Portfolio — Marketing Data Science
Marketing
Incrementality
& Predictive LTV
Kraemer A. Lovelace · klovelace@gmail.com · glyphanalytics.com · Causal inference · LTV modeling · Budget optimization
True Incremental ROAS
2.4×
↑ 0.3× vs prior quarter
Non-Incremental Spend
$1.2M
budget at risk quarterly
Avg Incremental Lift
38%
↑ across paid channels
Channels Tested
7
2 geo-holdouts active
How to read this: Each bar shows the total attributed ROAS a channel claims. The orange segment is what causal measurement confirms as truly incremental lift. The gray segment on top is non-incremental spend — budget consumed by users who would have converted organically anyway. Retargeting and Display have high attributed ROAS but tiny true lift — the gray dominates.
Attributed vs. True Incremental ROAS
Non-incremental portion shown stacked above true causal lift. Hover a bar for breakdown.
Method
True incremental ROAS
Non-incremental (attributed, not causal)
ChannelSpendAttributed ROAS True ROASIncrementalityRating
Avg Predicted LTV (12 mo)
$284
↑ 14% vs cohort avg
High-Value Segment
22%
of new user acquisitions
LTV : CAC Ratio
3.8×
Target ≥ 3× ✓
Model AUC
0.83
Gradient boosting
LTV Distribution by Segment
Predicted LTV bins for newly acquired users by value tier.
Window
High-value (top 20%)
Mid-value
Low-value
Cohort Actual vs. Predicted
Model calibration by acquisition quarter. Close tracking validates reliability.
Actual LTV
Predicted LTV
Top LTV Predictors — Feature Importance
Early behavioral signals outweigh acquisition channel, meaning post-signup experience drives long-run value.
Optimized Budget
$4.8M
quarterly total
Projected ROAS Lift
+31%
vs. current allocation
Channels Increased
3
high incrementality
Channels Reduced
3
low true lift
The logic: Same total budget, smarter distribution. Channels with high measured incrementality receive increased investment. Channels where causal testing revealed low true lift — particularly retargeting and display, where attribution significantly overstates impact — are reduced. The projected +31% ROAS improvement comes from reallocation alone, not additional spend.
Current vs. Recommended Allocation
Derived from geo holdout incrementality findings. Hover for rationale.
Current spend
Recommended spend
ChannelCurrentRecommendedChangeEvidence
Paid social$1.4M$1.8M+$400K62% incremental lift; top LTV cohort
Connected TV$0.5M$0.9M+$400KGeo holdout: 45% lift on new signups
Influencer$0.3M$0.6M+$300KUnderinvested vs. measured causal impact
Email$0.2M$0.2MHigh efficiency; channel already efficient
Retargeting$1.2M$0.7M−$500K12% incrementality — 88% organic cannibalization
Display$0.9M$0.5M−$400KView-through inflated; geo test shows 8% true lift
Paid search (brand)$0.5M$0.1M−$400KBrand terms capture intent, don't create it