Customer Acquisition. Lookalike Modeling. Performing Arts

Cut Mailing Costs by 30%
Without Losing Performance

A regional performing arts organization replaced broad saturation mailings with a machine learning-driven targeting strategy, eliminating 30% of their mailing volume while matching or exceeding prior campaign results.

30%
Reduction in
Mailing Volume
and Cost
Matched
or Better
Campaign Results vs.
Prior Full-Saturation
Mailing Efforts
100%
Budget Spent on
High-Propensity
Households Only
The Situation

Broad Reach.
Wasted Budget.

A regional performing arts organization had long relied on saturation direct mail to fill seats. While the reach was wide, the efficiency was not. Rising costs and limited audience insight forced a rethink of their entire acquisition strategy.

🎭

The Client

A regional performing arts organization running ongoing direct mail campaigns to drive ticket sales. Their audience is local and geographically defined, making targeted direct mail a core acquisition channel.

Performing Arts Ticket Sales Regional Audience

The Challenge

Rising direct mail costs with limited visibility into which households were most likely to respond. Traditional saturation mailings were broad and expensive, and the organization feared that reducing volume would lead to a drop in ticket sales.

High Mailing Costs Saturation Approach No Audience Insight
🤖

The Solution

DS Graphics Universal Wilde partnered with the organization to apply machine learning to their customer data. Rather than mailing everyone, the team built a lookalike model to score every household by likelihood to convert, then removed low-propensity addresses from the list entirely.

Machine Learning Lookalike Modeling Propensity Scoring
📬

The Approach

Historical ticket-buyer data was used to train a model that identified behavioral and demographic patterns among existing customers. Each household in the target region received a likelihood-to-convert score. Only high-propensity households made the final mailing list.

Historical Data Modeling Household Scoring List Suppression
The Methodology

A Three-Step
Machine Learning Strategy

The approach replaced guesswork with a repeatable, data-driven system for identifying which households deserved a spot on the mailing list.

Step 1

Build the Lookalike Model

Using historical ticket-buyer data, a machine learning algorithm identified patterns and behaviors common among existing customers. The model learned what a high-value audience member actually looks like at a household level.

Steps 2 + 3

Score and Suppress

The model evaluated every household in the target region, assigning a likelihood-to-convert score based on how closely each matched known customer profiles. Households below the conversion threshold were removed from the list entirely.

  • 2
    Score every householdPropensity score assigned based on fit with the customer profile model.
  • 3
    Eliminate low-propensity addressesBudget focused only on households with real conversion potential.
Why It Worked

Data Over Assumptions

No guesswork. No blanket coverage. Every decision was backed by real customer data, ensuring that every dollar in the mailing budget was directed at a household that had demonstrated meaningful similarity to an existing ticket buyer.

The Outcome

Smarter Segments, Better Results

Resources previously wasted on low-propensity households were reinvested into stronger audience segments. The result was a leaner, more efficient campaign that proved volume and performance are not the same thing.

Results Visualized

Less Volume,
Same Impact

Mailing Volume: Before vs. After
Saturation Mailing (Before)100% of list
ML-Targeted Mailing (After)70% of list

30% of the mailing list was removed after being identified as low-propensity, cutting print and postage costs by 30% without touching the high-value segment.

Cost Efficiency: Budget Allocation
Budget on Low-Propensity HHs (Before)30%
Budget on Low-Propensity HHs (After)0%
Budget on High-Propensity HHs (After)100%

After suppression, every dollar went to households most likely to engage, increasing overall marketing efficiency with the same creative and offer.

Campaign Performance Comparison
Saturation Campaign ResultsBaseline
ML-Targeted Campaign ResultsEqual or Better

Despite mailing 30% fewer pieces, campaign results matched or exceeded prior saturation efforts, confirming that the removed households were genuinely low-value.

Where the Savings Go
Wasted on Low-Propensity (Before)30% of budget
Reinvested in High-Value SegmentsRedirected

Freed-up budget can be reinvested into additional high-propensity outreach, creative upgrades, or complementary digital channels to further amplify results.

Key Results

The Impact,
By the Numbers

✂️
30%
Reduction in Total
Mailing Volume and Cost
🎯
100%
Of Budget Directed at
High-Propensity Households
📈
Matched
or Better
Campaign Performance vs.
Prior Saturation Efforts
🤖
ML
Machine Learning Model
Trained on Historical Buyers
🏠
0
Low-Propensity Households
Remaining on Final List
💡
3x
Sharper Audience Focus
vs. Saturation Baseline
The Analysis

Why Machine Learning
Changes the Math

Volume Is Not Value

Saturation mailing assumes more reach equals more response. The shift this organization made was fundamental: by treating every household as a data point, they discovered that 30% of their list was simply not worth reaching. Not bad addresses — just households whose behavioral and demographic signals indicated low conversion potential.

How the Lookalike Model Works

The model was trained on historical ticket-buyer data, identifying signals common among households that had actually converted. Those patterns were applied to every household in the target region. Rather than guessing who might be interested, the model identified who looks like someone who already was.

The Performance Paradox

Mailing fewer people did not reduce results. Campaign performance matched or exceeded prior saturation efforts, validating the core hypothesis: the removed households were not contributing to outcomes. When you eliminate the noise, the signal gets stronger — and the remaining households are more likely to respond.

A Self-Improving System

Unlike a one-time list cleanse, this model improves with each campaign. New response data retrains the model, making future scoring more accurate. Historical data becomes a compounding competitive advantage rather than a static reference point.

Performance Snapshot

Mailing Volume Change-30%
Cost Change-30%
Campaign PerformanceMatched or Better
Low-Propensity HHs Removed30% of list
Budget Efficiency100% on target
Data FoundationHistorical buyers
Modeling ApproachML Lookalike

What Was Eliminated

GuessworkRemoved
Blanket CoverageRemoved
Low-Value SpendReinvested
Wasted ImpressionsEliminated
Conclusion

Smarter Segments.
Better Outcomes.

A 30% reduction in mailing volume with no loss in performance. Machine learning did not shrink this campaign. It sharpened it.

The households this organization stopped mailing were never going to convert. The ones they kept were exactly who they needed. Machine learning did not shrink this campaign — it sharpened it, delivering equal results at 30% lower cost with every dollar focused on buyers who matched the profile of someone who had already said yes.

For any organization relying on direct mail, this is the playbook: stop mailing everyone. Start mailing the right people.