An attribution model is the specific rule or algorithm that decides how conversion credit is split across every touchpoint a user took before converting. It’s the engine sitting under every channel-level revenue number in attribution reports. Pick a different model and the same conversion path produces a different credit allocation — sometimes 40%+ different. This guide focuses on the models themselves: the six rule-based options, GA4’s data-driven default, the 2024–2026 changes that killed cross-channel rule-based models in GA4, the math behind data-driven attribution, and a practical decision framework for picking one.
What Is an Attribution Model?
An attribution model is the formula that answers a single question: given a sequence of touchpoints leading to a conversion, what percentage of credit does each touchpoint deserve? The model is just the rule for splitting credit — separate from the data collection layer (cookies, server logs, the Measurement Protocol) and separate from the attribution settings UI in your analytics tool.
Two broad families exist. Rule-based models apply a fixed mathematical rule regardless of what the data actually shows — first-click gives 100% to the first touch, linear splits evenly, and so on. Data-driven models learn the rule from your historical paths and assign credit based on each touchpoint’s measured contribution to conversion probability. Rule-based models are deterministic and auditable; data-driven models are adaptive but opaque.
The model is purely an analytical layer. It doesn’t change which touchpoints were collected, doesn’t change conversion counts, and doesn’t change raw clicks or sessions. It only changes the percentage attached to each channel in the resulting revenue split — but that single number drives every downstream budget decision.
Rule-Based Attribution Models: First-Click, Last-Click, Linear, Time-Decay, Position-Based
Five rule-based models dominate practical use. Each makes a different assumption about which touchpoints “deserve” the credit. The table below compares them on a four-touchpoint path: Display ad → Organic search → Email click → Direct visit, ending in a $100 conversion.
| Model | Credit rule | Display ad | Organic | Direct | |
|---|---|---|---|---|---|
| First-Click | 100% to first touchpoint | $100 | $0 | $0 | $0 |
| Last-Click | 100% to last touchpoint | $0 | $0 | $0 | $100 |
| Linear | Equal split across all touches | $25 | $25 | $25 | $25 |
| Time-Decay | Exponential weight toward recent (7-day half-life) | $8 | $17 | $33 | $42 |
| Position-Based | 40% first + 40% last + 20% middle | $40 | $10 | $10 | $40 |
| Data-Driven (DDA) | Algorithmic — learned from path data | $22 | $31 | $28 | $19 |
The five rule-based models are interchangeable as opinions — there’s no objective answer to which is “correct” because each enforces a different theory of marketing influence. First-click champions discovery; last-click champions closing; linear refuses to take a side; time-decay assumes recency equals intent; position-based negotiates between the two extremes. Data-driven, in contrast, derives the answer from observed lift rather than assumption.
Data-Driven Attribution (DDA) in GA4: Now the Default
Data-driven attribution is GA4’s default model for every property created since late 2023. Unlike rule-based models, DDA doesn’t apply a fixed formula. It compares paths that converted with paths that didn’t, then estimates how much each touchpoint actually moved the conversion probability. The result is a per-touchpoint credit weight that’s specific to your account, your conversion type, and your current data window.
DDA is supported across every channel GA4 recognizes: UTM-tagged campaigns, organic search, paid search, email, social, referral, display, and direct. Google’s official DDA documentation states there’s no minimum conversion threshold to enable it, but the underlying ML model needs roughly 300 conversions per conversion type per 30 days to produce stable weights. Below that threshold GA4 falls back to weighted cross-channel logic that approximates position-based.
GA4 Attribution Model Changes 2024–2026 (Last-Click Sunset, DDA Default, Paid+Organic Only)
Three major attribution-model changes have reshaped GA4 between 2023 and 2026. Anyone running historical comparisons should know all three:
- Late 2023 — DDA becomes the new default. All new GA4 properties created from October 2023 ship with data-driven attribution as the reporting default, replacing last-click that had ruled Google Analytics for over a decade.
- April 2024 — cross-channel rule-based models removed. Google retired first-click, linear, time-decay, and position-based as cross-channel reporting models in GA4. They remain available only for paid Google Ads channels, not for cross-channel comparisons. Last-click and DDA are the only two cross-channel options.
- 2025–2026 — paid + organic last-click as the alternative. The “last-click” option in GA4 today is officially “paid and organic last-click”, which strips out direct visits when crediting and is closer to a non-direct last-touch model than the historical pure last-click logic.
The practical effect: in modern GA4, the realistic choice is between data-driven attribution (default) and paid-and-organic last-click. The other rule-based models live on inside Google Ads’ standalone attribution reports but no longer appear in GA4’s cross-channel views. Auditing a model switch in 2024+ data requires checking GA4’s release notes for the exact reporting cut-over.
How DDA Works: Shapley Value + Markov Chain Logic
The internals of DDA aren’t fully public, but Google has confirmed the model uses concepts from two well-known algorithms: Shapley value from cooperative game theory and Markov chains from probability theory. Together they form the backbone of every modern data-driven attribution engine, including ChannelMix, Adobe DDA, and the open-source marketing analytics community’s reference implementations.
