Lightweight Business Intelligence in Ruby

Long the provence of specialists in the hermetic world of enterprise software development, business intelligence (BI) is increasingly important to smaller, more agile companies and startups who need access to near-real-time information to make critical business decisions. With its support for aggregation and reduction of massive amounts of data and its flexible schemas, MongoDB is a great choice for creating lightweight, denormalized data stores optimized for BI, with the added bonus of peaceful co-existence with transactional data stores. In this talk I will explore how Trunk Club captures and analyzes customer information, monitors user behaviour, feeds machine-learning algorithms for decision support, and delivers value to business stakeholders through simple querying and reporting interfaces.

Transcript

1. LIGHTWEIGHT BUSINESS INTELLIGENCE COREY EHMKE APRIL 2013
2. WHO AM I? A developer with a long memory & a longer history An active open source contributor A senior engineer at Trunk Club A lifelong learner
3. Service-oriented startup Software optimizes and streamlines business processes Technology is a key differentiator Engineering provides leverage for scaling the business TRUNK CLUB
4. OUR ENGINEERING GOALS Startups make critical decisions on a daily basis Better data leads to better decision-making Our mission is ∴ to provide this data in a timely and useful form
5. SOME CRITICAL DATA POINTS Marketing campaign performance Member on-boarding funnel Trunk lifecycle Stylist interactions Product performance
6. BUSINESS INTELLIGENCE: THREE APPROACHES
7. WHAT IS BUSINESS INTELLIGENCE? Collection & organization of mission-critical knowledge Historical view of business operations Tools to support decision making
8. THE NAÏVE APPROACH
9. THE NAÏVE APPROACH Reporting out of the transactional database Raw SQL embedded in your code for “performance” Granting direct db access to stakeholders
10. THE NAÏVE APPROACH: SHORTCOMINGS Fighting the schema with complex joins Poor performance Impact on production resources
11. DON’T TRUST THIS GUY WITH YOUR DATABASE.
12. THE ENTERPRISE APPROACH
13. THE ENTERPRISE APPROACH Distinct BI database Nightly ETL (extract, transform, load) Schema designed for reporting Separate hardware and software stack Combination of static and dynamic reports
14. THE ENTERPRISE APPROACH: SHORTCOMINGS 24 hour delay in information Expensive to configure and maintain Requires highly specialized resources Hard to change your mind or adapt Enterprise-y
15. VERY, VERY ENTERPRISE-Y
16. DON’T TRUST THESE GUYS WITH YOUR DATABASE, EITHER.
17. TRADITIONAL BI: THE WRONG HAMMER
18. TRADITIONAL BI: THE WRONG HAMMER Waterfall approach Painful data migrations Brittle ETL processes No automated testing Logic embedded in the data store
19. IN SHORT... TRADITIONAL BI DOESN’T FEEL AGILE.
20. A LIGHTWEIGHT APPROACH TO BUSINESS INTELLIGENCE
21. THE FOUR NOBLE GOALS OF THE LIGHTWEIGHT BI APPROACH Provide real-time data to support decisions Leverage familiar technology & infrastructure Use existing development staff Support an iterative, agile approach
22. TRUST YOUR DEVELOPERS.
23. GETTING STARTED Collaborate! Determine KPIs that actually matter Figure out what sort of questions the business is asking on a daily basis
24. TURN INFERENCES INTO FACTS Find a way to tell a story with your data Design your schema based on facts De-normalize like a boss
25. PRESENT ANSWERS Provide a central, single source of truth Present the data wherever it’s needed Dashboard design is harder than you think Plan for iterations & ongoing collaboration
26. PLEASE GODS NO.
27. DON’T MAKE THESE.
28. BRINGING RUBY TO THE PARTY Makes agile, test-driven development easy Quickly deploy new apps Plenty of visualization libraries Powerful SQL & NoSQL ORMs Great data munging capabilities
29. LEVERAGING MONGODB Flexible and dynamic schemas Support for native datatypes Powerful querying and aggregation Fast and performant Easy to scale up
30. GO AHEAD... TRY THIS AT HOME LIGHTWEIGHT BI TECHNIQUES
31. PARALLEL DB DEPLOYMENT Modern frameworks support multiple orms Use SQL for transactions Use NoSQL for reporting Business logic in your apps, not in your database
32. STATISTICAL MODELS Collections of facts, not attributes Data spanning deep and wide object graphs De-normalized and optimized for reporting
