PowerPoint Presentation
1machine learning + measurement science -> big data insights
Conal Sathi, data alchemistBrad Pflum, data scientist
Slice Intelligence Confidential Do not copy or distribute without prior written consenthow much do you spend on jeans?who spends about $40 on a pair?who spends less than that?who spends $40 to $100?who spends more than $100?so then the average price of jeans is . . . average price of jeans is $20
Slice Intelligence Confidential Do not copy or distribute without prior written consent
2
who is wearing a wearable?
Slice Intelligence Confidential Do not copy or distribute without prior written consentnot so fast!enduring challenge of generalizing from sample to population
especially when the sample is biased
especially when you dont have perfect information about the sample and the population
which is always
so thats what well talk about in the middle of this talk!
Slice Intelligence Confidential Do not copy or distribute without prior written consent
Slice overviewfounded in 2010, Slice has amassed a 4.4-million person panel of online shoppersdiverse, massive panel recruited from a mix of sourceswholly-owned, independent subsidiary of Rakutenheadquartered in Silicon Valley
Slice Intelligence has the most comprehensive online shopping data anywhere.Over 400 million consumer purchases processed$5 billion in purchasesleadership team is a mix of technology veterans and market research experts
Slice Intelligence Confidential - Do not copy or distribute without prior written consent
Slice Intelligence Confidential Do not copy or distribute without prior written consentSliceYour Smart shopping assistantTrack shipments automaticallyStore shopping receiptsSave money with price drop alertsStay safe with recall alerts1234
Slice Intelligence Confidential Do not copy or distribute without prior written consentUnroll.meClean up your inboxUnsubscribe with one clickCombine favorite subscriptions into one emailRead what you want, when you want123
Slice Intelligence Confidential Do not copy or distribute without prior written consent
email receipt based data insightsThe secret sauce TO UNIQUE PURCHASE INSIGHTS
Receipt DiscoveryField ExtractionData Science
Market Research
Womens Clothing
MensClothing
Slice Intelligence Confidential Do not copy or distribute without prior written consentusing machine learning to build the purchase graphPanasonic 35 - 100mm f / 2.8 X OIS Samsung Galaxy S4 SPH - 16GB - Purple
Slice Intelligence Confidential Do not copy or distribute without prior written consentusing machine learning to build the purchase graphPanasonic 35 - 100mm f / 2.8 X OIS
Samsung Galaxy S4 SPH - 16GB - Purple
Electronics & Accessories Cameras & PhotoCamera Lenses
Brand: Panasonic Focal Length: 35 - 100mm Aperture: 2.8Electronics & Accessories Telephones & Mobiles Mobile Phones
Brand: SamsungMemory: 16GBSub-brand: GalaxyColor: Purple Model: S4
Slice Intelligence Confidential Do not copy or distribute without prior written consentmeasurement science at slicebringing statistical validity to analyticsso we can talk about the US online shopping population, not just the slice panel
recognize:no data is perfectly completeno collection methodology is perfectly errorlessno sample is perfectly representative
weighting and projection can help with the last partmold our subset into a shape that better resembles the population
Slice Intelligence Confidential Do not copy or distribute without prior written consentweighting & projectionperform a targeting surveysmall but very-high qualityattach characteristics to your sampleand compute from your dataweight your sample to match targetsover-represented profiles get a lower weight; vice-versatypically in terms of demographics, and other e.g. behavioral marginsuser classification and panel eligibility also important
Slice Intelligence Confidential Do not copy or distribute without prior written consentvalidated by industry sources
amazon.com correlation to slice data: .99 overall correlation to slice data: .98
Slice Intelligence Confidential Do not copy or distribute without prior written consentSlice Confidential - Do not copy or distribute without prior written consent14
our data in action!news and applications
Slice Intelligence Confidential Do not copy or distribute without prior written consentapple watch dataslice was able to measure the number of people who bought an apple watch, how many they bought, what they spent, and when they bought them
Slice Intelligence Confidential Do not copy or distribute without prior written consentapple customers held outprior to apple watch launch:
only 1.5% apple customers* had purchased a smartwatch
vs. 4.6% of samsung (mobile) customers
*customer = bought an apple/samsung mobile product in past 2 years
Slice Intelligence Confidential Do not copy or distribute without prior written consent
purchasing flurry in first hour
957,000 people first day, 1,039,000 first week~1.3 watches per person, 1.25M watches first day (~800k in first hour, 62%) 62% were cheapest sport model
Slice Intelligence Confidential Do not copy or distribute without prior written consentslice data answers core business questionsfor whatever youre interested in selling:at what price? color? capacity? configuration? whats growing? how does it compare to how youre growing?
what opportunities am I missing?am I missing emerging product/category opportunities?how much revenue am I bleeding to competitors?am I under or over-charging for certain products?am I under or over-charging for shipping and handling?am I under-penetrated in certain categories
what roads should I not take?what brands, products, or categories are just not selling well online?what promotions dont work?what prices dont work?
