Skip to content

Instantly share code, notes, and snippets.

@OleksandrKucherenko
Created October 21, 2022 12:04
  • Star 0 You must be signed in to star a gist
  • Fork 0 You must be signed in to fork a gist
Star You must be signed in to star a gist
Embed
What would you like to do?
The template of Experiment definition for Data Driven approaches. ref: https://medium.com/@olku/my-data-driven-approach-framework-900e8ac710cd

Framework

Step Questions
Change Request Why? What? How?
Confirmation Metrics, Success/Failure criteria;
Execution Given/When/Then? How Long? How Many? What is Required? Inputs/Outputs/Trends?
Decisions and Artifacts Automated metric grabber (per hour, per day); Side effects

Example

title: WHAT - “How might we increase the number of orders?”

STEP 1: Change Request

title: WHY: "We need business to grow."

  `````ad-info
  title: WHAT: "Increase the number of sales per week."
  
    ````ad-question
    title: HOW:
    color: 200, 200, 200      
    ```ad-info
    title: Assumptions: Results of the brainstorm
    - [x] A1: Web traffic can be converted to sales
    - [ ] A2: Customer reviews can trigger sales
    - [ ] A3: Discount with an expiration date can trigger sales
    - [ ] A4: Payback with an expiration date can trigger sales
    - [ ] A5: Instagram train/giveaway can trigger sales
    - [ ] A6: ...continue...
    ```

    ```ad-note
    title: Voting Results: selected A1 as the primary action
    ```
    ````
  `````

STEP 2: Confirmation

title: Confirmation 

  ``````ad-faq
  title: METRICS:
  - [ ] M1: number of page views/impressions
  - [ ] M2: number of orders
  - [ ] M3: $$$ order amount
  - [ ] M4: number of drops/exits/cancelations
  - [ ] M5: $$$ cost of the customer
  - [ ] M6: conversion rate (%)
  ``````

  ``````ad-info
  title: Side effects/artifacts:
  - [ ] E1: Why was the order canceled? (examples: Price too high, not proper quality, ...)
  - [ ] E2: Where does the user exit from the "making order" sequence? (examples: too many options in order, on payment step, on personal details, ...)
  - [ ] E3: Session duration
  - [ ] E4: Bounce Rate
  - [ ] E5: New vs. Returning visitor
  - [ ] E6: Average pages/steps per session 
  - [ ] E7: Cost of Execution
  ``````

  `````ad-faq
  title: Success/Failure criteria
  - [ ] Fail. Customer cost is too high (>15$ per customer)
  - [ ] Success. Conversion rate >10%
  - [ ] Success. Customer cost is low (<5$ per customer)
  - [ ] Success. Too Many orders (stop the experiment if needed, reduce the size)
  `````

  `````ad-question
  title: Should we execute it?
  - [ ] is the experiment atomic?
  - [ ] are risks of failure identified (roadblocks)? What can we change to reduce risks?
  - [ ] is implementation time short?
  - [ ] can we score the test's priority/value/impact?
  `````

STEP 3: Execution

title: Execution

  ``````ad-important
  title: Required for execution: select one option only
  - [ ] Organic traffic
  - [ ] Purchased traffic (ads networks)
  - [ ] Purchased traffic (3rd party, freelance)
  - [ ] Purchased traffic (social networks ads)
  ``````

  ``````ad-info
  title: Duration: 1 week (or till we reach 10'000 impressions)
  ``````

  ``````ad-info
  title: Size: 10'000 impressions
  ``````

  ``````ad-cite
  title: Test
    `````ad-info
    title: GIVEN
    - [ ] website is ready to accept traffic in such an amount
    - [ ] business is ready to be shortly overloaded (too many orders)
    - [ ] website proper functioning in "order sequence"
    - [ ] metrics collected by analytics from all pages (M1-M...)
    - [ ] can we divide the experiment into two equal groups? (One group **without** experiment (clean group), another with experiment (test group))
    `````
    
    `````ad-info
    title: WHEN
    - [ ] clean run (no other experiments executed)
    - [ ] metrics extracted every hour and saved to the database
    - [ ] tested something that can trigger sales (texts, graphics, layouts, colors, etc.)
    `````
    
    `````ad-info
    title: THEN
    - [ ] collected metrics are visualized with calculated trends
    - [ ] assumption on trend defined for detecting anomaly behavior 
    - [ ] comparison to the group clean group results visualized
    `````
    
  ``````

STEP4: Decisions and Artifacts

title: Decisions (Stop|Adjust|Scale up)

  `````ad-question
  title: Is Success? How can we apply this in scale?
  - [ ] convert experiment to a business rule
  - [ ] identify failure criteria that should switch off this business rule 
  `````

  `````ad-question
  title: Is Failure? What is the reason of failure?
  color: 255, 0, 0
  - [ ] clarify that no outside forces are involved in test (for example: infrastructure failure)
  `````

  `````ad-question
  title: Results are unclear? What influenced the experiment? Can we eliminate the side risks and repeat the experiment?
  color: 255, 251, 45
  - [ ] test is not atomic
  - [ ] test does not produce a statistically significant difference
  - [ ] test results are under the influence of outside force
  `````

  `````ad-info
  title: Artifacts
  color: 48, 96, 255
  - [ ] Metrics collection tech should become a part of MAIN code
  - [ ] Any findings should be documented and become a part of the backlog
  `````

@OleksandrKucherenko
Copy link
Author

data-driven-framework-single-file-v1

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment