Anzeige
Anzeige

Más contenido relacionado

Similar a How GetNinjas uses data to make smarter product decisions(20)

Anzeige

How GetNinjas uses data to make smarter product decisions

  1. Snowplow Meetup @ GetNinjas Bernardo Srulzon bernardo@getninjas.com.br
  2. How does it work? 1. Tell us what service you need 2. We’ll match you with up to 4 pros 3. See the reviews and hire the best :)
  3. OVERVIEW 1,500,000 services requested in the last 12 months 200,000 registered pros R$ 190M+ in transactions (GMV), last 12 months 100+ categories, from electricians to wedding photographers
  4. OVERVIEW Each dot is a “mini GetNinjas”, with 100 categories 300 main cities 100 categories x 30k combinations for which we have to maintain a healthly supply vs. demand balance
  5. #1 In this context, what should be the role of Business Intelligence?
  6. We’re focused on creating a great experience for clients and pros; to achieve that, we measure each step of the customer lifecycle Pirate Metrics Pirate Metrics by Dave McLure @ 500 Startups
  7. We’re focused on creating a great experience for clients and pros; to achieve that, we measure each step of the customer lifecycle Pirate Metrics Are we balancing supply and demand correctly? What acquisition channels should are more cost-effective? Pirate Metrics by Dave McLure @ 500 Startups
  8. We’re focused on creating a great experience for clients and pros; to achieve that, we measure each step of the customer lifecycle Pirate Metrics Are we matching clients with great pros? Can pros close their first service a week after signup? Pirate Metrics by Dave McLure @ 500 Startups
  9. We’re focused on creating a great experience for clients and pros; to achieve that, we measure each step of the customer lifecycle Pirate Metrics Will clients come back to request a service in a different category? Are pros consistently interacting with clients & closing services? Pirate Metrics by Dave McLure @ 500 Startups
  10. We’re focused on creating a great experience for clients and pros; to achieve that, we measure each step of the customer lifecycle Pirate Metrics Do clients share the experience and bring new clients? Are successful pros referring us to their colleagues? Pirate Metrics by Dave McLure @ 500 Startups
  11. We’re focused on creating a great experience for clients and pros; to achieve that, we measure each step of the customer lifecycle Pirate Metrics Is the lead price fair, and bring significant ROI to pros? Are we extracting the maximum value from our leads? Pirate Metrics by Dave McLure @ 500 Startups
  12. If the Business Intelligence role is to answer complex questions, the traditional mindset doesn’t really work... A more traditional BI team would... Collect data Build reports Send report to decision-maker Sense of ownership Business impact Autonomy Out-of-the-box thinking
  13. Our role is to transform data into actionable insights; in practice, we start from a hypothesis and participate in the entire decision process Data science only matters if data drives action Jeremy Stanley, VP Data Science @ Instacart Hypothesis Exploration & Validation Decision Team start from a hypothesis We should adjust our pricing so that pros can close at least one job on their first month in the platform. This should increase our retention. Come up with a plan to validate it Analyzing historical data for closed jobs and applying a statistical model, we should be able to determine the pricing options Participate in decision-making We concluded that the least expensive plan should allow pros to purchase 30 leads. We’ll move forward with the implementation and monitor results
  14. #2 From 20k to 170k service requests How did we scale?
  15. In the past 3 years, we evolved our focus and structure to reflect the company priorities 2014 2 Focus Product-market-fit: making sure our product actually solves a market pain point • Google Analytics setup • First version of the matching algorithm • Analysis focused on transactional data
  16. In the past 3 years, we evolved our focus and structure to reflect the company priorities 2014 20152 4 Focus Product-market-fit: making sure our product actually solves a market pain point • Google Analytics setup • First version of the matching algorithm • Analysis focused on transactional data Focus Have a deeper understanding of how clients and pros interact with the platform. Question the details. • Got stuck with GA limitations, went forward with Snowplow • Invested a lot of energy in training the team • Split the team workload into acquisition & closing
  17. In the past 3 years, we evolved our focus and structure to reflect the company priorities 2014 2015 2016-2017 Focus Product-market-fit: making sure our product actually solves a market pain point Focus Have a deeper understanding of how clients and pros interact with the platform. Question the details. Focus Build a comprehensive view of the user, and democratize access to data 2 4 10 • Google Analytics setup • First version of the matching algorithm • Analysis focused on transactional data • Got stuck with GA limitations, went forward with Snowplow • Invested a lot of energy in training the team • Split the team workload into acquisition & closing • Integration with customer support & sales systems • Team structured in multi- functional “squads” • More complex matching and marketing algorithms
  18. #3 Web analytics + Data warehousing
  19. We deep dive into web analytics data to understand the behavior of clients and pros What’s the conversion rate of this page?
  20. We deep dive into web analytics data to understand the behavior of clients and pros What’s the conversion rate of this page? Desktop converts more than mobile?
  21. We deep dive into web analytics data to understand the behavior of clients and pros What’s the conversion rate of this page? Desktop converts more than mobile? State capitals convert more vs. countryside?
  22. We deep dive into web analytics data to understand the behavior of clients and pros What’s the conversion rate of this page? Desktop converts more than mobile? State capitals convert more than countryside? What % of users load in < 5 sec?
  23. We deep dive into web analytics data to understand the behavior of clients and pros What % interact with the form?
