SLEEP HELPED ANDRE’S OFFENSE - 37% - 45% + 29% + 8.9% 0.250 <7 HRS 0.462 7-8 HRS 0.545 >8 HRS POINTS PER MINUTE FREE THROW TURNOVERS PER FOULS 3 POINT PERCENTAGE

CONCLUSIONS • WE DO BIG THINGS AT SMALL COMPANIES • MORE IMPORTANT TO FIGURE OUT WHAT TO DO THAN DO IT • DATA SCIENCE FOR “GOOD” • IMPROVE HEALTH WITH BIG DATA TOOLS • USER EXPERIENCE COMES FIRST • THE FIRST ML MODEL ISN’T A MODEL • DEMOCRATIZE DATA PRODUCTS

BINGO! DINO DNA! HEALTH IDENTITY FROM THE WRIST BRIAN WILT HEAD OF DATA SCIENCE AND ANALYTICS @BRIANWILT