Shapley value answers: “If channel X were absent, how much would total conversions drop?” It’s calculated by enumerating every possible coalition of channels and measuring channel X’s average marginal contribution. The credit each channel receives equals its average lift across all subsets — a mathematically fair split that ignores ordering.
Markov chain attribution models the customer journey as a state machine: each channel is a state, conversion is the absorbing state, and transition probabilities come from observed paths. Credit equals each channel’s “removal effect” — the drop in total conversion probability if that channel state were deleted from the graph. Markov captures path order; Shapley captures presence/absence. DDA blends both signals.
The output is non-obvious. A channel with high path frequency but low marginal lift (direct visits from existing customers) gets less credit than its raw count suggests. A channel that rarely shows up but consistently lifts conversion when it does (a niche referral, a high-quality podcast mention) gets disproportionately more. That’s why DDA reports look unfamiliar after years of last-click — they reflect causation rather than recency.
When to Use Each Model: Brand vs Performance vs Awareness
Different campaign goals favour different attribution models. Map your current optimization target to the model that exposes the right signal:
- Performance / direct response — short single-session conversions. Last-click is acceptable as a sanity check; DDA still preferred because even impulse buyers see priming touchpoints.
- Brand awareness — display, sponsorships, top-of-funnel content. Use first-click as a complementary view to spotlight which channels open journeys, but never as the sole optimization model.
- Demand generation / mid-funnel — content, retargeting, nurture email. Position-based (40/20/40) recognizes both opening and closing channels and is interpretable for stakeholders.
- Long consideration cycle (B2B, finance, real estate) — DDA with a 90-day lookback window. Rule-based models systematically miss upper-funnel touches in long cycles because of how they weight or ignore early touchpoints.
- Mature, high-volume e-commerce — DDA. Conversion volume is the bottleneck for DDA stability, and high-volume accounts are where it shines.
- Reporting baseline / sanity check — linear. Never an optimization model, but useful for stakeholders who want to see a non-opinionated split alongside DDA.
Cross-Channel Attribution Limitations (Cookie Loss, Walled Gardens)
No attribution model — rule-based or data-driven — fixes the underlying data gaps. The model only describes the touchpoints it can see. Five structural limitations apply to every model:
- Cookie loss. Safari’s ITP caps third-party cookie life at 7 days and progressively limits first-party cookies. A user researching on iPhone over three weeks loses touchpoint stitching well before any 30-day window expires.
- Walled gardens. Meta, TikTok, and Google’s own platforms keep raw user-level path data inside their reporting environments. GA4 sees clicks but not impressions, so view-through influence from social and display is structurally invisible.
- Multi-device journeys. A user researching on mobile and converting on desktop appears as two separate users without cross-device tracking via Google Signals or User-ID. Attribution then double-counts both fragments as separate paths.
- Dark traffic. Direct visits from messaging apps, podcast mentions, and offline word-of-mouth land in GA4 as (direct) / (none). The upstream channel that caused them is invisible to every attribution model.
- Lookback cut-offs. Touchpoints outside the configured window simply don’t exist for the model. A 30-day window in a B2B account routinely misses the display ad and content visit that opened the journey six weeks earlier.
Server-Side and Conversions API (CAPI) for Better Attribution
The most reliable fix for cookie loss isn’t a new attribution model — it’s better data collection. Server-side tagging, the Measurement Protocol, and platform-specific conversion APIs (Meta CAPI, TikTok Events API, LinkedIn Conversions API) ship conversion events directly from your server to the ad platforms, bypassing the browser entirely. The result: a deduplicated conversion stream that survives ad blockers, ITP, and consent withdrawal.
For DDA specifically, server-side data improves training quality because the model learns from a more complete sample of paths. Higher path coverage means the Shapley/Markov calculations stabilize faster and produce credit weights with less variance. Avinash Kaushik’s 2024 essays on attribution and incrementality argue this is now the largest single improvement most accounts can make to attribution accuracy — bigger than any model choice.
Server-side adds engineering cost (gtag-server, BigQuery export, event deduplication logic), so it’s a Tier-1 priority for accounts spending $50k+/month on paid media and a Tier-2 priority below that. For implementation walkthroughs see our BigQuery guide on server-side event modelling.
Multi-Touch vs Single-Touch Models
The most useful taxonomy is single-touch vs multi-touch. Single-touch models give 100% to one touchpoint (first-click, last-click). Multi-touch models distribute credit across all touchpoints (linear, time-decay, position-based, DDA). The split matters because it changes which channels appear profitable.
| Family | Models | Best for | Worst for |
|---|---|---|---|
| Single-touch | First-click, Last-click | Short single-session journeys; reporting baselines | B2B, long-cycle, multi-channel campaigns |
| Rule-based multi-touch | Linear, Time-decay, Position-based | Mid-length B2C journeys; explainability to non-analysts | Accounts with stable conversion volume where DDA works |
| Algorithmic multi-touch | Data-driven (Shapley/Markov hybrid) | High-volume conversion accounts; cross-channel optimization | Low-conversion accounts (<300/month); transparency-critical reporting |
The portfolio standard since 2024: a multi-touch model is mandatory for cross-channel decisions, and DDA is the default unless conversion volume is too low. Rule-based multi-touch (linear, time-decay, position-based) survives mainly inside Google Ads’ standalone attribution reports rather than in GA4’s cross-channel views.
