Wave Analytics: Developing Predictive Business Intelligence Apps

Wave
Analy*cs:

Developing
Predic*ve
Business
Intelligence
Apps

Built
on
the
Salesforce1™
Pla?orm

Quote
Contract
Revenue
Opportunity
ERP

CONTRACT

LIFECYCLE

MANAGEMENT

ORDER

MANAGEMENT
CONFIGURE

PRICE

QUOTE
BILLING

REBATE

MANAGEMENT

DEAL
MANAGEMENT

E-‐COMMERCE

QUOTE-‐TO-‐CASH
INTELLIGENCE

X-‐AUTHOR

ApQus
Quote-‐to-‐Cash
Footprint

Why
Predic*ve
is
Important…

• IoT
and
the
Age
of
Data

• New
Tools
are
needed

•  Descrip7ve

•  Predic7ve

•  Prescrip7ve
Analy7cs

• Ac7on
from
Insight

The
Ways
Wave
helps…

• Visualizing
descrip7ve
and
predic7ve
analysis

• Delivering
insights
to
Salesforce
users

If
you
have
one
of
these…..

Then
you
might
like
these…

Story:
Growing
Revenue
and
Margin

•  Business
wants
to
grow
revenue
with
cross-‐sell
and
up-‐sell

recommenda7ons
based
on
data
analysis

•  Marke7ng
uses
Wave
to
es7mate
the
poten7al
opportunity

•  Sales
rep
receives
a
recommenda7on
for
her
customer
from
ApHus
that
will

increase
the
size
of
the
deal

•  Wave
provides
the
Sales
rep
with
addi7onal
insight
into
the

recommenda7on
for
the
customer
which
helps
her
be
more
consulta7ve

•  Marke7ng
uses
Wave
to
track
the
business
impact
of
the

recommenda7ons

which
convert
into
sales

Add
Products
to
Cart

Browse
Recommended

Products

Adds
Recommended

Products
to
Cart

Analyze
Purchase

History
of
Similar

Customers

Predict
Recommended

Products

ApQus
Use
Case

Display
chart
depic7ng

Recommended
Products
by

Customer
cluster

Display
chart
depic7ng

$

from
Recommenda7ons

added
to
cart

Salesforce

Database

Frequency

Similarity

Predic7on

Marke7ng

Sales

Wave

Database

Explain
Product

Recommenda7ons

Wave
API

Recommender
API

ApHus
CPQ

ApHus
Cross-‐Sell

Recommender

Wave

Machine-‐Learning
Algorithm

Algorithm
–  ML uses statistics to analyze and predict patterns in data
–  Item-based Collaborative Filtering
–  Initial model developed in “R” uses KNN (K Nearest Neighbor)
–  3 main components of the algorithm
▪  Frequency
▪  Similarity
▪  Prediction
–  Capable of unsupervised learning
Frequency Table
–  Count number of Products (items) purchased by each Account

Recommenda*on

•  Select products (items) which are similar to the product added to the Cart
•  Given an account, choose top-N products among the similar products based on
the estimated rates
•  Products 15 & 20 are recommended to the user
Added Product 1
to the Cart
Similar Products
Product 3
Product 8
Product 10
Product 15
Product 20
User
Prediction Rate
0.5
1
0.3
2
3
User