33. STATISTICAL MODELS module  MemberDataProfiles    class  Performance        include  Mongoid::Document        include  Mongoid::Timestamps        field  :member_id,                                  :type  =>  Integer        field  :annual_trunk_frequency,        :type  =>  Integer,    :default  =>  0        field  :average_trunk_value,              :type  =>  Float,        :default  =>  0.0        field  :is_customer,                              :type  =>  Boolean,    :default  =>  false        field  :keep_rate,                                  :type  =>  Float,        :default  =>  0.0        field  :last_transaction_date,          :type  =>  Date        field  :member_creation_date,            :type  =>  Date        field  :total_value,                              :type  =>  Float,        :default  =>  0.0        field  :total_number_of_trunks,        :type  =>  Integer,    :default  =>  0    end end
34. STREAMING ETL Event-triggered, continuous data extraction Calculations are defined in code rather than SQL or ETL scripts Allow resource-intensive data munging to happen in the background Provide near-real-time data
35. STREAMING ETL #  rabbit_notifier.rb class  RabbitNotifier        def  self.notify_of_action(model,  action,  extras  =  {})            notify(                :model_actions,                  model,                headers(model,  extras.merge('action'  =>  action.to_s))            )          end ... end
36. STREAMING ETL #  listeners.rb ListenerConfig.map  do    route  :product_shipped,      :to  =>  UpdatesProduct    route  :product_returned,    :to  =>  UpdatesProduct end
37. STREAMING ETL #  updates_product.rb def  self.with(json)    product  =  Product.init_with_params(extracted_params(json))    product.update_stats! end def  self.extracted_params(json)    JSON.parse(json,  :symbolize_names  =>  true) end
38. STREAMING ETL #  product.rb class  Product    include  Mongoid::Document    include  Mongoid::Timestamps    include  Products::Calculations    field  :name    field  :price,                          :type  =>  Float,      :default  =>  0.0    field  :cost,                            :type  =>  Float,      :default  =>  0.0    field  :profit_margin,          :type  =>  Float,      :default  =>  0.0    field  :keep_rate,                  :type  =>  Float,      :default  =>  0.0    field  :quantity_shipped,    :type  =>  Integer,  :default  =>  0    field  :quantity_returned,  :type  =>  Integer,  :default  =>  0    field  :profitability,          :type  =>  Float,      :default  =>  0.0    field  :trunkability,            :type  =>  Float,      :default  =>  0.0    field  :positive_feedback,  :type  =>  Array,      :default  =>  []    field  :negative_feedback,  :type  =>  Array,      :default  =>  [] ...
39. STREAMING ETL #  product.rb  cont’d        ...    def  self.init_with_params(params={})        product  =  Product.where(:name  =>  params[:name]).first            product  ||=  Product.new(params)    end    def  update_stats!        calculate_stats        save    end end
40. APIS EVERYWHERE Provide easy access to your data Allow reuse of data in novel ways Quickly build dashboards and data explorers Use the Faceted gem to make building APIs easy
41. LIGHTWEIGHT BUSINESS INTELLIGENCE: EXAMPLES
42. LEVIATHAN
43. OUR TECHNOLOGY ECOSYSTEM
44. LEVIATHAN Records events from all applications Subscribes to all message queues Collects and displays real-time data Browse, search, & drill-down interface Longitudinal analysis with dynamic cohorts
45. LEVIATHAN: EVENT MODEL class  Event    include  Mongoid::Document    include  Mongoid::Timestamps    field  :label    field  :application    field  :details,  :type  =>  Hash,  :default  =>  {}    def  self.record!(label,  application,  params={})        Event.create(            :label  =>  label,            :application  =>  application,            :details  =>  params        )    end    def  self.search_details(criteria={})        where("details.#{criteria.keys.first}"  =>  /#{criteria.values.first}/i)    end end
53. LEADERBOARD
54. LEADERBOARD Real-time sales performance Prominently placed in sales team area Visibility = competition
56. PRODUCT INFORMANT
57. PRODUCT INFORMANT Inventory performance metrics Extracts facts from a dozen RDBMS tables Delivers on decision support
60. SUMMING UP Real-time data is valuable and possible Step outside of purely relational thinking Use the technologies you’re most familiar with
61. SUMMING UP Wrap your data in easy-to-use APIs Build micro-apps to deliver stakeholder-specific dashboards Be agile, not enterprisey
62. KEEP IN TOUCH! corey@trunkclub.com @bantik on twitter bantik.github.com trunkclub.com/corey