# S-Shape vs Concave

## S-Shape (ICP) vs Concave (ADBUG)

### Choosing the Right Curve Type for Your Marketing Channels

MixModeler offers two fundamental curve types for modeling saturation: **S-Shape (ICP)** for threshold-based effects and **Concave (ADBUG)** for immediate diminishing returns. Understanding when to use each is critical for accurate marketing attribution.

***

### S-Shape Curves (ICP)

#### Visual Pattern

```
Effect
  |           ╱────────  (Saturation)
  |         ╱
  |       ╱              (Rapid growth)
  |     ╱
  |   ╱                  (Slow start)
  | ╱
  |╱____________________
       Spend →
```

**Three Distinct Phases:**

1. **Slow Start:** Minimal effect at low spend (below threshold)
2. **Rapid Acceleration:** Steep effectiveness increase (threshold crossed)
3. **Saturation:** Diminishing returns, approaching maximum

***

#### When Marketing Shows S-Shape Behavior

**Threshold Effects:** Marketing needs critical mass before becoming effective

**Network Effects:** Value increases as more people are reached (social phenomena)

**Viral Potential:** Slow build, then rapid spread, then saturation

**Brand Awareness:** Need sufficient reach before brand recall kicks in

***

#### Real-World S-Shape Examples

**Social Media Campaigns:**

* Posts need engagement threshold to trigger algorithm amplification
* Slow initial shares → viral tipping point → saturation
* **Use ICP curves**

**New Product Launches:**

* Initial adopters slow (awareness building)
* Word-of-mouth acceleration phase
* Market saturation eventually
* **Use ICP curves**

**Brand TV Campaigns:**

* Need frequency/reach threshold before brand lift
* Middle campaign period shows peak effectiveness
* Late campaign shows saturation
* **Use ICP curves**

**PR and Sponsorships:**

* Slow credibility building phase
* Recognition acceleration once established
* Saturation at high awareness levels
* **Use ICP curves**

***

#### S-Shape Parameters (ICP)

**Available in MixModeler:**

**CDR Formula (3 parameters):**

* Alpha: Controls overall saturation level
* Beta: Controls steepness of middle acceleration
* Gamma: Controls switch point location (threshold)

**ATAN Formula (2 parameters):**

* Alpha: Controls saturation level
* Power: Controls curve shape (power > 1.5 creates S-shape)

***

### Concave Curves (ADBUG)

#### Visual Pattern

```
Effect
  |╱──────────────────  (Saturation)
  |
  |
  |                      (Immediate diminishing returns)
  |
  |
  |____________________
       Spend →
```

**Single Phase:**

* High initial impact
* Immediate and continuous diminishing returns
* Gradual approach to maximum (no inflection point)

***

#### When Marketing Shows Concave Behavior

**Audience Sorting:** Best prospects captured first, each additional reach less valuable

**Frequency Fatigue:** First impression most impactful, additional exposures less effective

**Search Intent:** High-intent keywords exhausted first, expanding reach lowers conversion

**Price Sensitivity:** Most price-sensitive customers respond first to promotions

***

#### Real-World Concave Examples

**Paid Search:**

* Best keywords (high intent) captured immediately
* Expanding to broader keywords lowers conversion rate
* Each additional keyword less efficient
* **Use ADBUG curves**

**Performance Display:**

* High-intent audiences targeted first (retargeting, lookalikes)
* Audience expansion reaches less-qualified users
* Frequency caps cause diminishing impact
* **Use ADBUG curves**

**Email Marketing:**

* Best customers open/convert immediately
* Additional sends show declining response rates
* List fatigue sets in
* **Use ADBUG curves**

**Price Promotions:**

* Most price-sensitive shoppers respond first
* Deep discounts capture progressively less-sensitive segments
* Immediate saturation at high discount levels
* **Use ADBUG curves**

**Direct Mail:**

* Best prospects (highest propensity) mailed first
* Additional targeting reaches lower-propensity segments
* Immediate diminishing returns per piece
* **Use ADBUG curves**

***

#### Concave Parameters (ADBUG)

**Available in MixModeler:**

**CDR Formula (3 parameters):**

* Alpha: Controls saturation ceiling (typically 0.8-1.0)
* Beta: Controls steepness of diminishing returns
* Gamma: Controls how quickly saturation is reached

**ATAN Formula (2 parameters):**

* Alpha: Controls saturation level
* Power: Set to 1.0 for pure concave (no S-shape)

***

### Direct Comparison

#### Key Differences

```
```

***

#### Visual Side-by-Side

**S-Shape (ICP):**

```
      ╱────
    ╱
  ╱
╱___________
  ↑
  Switch point (threshold)
```

**Concave (ADBUG):**

```
╱───────────
  ↓
  Immediate diminishing returns
```

***

### Decision Framework

#### Choose S-Shape (ICP) When:

✅ **Threshold Exists** Need minimum investment before effectiveness begins

✅ **Brand Building** Awareness campaigns, sponsorships, PR

✅ **Network/Viral Effects** Social media, word-of-mouth dependent

✅ **Long Sales Cycles** B2B campaigns with awareness → consideration → conversion funnel

✅ **New Market Entry** Building presence from scratch

**Typical Channels:**

* Social media (Facebook, Instagram, TikTok)
* Brand TV campaigns
* Sponsorships and events
* Influencer marketing
* PR campaigns