  24. We deep dive into web analytics data to understand the behavior of clients and pros What % interact with the form? How much time do they take?
  25. We deep dive into web analytics data to understand the behavior of clients and pros What % interact with the form? How much time do they take? How many have validation issues?
  26. We deep dive into web analytics data to understand the behavior of clients and pros What % interact with the form? How much time do they take? How many have validation issues? Are recurrent users any different?
  27. To answer these questions, we need a web analytics structure that allows a very granular segmentation of data Channel Landing page type Device A/B Test Category Funnel Load TimeCity Segmentation is key! Tech Events Classes Consulting What’s the site conversion rate? WiFi/3GRecurring All data in aggregate is crap Avinash Kaushik
  28. Trackings generate insights into the behavior of the group, and the behavior of each user Where are the frictions and what are the hypotheses? Visit Form interaction 2nd step 100% Interaction on 2nd step Conversion! 50% 40% 20% 30% How’s the funnel evolving over time? What’s the group behavior?
  29. Trackings generate insights into the behavior of the group, and the behavior of each user Sign-up (SEO) Purchased a lead Received a review Downloaded app Purchased credits Received onboarding call E-mail marketing D+0 D+1 D+5 D+7 D+20 D+25 D+30 Plan (re)marketing campaigns Event-based communication What do users do before downloading app? More info to sales team What’s the behavior of a specific user? Debugging Where are the frictions and what are the hypotheses? Visit Form interaction 2nd step 100% Interaction on 2nd step Conversion! 50% 40% 20% 30% How’s the funnel evolving over time? What’s the group behavior?
  30. These complex questions eventually led us to hit the wall with Google Analytics, and we went out exploring alternatives Hard to segment data correctly No form tracking Hard to identify users Limited cross-device tracking Hard to integrate with other sources Can’t apply our business logic Some of the limitations...
  31. These complex questions eventually led us to hit the wall with Google Analytics, and we went out exploring alternatives SEM SEO Direct Direct Direct Direct Direct SEM SEO SEO SEO SEO SEO SEO Session #1 #2 #3 #4 #5 #6 #7 Channel attribution Reality GA was reporting 50% less traffic compared to what we expected Hard to segment data correctly No form tracking Hard to identify users Limited cross-device tracking Hard to integrate with other sources Can’t apply our business logic Some of the limitations...
  32. We decided to implement Snowplow, an open source platform for product analytics Snowplow is an enterprise-strength marketing and product analytics platform https://github.com/snowplow/snowplow Identifies users, and tracks the way they engage with the site & app Stores your users' behavioral data in a scalable "event data warehouse" Leverage BI & big data tools to analyze data We want to own our data We want to track users on the web and on the app We want to have a comprehensive view of users We want to answer EVERY question! Why?
  33. We decided to implement Snowplow, an open source platform for product analytics Snowplow is an enterprise-strength marketing and product analytics platform https://github.com/snowplow/snowplow Beanstalk EMR Redshift Redshift Identifies users, and tracks the way they engage with the site & app Stores your users' behavioral data in a scalable "event data warehouse" Leverage BI & big data tools to analyze data
  34. ...at a much lower cost than any other SaaS alternatives
  35. We decided to implement Snowplow, an open source platform for product analytics Apps SMS CRM/Sales Payments + Transactional data Sync MySQL -> Redshift via Amazon DMS Web Web/App analytics via Snowplow Push Support Server-side events via Snowplow External integrations via Stitch DATA MODEL 300M events per month Transforms atomic, unopinionated data into models that have business logic applied Email
  36. BEFORE | Google Analytics aggregate conversion data
  37. AFTER | Snowplow highly segmented data Landing page #1 Landing page #2 Landing page #3 Tech Support Events Home Services Home Improve ment Others
  38. Redshift e Tableau work great together – connections are super fast and allow fine segmentation Billions of Redshift data available for drag&drop segmentation Integration with other data sources (CSV, Excel, MySQL, etc) Different ways to visualize your data Flexibility for customized calculations & formulas
  39. Metabase | Online dashboards https://github.com/metabase/metabase
  40. #4 Team structure & profile
  41. We use data to make business decisions, but also to build models & algorithms Decision Science 2014-2015 2016 Centralized team Product-market-fit: making sure our product actually solves a market pain point Analysts within cross-functional teams Build a complete view of the user, and democratize access to data Focused on making smarter business decision based on data. Overlaps with the product manager role Skills Business + SQL + Excel + Tableau + Basic Programming Profile Intern from top Engineering schools 5 Data Science5 Focused on developing models & algorithms to provide a better customer experience Skills Business + Modelling + Programming Profile Masters/PhD + few years work experience
  42. We’re structured in cross-functional teams, with the specialists meeting every week to exchange good practices Tech Product Design BI Content “Specialist alignment” every week Cross-functional teams Focused on specific OKRs
  43. #5 What’s in the future?
  44. We should increasingly empower other areas to create & validate hypotheses, and make smarter business decisions Business Intelligence Build & maintain the data pipelines/infrastructure Model data, applying business logic Offer training & coaching Continue to create and explore hypotheses Product Marketing Tech Support & Sales Decisions focused on the clients’ and pros’ experience with the app Budget allocation, attribution model, return- on-investment Page speed, debugging, cache hits Team performance, commission model, lead scoring
  45. Snowplow Meetup @ GetNinjas Bernardo Srulzon bernardo@getninjas.com.br
Anzeige