The
Code

public void callWaveAPI() {
Http http = new Http();
HttpRequest req = new HttpRequest();
HttpResponse res = new HttpResponse();
// Wave REST API calls
String WAVE_BASE = 'https://gs0.salesforce.com/services/data/v34.0/wave';
String WAVE_DATASETS = WAVE_BASE + '/datasets';
String WAVE_QUERY = WAVE_BASE + '/query';
try {
// send Wave API request to query dataset
req = new HttpRequest();
// set request endpoint
req.setEndpoint(WAVE_QUERY);
req.setMethod('POST');
// set request headers
req.setHeader(HEADER_AUTHORIZATION, HEADER_AUTHORIZATION_PREFIX + ' ' + oAuthAccessToken);
req.setHeader(HEADER_CONTENT_TYPE, HEADER_CONTENT_TYPE_JSON);
// set request body
body = '{"query":"q = load "0FbB00000005ZQPKA2/0FcB00000006PlCKAU"; q = foreach q generate 'Prediction_Rate' as 'Prediction_Rate',
'Recommended_Product_Name' as 'Recommended_Product_Name', 'Related_Product_1_Name' as 'Related_Product_1_Name', 'Related_Product_2_Name' as 'Related_Product_2_Name'; q
= order q by 'Prediction_Rate' desc; q = limit q 100;"}';
req.setBody(body);
// send request
res = http.send(req);
// parse JSON response using JSON2ApexQueryResult2 class
queryResult = JSON2ApexQueryResponse2.parse(res.getBody());
List<JSON2ApexQueryResponse2.Records> records = queryResult1.Results.records;
for (JSON2ApexQueryResponse2.Records r : records) {
debugStr2 += 'Recommended = ' + r.Recommended_Product + '; Related 1 = ' + r.Related_Product1 + ', Related 2 = ' + r.Related_Product2 + 'n';
}
} catch(System.CalloutException ex) {
system.debug(LoggingLevel.INFO,'Callout error: '+ex);
system.debug(LoggingLevel.INFO,res.toString());
}
}

The
Code

// set request body
body = '{"query":"q = load
"0FbB00000005ZQPKA2/0FcB00000006PlCKAU"; q =
foreach q generate 'Prediction_Rate' as
'Prediction_Rate', 'Recommended_Product_Name' as
'Recommended_Product_Name', 'Related_Product_1_Name
' as 'Related_Product_1_Name',
'Related_Product_2_Name' as 'Related_Product_2_Name
'; q = order q by 'Prediction_Rate' desc; q = limit
q 100;"}';

The
Code
/**
* compute prediction score for a given product and account
* @param accountId the id of the account to get a prediction score for
* @param productId the id of the product to get a prediction score for
* @return the prediction score for the given account and product
*/
private Decimal computePredictionScore(ID productId, ID accountId) {
Decimal score = 0.00000;
Decimal sumOfSim = 0.00000;
Decimal sumOfRateSim = 0.00000;
// get product similarity map for the given product
Map<ID,Decimal> simMap = productSimMap.get(productId);
if (!productSimMap.containsKey(productId)) {
return score;
}
// iterate over products in product similarity map
for (ID prodIdKey : simMap.keySet()) {
// get the product similarity
Decimal productSim = simMap.get(prodIdKey);
// keep track of the sum of similarties
sumOfSim += Math.abs(productSim);
// keep track of the sum of zrate similarities
if (productZRateMap.containsKey(prodIdKey)) {
Map<ID,Decimal> productZRate = productZRateMap.get(prodIdKey);
if (productZRate.containsKey(accountId)) {
Decimal accountSim = productZRate.get(accountId);
sumOfRateSim += productSim * accountSim;
}
}
}
// calculate the prediction score
if (sumOfSim != 0.00000) {
// get product average and std deviation
Map<String,Decimal> productStats = productStatsMap.get(productId);
Decimal productAvg = productStats.get(KEY_AVG);
Decimal productStd = productStats.get(KEY_STD);
// calculate the score
score = productAvg + (productStd * (sumOfRateSim / sumOfSim));
}
// return the prediction score
return score;
}

Lessons
Learned

•  Leveraging
Wave

•  Use
Salesforce
to
Wave
data
ﬂows
to
store
historical
data
and
trends

•  Use
Wave
API
to
query
for
recommended
products

•  Use
Wave
dashboards
for
visualiza7on
of
recommenda7ons

•  Implemen7ng
Machine
Learning
on
Salesforce

•  With
Apex-‐
breakup
into
batch
process
to
stay
within
governor
limits

•  With
Heroku
–
fewer
restric7ons
but
introduces
data
security
challenges

•  Changes
to
the
model
once
coded
are
diﬃcult
and
present
design

considera7ons
for
produc7za7on

Wave Analytics: Developing Predictive Business Intelligence Apps

Recommended

More Related Content

What's hot (20)

Similar to Wave Analytics: Developing Predictive Business Intelligence Apps (20)

More from Salesforce Developers (20)

Recently uploaded (20)

Wave Analytics: Developing Predictive Business Intelligence Apps