***

#### Choose Concave (ADBUG) When:

✅ **Immediate Response** First exposure drives immediate action

✅ **Performance Marketing** Direct response, conversion-focused

✅ **Audience Targeting** Best prospects captured first

✅ **Short Sales Cycles** Transactional, low consideration purchases

✅ **Frequency-Based** Multiple exposures show declining effectiveness

**Typical Channels:**

* Paid search (Google Ads, Bing)
* Performance display/programmatic
* Email marketing
* Price promotions
* Retargeting campaigns
* Direct mail

***

#### Test Both If:

⚠️ **Unclear Pattern** Can't determine from business logic alone

⚠️ **Mixed Behavior** Channel shows both threshold and immediate response characteristics

⚠️ **New Channel** No historical precedent to guide decision

⚠️ **Experimental** Testing new creative or targeting approach

**Solution:** Use Curve Testing interface to test both, choose based on statistical fit (t-statistics, R²)

***

### Testing in MixModeler

#### Curve Testing Workflow

**Step 1: Select Variable** Choose marketing channel to test

**Step 2: Test S-Shape (ICP)**

* Run curve tests with ICP type
* Test multiple parameter combinations
* Note best t-statistic and R²

**Step 3: Test Concave (ADBUG)**

* Run curve tests with ADBUG type
* Test multiple parameter combinations
* Note best t-statistic and R²

**Step 4: Compare Results**

* Which curve type has higher t-statistic?
* Which shows better R² improvement?
* Which makes more business sense?

**Step 5: Select Winner**

* Use curve type with best statistical fit AND business logic alignment

***

### Common Patterns by Channel

#### TV Advertising

**Brand TV:** S-Shape (ICP)

* Need frequency for brand lift
* Switch point around 500-1000 GRPs

**Direct Response TV:** Concave (ADBUG)

* Immediate phone/web response
* Diminishing returns with frequency

***

#### Digital Display

**Programmatic:** Concave (ADBUG)

* Audience exhaustion
* Frequency caps
* Immediate diminishing returns

**High-Impact (Takeovers):** S-Shape (ICP)

* Need threshold impressions for impact
* Awareness building

***

#### Social Media

**Organic/Viral:** S-Shape (ICP)

* Engagement threshold for algorithm boost
* Network effects

**Paid Social (Performance):** Concave (ADBUG)

* Best audiences first
* Auction saturation

***

#### Radio

**Brand Radio:** S-Shape (ICP)

* Need frequency for recall
* Threshold effects

**Direct Response Radio:** Concave (ADBUG)

* Immediate call-to-action response
* Frequency fatigue

***

### Parameter Implications

#### S-Shape Parameters

**Alpha (higher values = more saturation):**

* 2.0-3.0: Moderate saturation
* 3.0-4.0: Strong saturation

**Power (ATAN formula):**

* 1.5-1.8: Gentle S-shape
* 1.8-2.0: Moderate S-shape
* 2.0+: Strong S-shape (steep middle)

**Switch Point:** Location where acceleration begins (automatically calculated)

* Should align with business threshold expectations

***

#### Concave Parameters

**Alpha (saturation ceiling):**

* 0.8-0.9: Lower ceiling (strong diminishing returns)
* 0.9-1.0: Higher ceiling (moderate diminishing returns)

**Power (ATAN formula):**

* Always 1.0 for pure concave shape

**Steepness:** How quickly diminishing returns kick in

* Controlled by beta (CDR) or implicit in ATAN

***

### Interpretation Differences

#### S-Shape Model

```
Sales = β₀ + β₁ × ICP(TV_Spend)
```

**If β₁ = 0.5:**

* Below threshold: TV has minimal impact
* At switch point: TV shows rapid effectiveness increase
* Above saturation: TV impact plateaus

**Business Insight:** "We need $X minimum spend for TV to work, then it's highly effective until $Y where it saturates"

***

#### Concave Model

```
Sales = β₀ + β₁ × ADBUG(Search_Spend)
```

**If β₁ = 0.6:**

* First dollars: Very high ROI
* Additional dollars: Continuously declining ROI
* High spend: Approaches ceiling

**Business Insight:** "Search is immediately effective but quickly shows diminishing returns. Optimal spend is around $X before saturation."

***

### Common Mistakes

#### Mistake 1: Using S-Shape for Performance Channels

**Problem:** Applying ICP to paid search or performance display

**Why Wrong:** These channels don't have threshold effects - they work immediately

**Fix:** Use Concave (ADBUG) for performance marketing

***

#### Mistake 2: Using Concave for Brand Campaigns

**Problem:** Applying ADBUG to brand TV or sponsorships

**Why Wrong:** Miss the threshold effect and acceleration phase

**Fix:** Use S-Shape (ICP) for brand/awareness campaigns

***

#### Mistake 3: Not Testing Both

**Problem:** Assuming curve type without validation

**Why Wrong:** May miss better-fitting alternative

**Fix:** Always test both types in Curve Testing, let data guide

***

#### Mistake 4: Ignoring Business Logic

**Problem:** Choosing curve based solely on statistics

**Why Wrong:** May select mathematically better but illogical curve

**Fix:** Statistics + business judgment = best choice

***

### Summary

**Key Takeaways:**

📈 **S-Shape (ICP) = Threshold effects** - slow start, rapid middle, saturation

📉 **Concave (ADBUG) = Immediate diminishing returns** - high start, continuous decline

🎯 **Brand → S-Shape** - awareness, viral, network effects

💰 **Performance → Concave** - direct response, audience sorting

🧪 **Test both when uncertain** - let statistics + logic guide

📊 **Different parameters** - Alpha, Power/Beta/Gamma behave differently per type

✅ **Match to marketing behavior** - curve should reflect reality

Choose the curve type that matches your channel's actual behavior, not the one that just fits the data best. The goal is accurate representation of marketing dynamics, not just mathematical fit!