Introduces predictive analytics to game developers. Tips and lessons from other industries. Case studies showing 63% to 150% higher freemium conversion rates.
4. Life Cycle Management
• Communication tailored to customer’s stage:
1) Welcome & educate. (“Here’s how”)
2) Upsell. Seek referrals
3) Seek renewal, or give retention pitch
• How to know what phase they’re in?
– Sometimes, it’s easy (first-time player)
– Otherwise, predictives usually used
5. Best Practices from Telcos
These companies learned:
• More-engaged customer Easier to upsell
• More upselling Lower churn
For best results:
• Limit customer communications,
and deliver the right message at each stage
– Upselling too soon will overwhelm or annoy
– Customers are receptive during brief windows
10. “Why not just promote to everybody?”
Why does tight targeting raise total revenue?
• If you spam with conversion/upsell offers…
– Players become numbed to your messages
– Annoyed, players opt-out, or stop playing
(You’ve expedited your churn)
– Players are not focused onto the most-appropriate
message for their life-cycle stage
– You waste money (communication, discounts)
– You hurt your reputation & degrade trust
14. The Purpose of Predictives
Focusing promos on those most likely to buy/etc
• Communicate to fewer customers
– Reduce opt-outs, burn-out, churn
• Reach many of the target group
– Send the offer to those who want it
• Reach others who are similar to the targets
– Share the offer with those “on the fence”
21. Observations from AT&T Case
• The social graph – if available – helps greatly
• The combination of behavioral & SNA
outperforms the sum of their contributions
23. Pragmatics: What data?
What data is used for social-game predictives?
1. User-specific (not personal)
– Demographics (if available). Location (IP#)
2. Game events
– Session starts/stops. Achievements, purchases
3. Social-graph data
– Invites. Gifting. PvP actions. “Visiting”. Etc.
24. Tech: Algorithms used
• Neural network, with back propagation
• Support vector machines
• Random forests, with entropy reduction
• Graph-theoretic methods
– Including: social graph analysis
• Machine learning
30. What went right?
Overall increase in conversions… WHY?
• Similar players get similar predictive ratings
– “Marginal converters” are rated similarly to
inevitable converters.
• Promotions go to a smaller group
– Less promo-fatigue & irritation; fewer opt-outs
– Tightly-targeted emails get huge open rates & CTR