Adaptive Pricing with Machine Intelligence
1 like3,527 views
This document discusses using machine learning and fuzzy logic for adaptive pricing in e-commerce. It describes fuzzy logic as an option for adaptive pricing that uses intuitive business rules rather than large datasets. Fuzzy logic allows pricing to vary smoothly based on multiple factors like product popularity, user recency, and market penetration. The document provides examples of how these input factors could map to an output discount percentage using fuzzy rules and explains how fuzzy logic could integrate with store software for adaptive pricing.
Adaptive Pricing with Machine Intelligence
- 1. A D A P T I V E P R I C I N G W I T H M A C H I N E I N T E L L I G E N C E MTL+ECOMMERCE
- 2. E VA N P R O D R O M O U β Founder & CTO, Fuzzy.io β Former CTO, Breather β Founder, StatusNet β Founder, Wikitravel
- 3. W H O I S T H I S TA L K F O R ? β Involved in e-commerce β βproducts or services on the Web or mobileβ β Technical understanding β Decision-making power
- 4. W H AT I S β A D A P T I V E P R I C I N G β ? β’ Changing the price of a product β’ Based on the situation β’ User attributes β’ Product attributes β’ Business attributes
- 5. W H Y A D A P T I V E P R I C I N G ? β Profit maximization β Competition β Many large retailers use it β Guide user behaviour β Activation β Retention β Referral
- 6. R I S K S β Too high = donβt convert β Too low = cut into margin β May be worthwhile to activate a customer β Perception of fairness
- 7. I M P L E M E N TAT I O N O P T I O N S β Procedural code β Markets β Machine learning β Fuzzy logic
- 8. IF THEN ELSEIF ELSE IF THEN ELSEIF ELSE IF THEN ELSEIF ELSE IF THEN ELSEIF ELSE IF THEN ELSE IF THEN ELSE IF THEN ELSE IF THEN ELSE IF THEN ELSE IF THEN ELSE IF THEN ELSE IF THEN ELSE IF THEN ELSEIF ELSE IF THEN ELSEIF ELSE IF THEN ELSEIF ELSE IF THEN ELSEIF ELSE IF THEN ELSE IF THEN ELSE IF THEN ELSE IF THEN ELSE IF THEN ELSE IF THEN ELSE IF THEN ELSE
- 9. P R O B L E M S W I T H P R O C E D U R A L C O D E β Gets very complicated with multiple inputs β Brittle β Hard to debug β Hard to maintain β Thresholds
- 11. M A R K E T- B A S E D S O L U T I O N S β Require a market β Require something close to real-time bidding β Require fungible product or service β One seller is equivalent to another
- 13. M A C H I N E L E A R N I N G β Requires corpus of training data β May not be collected β May be difficult to experiment β Requires training process β Unintuitive results β Harder to audit β Staff are expensive
- 14. F U Z Z Y L O G I C β Fuzzy sets β Intuitive categories like βoldβ, βnewβ, βgoodβ, βwarmβ β Degrees of membership β 0 to 100% β Real-world wisdom β IF userAge IS new THEN discount IS high
- 15. F U Z Z Y L O G I C F O R A D A P T I V E P R I C I N G β’ Pros β Uses explicit business rules β Doesnβt require large training corpus β Smoothly-varying output β no discontinuities with thresholds β Handles contradictions well β Adding and removing inputs well β Missing data works well β Easier to audit
- 16. F U Z Z Y L O G I C F O R A D A P T I V E P R I C I N G β’ Cons β’ Requires numerical inputs
- 17. D I S C O U N T β Not a fixed price β Can use the same agent for multiple products β 0% = full price, 100% = free β Bounded to prevent outrageous prices
- 18. W H AT FA C T O R S C A N A F F E C T P R I C E ?
- 19. P R O D U C T P O P U L A R I T Y β Sales/week β Smoothes over variations by day-of-week β Ideally, pre-calculated for previous week
- 20. C AT E G O RY P O P U L A R I T Y β Similar to product popularity, but for product category
- 21. S I T E P E R F O R M A N C E β Site-wide sales for the week β Can be in dollars, or # of sales β Very site-specific
- 22. S A L E S P E R H O U R O F W E E K β Discrepancies between weekday/ weekend, night/day
- 23. S A L E S P E R W E E K O F Y E A R β Especially for seasonal products β Best for established stores β At least one year of sales!
- 24. U S E R R E C E N C Y β How long ago did the user sign up?
- 25. U S E R A C T I VAT I O N β Number of sales or dollars
- 26. M A R K E T P E N E T R AT I O N β For geographical markets β In number/million
- 27. O T H E R FA C T O R S β Influencer β Number of followers on Twitter β Number of friends on Facebook β Social network penetration β Percentage of followers on Twitter who have joined β Percentage of friends on Facebook who have joined
- 28. R U L E S β Map input factors to output discount β Usually linear β Occasionally inversely linear β More complex rules possible
- 29. I N T E G R AT I N G W I T H S T O R E S O F T WA R E β Using an SDK β Or a plugin
- 30. F U Z Z Y L E A R N I N G β In production β Feedback loop based on profit margin on the sale β 0% = no conversion β Varies fuzzy set boundaries β Varies weights of fuzzy rules
- 31. T H A N K S Evan Prodromou evan@fuzzy.io https://fuzzy.io/