Slice Intelligence Confidential Do not copy or distribute without prior written consentpurchase graphnodes are what and where people buy:retailers they shop frombrands they buy fromauthors they readproducts they purchase
edges are the similarities between nodesi.e. two nodes are similar if people who buy the first also buy the latter (correlation, not causation)
Slice Intelligence Confidential Do not copy or distribute without prior written consentitem level consumer purchase dataAcross merchants, channels and time
ECOMMERCE
Package Out for DeliveryiPad Air 2, Gold, 64GB, $599.00
TRAVELDepart San Francisco, CA7:30pm, Fri, April 14, 2015
Arrive Los Angeles, CA9:00pm, Fri, April 14, 2015
PAYMENTSPayment ConfirmationLarge Non-Fat Mocha to La Stacione Cafe
OFFLINE RETAIL
TICKETSReservation for VarekaiSat, Mar 11, 2015 @7:15pm
Purchased Wed, Mar 8, 2015Cashmere Blend Scarf, $98.00
Slice Intelligence Confidential Do not copy or distribute without prior written consent
how to define similarity?
intuitively, this would be the number of people who buy both products divided by the popularity of each product.
for example:
75 users buy brand A50 users buy brand B35 users buy both
Slice Intelligence Confidential Do not copy or distribute without prior written consenthave an example with numbers venn diagram and cos similarity value, talk about amazon21
understanding a graph visualization
very strong correlationweak/no correlationstrong correlationtwo nodes must be directly connected to be similar.the smaller the edge, the more similar the nodes.
Slice Intelligence Confidential Do not copy or distribute without prior written consent
22
ok, lets see one in action
Slice Intelligence Confidential Do not copy or distribute without prior written consent
ok, lets see one in action
Slice Intelligence Confidential Do not copy or distribute without prior written consent
ok, lets see one in action
Slice Intelligence Confidential Do not copy or distribute without prior written consent
ok, lets see one in actionAmazonwe call Amazon a hub(largest # of edges)
Slice Intelligence Confidential Do not copy or distribute without prior written consent
ok, lets see one in actionAmazonAppleiTunesPayPaleBayTargetGrouponLiving SocialNetflixGoogle PlayBest BuyNeweggWalmartZapposWootMonoprice.com
Slice Intelligence Confidential Do not copy or distribute without prior written consentColor edges and labels by category27
ok, lets see one in action
Slice Intelligence Confidential Do not copy or distribute without prior written consent
ok, lets see one in action
Gapgap is the apparel hub
Slice Intelligence Confidential Do not copy or distribute without prior written consent
birds of a feather cluster togetherGapMacysNordstromJ. CrewBloomingdalesSaks Fifth AvenueNeiman MarcusBanana RepublicL.L. BeanRalph LaurenAeropostaleExpressLevisDermStoreSkinStoreULTADrugstoreSephora
LOFTAnn TaylorOld Navygap is the apparel hub
Slice Intelligence Confidential Do not copy or distribute without prior written consent
what connects the two hubsAmazonGap
Slice Intelligence Confidential Do not copy or distribute without prior written consentIf Amazon wants to reach Gap users, they should advertise through Zappos31
what connects the two hubsAmazonGapZapposOld NavyTargetAmazon could reach Gap customers by advertising on Zappos
Slice Intelligence Confidential Do not copy or distribute without prior written consentIf Amazon wants to reach Gap users, they should advertise through Zappos32
so how could we leverage this?recommendation engineadvertising/marketinguser segmentationsearch engine optimization
Slice Intelligence Confidential Do not copy or distribute without prior written consentin conclusionmachine learning + measurement science leads to big data insights
need both to have high quality, structured, representative data
Slice Intelligence Confidential Do not copy or distribute without prior written consent35questions?
visit slice.com or email [email protected]@slice.com
or follow us on twitter:@adataalchemist@bayesybrad@sliceintel
Slice Intelligence Confidential Do not copy or distribute without prior written consent