One-Line Definition
An attribution model is a ruleset that assigns credit for a conversion across the multiple marketing channels and touchpoints a customer interacted with before purchasing — answering the question every DTC operator eventually asks: *"Which of my ads actually made this sale happen?"*
Real-Life Analogy
Imagine a group of five friends who each contribute to planning a birthday dinner. One suggests the restaurant, another drives everyone there, a third brings the cake, a fourth pays for the appetizers, and a fifth gives the toast that makes the night memorable. When the guest of honor says "thank you for the amazing evening," who deserves the credit?
- The person who suggested the restaurant? (first touch)
- The one who paid the bill? (last touch)
- Everyone equally? (linear)
- The one whose contribution mattered most? (data-driven)
Attribution models are simply different ways of answering that "who gets the credit?" question — except the dinner is a $120 order, the friends are your Meta ads, TikTok videos, email flows, and Google searches, and the guest of honor is your customer.
Core Formula
At its simplest, every attribution model distributes a fixed pool of credit (100%) across N touchpoints in a conversion path:
Credit_i = w_i × Conversion Value
Where:
- Credit_i = credit assigned to touchpoint *i*
- w_i = weight of touchpoint *i*, subject to Σ w_i = 1
- Conversion Value = the total value generated (revenue, signups, etc.)
The only thing that changes between models is how w_i is calculated:
| Model | Weight Rule (w_i) | Example: 4-touch path (Ad → Email → Search → Retarget) |
|---|---|---|
| Last Touch | 100% to final touch | Ad: 0% · Email: 0% · Search: 0% · Retarget: 100% |
| First Touch | 100% to first touch | Ad: 100% · Email: 0% · Search: 0% · Retarget: 0% |
| Linear | Equal split across all | 25% each |
| Time Decay | Exponential decay, half-life ~7 days | e.g., 5% / 15% / 30% / 50% |
| Position-Based (U-shaped) | 40% first, 40% last, 20% split middle | 40% / 10% / 10% / 40% |
| Data-Driven | Assigned by ML model (Shapley value, Markov chains) | Varies by observed incremental lift |
A concrete example: a customer sees a $0.80 CPC TikTok ad, clicks a $1.20 CPC Google search ad, opens a free email, then converts on a $2.50 CPC retargeting ad for a $120 order. Under last-touch, retargeting gets 100% of the credit and looks like a hero. Under linear, each channel gets 25% — and suddenly TikTok's role becomes visible.
Comparison with Related Terms
| Term | What It Is | How It Differs from Attribution Model |
|---|---|---|
| **Attribution Model** | A ruleset for distributing conversion credit | The framework itself |
| **Attribution Window** | The time period (e.g., 7-day click, 1-day view) in which touchpoints count | Defines *which* touchpoints are eligible; the model defines *how* they're weighted |
| **Marketing Mix Modeling (MMM)** | Statistical regression on aggregate spend vs. revenue | Works without user-level tracking; complements attribution in a post-cookie world |
| **Incrementality Testing** | Geo-holdout or A/B tests measuring true causal lift | The gold standard; attribution is a proxy, incrementality is ground truth |
| **Multi-Touch Attribution (MTA)** | Umbrella term for any model using >1 touchpoint | MTA is a category; specific models (linear, time decay) live inside it |
| **Last-Click Attribution** | A single-touch model giving 100% to the final click | A subset of attribution models — the simplest and most misleading |
Use Cases
1. Budget reallocation across paid channels. A DTC brand spending $50,000/month across Meta, Google, and TikTok uses a data-driven model and discovers that TikTok assists 38% of conversions but receives only 12% of last-click credit. Shifting 15% of budget to TikTok lifts blended ROAS from 2.1x to 2.6x.
2. Email and SMS program justification. Email often sits mid-funnel and gets crushed by last-click. Switching to position-based attribution reveals that welcome flows touch 22% of first-time buyers — justifying a $4,000/month Klaviyo spend that last-click said was "unprofitable."
3. Creative and content decisions. Time-decay models (with a 7-day half-life) show that upper-funnel video ads influence purchases 10–14 days later, prompting brands to stop killing "underperforming" awareness campaigns after 72 hours.
4. Cross-border market entry. When entering a new region (e.g., DE or JP), first-touch attribution highlights which discovery channel — influencer, PR, or paid social — actually introduces the brand to a cold market, where retargeting pools don't yet exist.
5. Post-iOS 14.5 measurement strategy. With signal loss from ATT, brands triangulate: MTA for on-platform decisions, MMM for channel-level budget, and incrementality tests for validation. No single model is trusted alone.
Misconceptions
"Attribution tells me what's incremental." No — attribution is *correlational*. If a user would have bought anyway, retargeting still gets credit. Only incrementality testing (holdouts, geo splits) measures true causal lift. A retargeting campaign showing 8x ROAS might have a 0.5x *incremental* ROAS.
"Last-click is objective; other models are biased." Last-click is the *most* biased model — it systematically overvalues bottom-funnel, brand-search, and retargeting, and undervalues discovery. Every model encodes assumptions; last-click just hides them.
"One model is correct." The right model depends on your question. Use first-touch for market entry, position-based for full-funnel budget splits, time-decay for short consideration cycles (impulse DTC), and data-driven for mature programs with sufficient conversion volume (typically 300+ conversions/month per channel for statistical reliability).
"Attribution works the same across borders." It doesn't. GDPR consent rates in the EU (often 60–70%) and different platform APIs mean European paths are more fragmented. Brands often run separate models per region.
"More touchpoints = better model." Overcounting (e.g., counting an impression and a click from the same ad as two touchpoints) inflates credit and distorts weights. Deduplication matters.
Related Terms
- Multi-Touch Attribution (MTA)
- Last-Click / First-Click Attribution
- Linear, Time-Decay, Position-Based Models
- Data-Driven Attribution (DDA)
- Shapley Value & Markov Chain Models
- Marketing Mix Modeling (MMM)
- Incrementality Testing / Geo-Holdouts
- Attribution Window (7-day click, 1-day view)
- Blended CAC & MER (Marketing Efficiency Ratio)
- Conversion Path & Touchpoint
- Server-Side Tracking & Post-Cookie Measurement