Choosing an Attribution Model: Decision Framework
A practical four-question decision tree handles 90% of accounts:
- Do you have ≥300 conversions per type per 30 days? If yes, DDA. If no, paid-and-organic last-click as a temporary baseline while the conversion volume grows.
- Is your sales cycle longer than 30 days? If yes, set the GA4 lookback window to 90 days regardless of model. The model is useless if it can’t see the upper-funnel touch.
- Are you running brand-awareness campaigns? If yes, add first-click as a complementary report (not the primary model) so brand investments are visible.
- Can you stomach a black-box model? If executives demand explainability, run linear or position-based alongside DDA in GA4’s Model comparison view. Show both, anchor decisions on DDA.
The framework deliberately doesn’t pick a single “best” model. Modern attribution practice runs multiple models in parallel — DDA as the optimization target, last-click as the audit baseline, first-click as the discovery view — and compares them in GA4’s Model comparison report. Treat any single model as directional, never as single-source-of-truth revenue accounting.
Frequently Asked Questions
What is an attribution model?
An attribution model is the rule or algorithm that decides how conversion credit is divided across every touchpoint a user touched before converting. It’s separate from the data collection layer — it’s purely the formula that turns a sequence of touchpoints into per-channel revenue numbers in your reports.
What’s the difference between rule-based and data-driven attribution?
Rule-based models apply a fixed formula — first-click gives 100% to the first touch, linear splits evenly, etc. — regardless of what the data shows. Data-driven attribution uses machine learning (Shapley value plus Markov chain logic) to derive credit weights from your actual conversion paths, so the rule itself is learned from your data instead of imposed on it.
Which attribution model is best for B2B?
For high-volume B2B accounts, data-driven attribution with a 90-day lookback window. For low-volume B2B (under 300 conversions/month), paid-and-organic last-click with a 90-day window is a reasonable fallback. The lookback window matters more than the model choice in B2B, because most rule-based models simply ignore touchpoints outside the window.
Why did GA4 remove first-click, linear, time-decay, and position-based models?
In April 2024 Google removed first-click, linear, time-decay, and position-based as cross-channel reporting models in GA4. These models still exist inside Google Ads’ standalone attribution reports for paid channels, but cross-channel comparisons in GA4 are now limited to data-driven attribution and paid-and-organic last-click. The reasoning: rule-based cross-channel reporting was systematically misleading and DDA is now mature enough to be the default.
How many conversions does data-driven attribution need?
Google’s documentation suggests roughly 300 conversions per conversion type per 30 days for stable DDA weights. Below that threshold GA4 still labels reports as “data-driven” but the underlying model falls back to weighted cross-channel logic similar to position-based until enough conversion volume accumulates.
Does data-driven attribution use Shapley values?
Yes — Google has confirmed DDA blends Shapley value (from cooperative game theory, measuring each channel’s average marginal contribution across all coalitions) with Markov chain logic (modelling the journey as a state machine and measuring each channel’s removal effect). Most modern DDA engines, including Adobe Analytics DDA and ChannelMix, follow the same hybrid approach.
Can I run multiple attribution models at the same time?
Yes, and you should. GA4’s Advertising → Model comparison report lets you compare data-driven, last-click, and (for paid Google Ads channels) any retired rule-based model side by side on the same conversions. Modern practice is DDA as the primary optimization model, last-click as the audit baseline, and first-click as a discovery-channel view.
Related Terms
- Attribution — the broader practice of crediting conversions across touchpoints (this page focuses specifically on the models).
- Conversion — the key event being attributed by the model.
- UTM — URL parameters that label inbound touchpoints so attribution models can identify them.
- Cookie — the storage mechanism most attribution models still depend on for path stitching.
- First-party cookie — cookie set by your domain, increasingly the only kind surviving ITP/ETP.
- Measurement Protocol — server-side event ingestion that bypasses the browser entirely.
- Cross-device tracking — the User-ID/Signals layer that prevents one user looking like two paths.
- Cohort — grouping users by acquisition source for attribution-validated retention analysis.
- Macro conversion — the high-value event most worth attributing accurately.
- Event — the GA4 primitive that conversions and touchpoints are built from.
- BigQuery — where to take attribution data when GA4’s UI runs out.
Bottom Line
An attribution model is just a rule for splitting conversion credit across touchpoints — but the choice of rule reshapes channel-level revenue by 15–40% and drives every downstream budget decision. The five rule-based models (first-click, last-click, linear, time-decay, position-based) all impose a fixed assumption; data-driven attribution learns the rule from your actual paths using Shapley value plus Markov chain logic. In modern GA4 the realistic cross-channel choice is DDA (default, requires ≥300 conversions/month) or paid-and-organic last-click (audit baseline). Run both in parallel via Model comparison, anchor decisions on DDA, and remember no model fixes the underlying cookie loss, walled-garden, and cross-device data gaps — server-side tagging via Conversions APIs does.
