Skip to main content

How NodeNarrative Attribution Works

No black boxes. No proprietary algorithms. Our attribution methodology is built on peer-reviewed mathematics and graph analytics, and we'll show you exactly how it works.

Explore the Models View Platform

Why Platform-Reported Metrics Can't Be Trusted

Every platform takes credit for the same conversion. Only one actually drove it.

Google Analytics 4

“Organic search drove 60% of revenue.”

GA4 defaults to last-click attribution. It credits the final touchpoint and ignores the paid ad, email, or social post that started the journey.

Meta Ads

“Our campaigns generated $50K in revenue.”

Meta counts any conversion within its attribution window, even if the customer would have purchased anyway. Self-reported, self-serving.

Google Ads

“Search ads delivered 8x ROAS.”

Google Ads takes credit for brand searches (people who were already going to buy). The ROAS looks impressive but doesn’t reflect incremental impact.

The result? Conflicting reports, wasted budget, and decisions based on whichever platform shouts loudest. There's a better way.

A Knowledge Graph-Based Approach to Attribution

Traditional tools analyse touchpoints in isolation. They see a list of clicks. NodeNarrative uses a graph database to map the relationships between touchpoints, revealing how channels work together to drive conversions.

From Spreadsheets to Knowledge Graphs

A spreadsheet shows you rows of clicks. A knowledge graph shows you connected journeys. Built on Neo4j, NodeNarrative represents every customer interaction as a node and every journey step as an edge, creating a queryable map of how your marketing actually works.

This graph structure is what makes advanced models like Shapley values and Markov Chain analysis possible at scale. Instead of aggregating data into summaries, we traverse individual journeys to find the patterns that matter.

Relationship Mapping

Every touchpoint is a node. Every journey is a path. Neo4j maps the connections between channels, revealing how they work together, not just in sequence.

Path Analysis

Traverse actual customer journeys, not aggregated summaries. See the exact sequence of interactions that lead to high-value conversions.

Cross-Channel Visibility

Graph queries surface patterns that relational databases can’t: which channel combinations produce the highest conversion rates, and where journeys stall.

Real-Time Processing

Sub-100ms query performance at the p95 level. Attribution data updates as journeys progress, with no overnight batch processing.

The Seven Attribution Models

Every paid plan includes all seven models. Run them simultaneously, compare where they agree, and make decisions with confidence.

Heuristic Models

Rule-based

First-Click

100% credit to the first touchpoint.

When a customer converts, the model looks back through their entire journey and gives full credit to the very first interaction, the touchpoint that started the relationship.

  • Easy to understand and explain
  • Good for measuring awareness and discovery channels
  • Useful as a baseline comparison
  • Ignores all mid-funnel and closing activity
  • Overvalues top-of-funnel channels
  • Can misattribute if the first touch was incidental

Best for: Understanding which channels drive initial discovery.

Last-Click

100% credit to the final touchpoint.

The model attributes the entire conversion to whatever touchpoint immediately preceded the purchase: the "closer" that sealed the deal.

  • Simple and widely understood
  • Effective for measuring bottom-of-funnel channels
  • Default in most analytics platforms (familiar to teams)
  • Ignores the full journey that led to the conversion
  • Overvalues retargeting and brand search
  • Creates a misleading picture of channel performance

Best for: Identifying which channels close sales.

Linear

Equal credit to every touchpoint.

If a customer had 5 touchpoints before converting, each touchpoint receives 20% of the credit. Every interaction is treated as equally important.

  • Acknowledges every channel in the journey
  • No single touchpoint is over- or under-valued
  • Good starting point for teams new to multi-touch attribution
  • Treats a casual blog visit the same as a high-intent product page
  • Doesn’t account for diminishing or increasing impact over time
  • Can dilute the signal from truly influential channels

Best for: Teams wanting a balanced, unbiased view of all channels.

U-Shaped

40% to first touch, 40% to last touch, 20% to the middle.

When a customer converts, the model gives 40% of credit to the first interaction that started the journey and 40% to the final interaction that closed the sale. The remaining 20% is distributed equally among all touchpoints in between.

  • Acknowledges both discovery and conversion channels
  • Gives proportional credit to mid-funnel nurturing
  • Good for businesses with strong brand and conversion focus
  • Fixed weighting may not reflect actual influence
  • Undervalues mid-funnel if the journey has many touchpoints
  • Assumes endpoints are always the most important interactions

Best for: Businesses with strong brand awareness and clear conversion channels.

Time-Decay

More credit to touchpoints closer to conversion.

The model applies an exponential decay function that weights recent interactions more heavily than earlier ones. A touchpoint one day before conversion gets significantly more credit than one from two weeks ago.

  • Reflects the recency effect in purchase decisions
  • Works well for short sales cycles and time-sensitive campaigns
  • Strong for evaluating promotional and urgency-driven marketing
  • Undervalues awareness channels that started the journey
  • Assumes recent touchpoints are always more important
  • Less useful for long consideration cycles

Best for: Short sales cycles, promotional campaigns, and urgency-driven marketing.

Advanced Models

Data-driven

Shapley Values

Fair credit based on each channel’s marginal contribution.

Uses cooperative game theory to calculate the fair contribution of each channel by analysing all possible combinations of touchpoints.

Imagine removing one channel from every possible journey combination. The average change in conversion rate across all those combinations is that channel’s Shapley value, its true marginal contribution to the outcome.

  • Mathematically proven to be the fairest distribution method
  • Accounts for synergies between channels
  • Based on peer-reviewed economics (Lloyd Shapley, 1953; Nobel Prize 2012)
  • Produces consistent, repeatable results
  • Computationally intensive with many channels (addressed by our graph engine)
  • Requires sufficient data volume for statistical significance

Best for: Budget allocation decisions where you need to know exactly what each channel contributes.

Markov Chain

Credit based on probabilistic removal effect.

Models the customer journey as a probabilistic state machine, then measures each channel’s impact by calculating what happens when it’s removed.

The model maps all observed journeys as transition probabilities between channels. It then removes each channel one at a time and measures the drop in overall conversion rate, the "removal effect". A larger drop means a more critical channel.

  • Captures the sequential nature of customer journeys
  • Identifies channels that are critical path dependencies
  • Works well with complex, multi-step funnels
  • Reveals hidden bottlenecks in the conversion path
  • Requires meaningful journey volume for reliable probabilities
  • Results can shift with seasonal traffic pattern changes

Best for: Understanding which channels are essential to the conversion path: not just present, but necessary.

How to Read Your Attribution Graph

Understanding your graph is straightforward. Here's what each element represents.

Step 1

Nodes Are Touchpoints

Each circle in the graph represents a customer interaction: a page view, ad click, email open, or purchase. Colour indicates the channel.

Step 2

Edges Are Journeys

Lines connecting nodes show the sequence of interactions. Follow an edge to see how a customer moved from one touchpoint to the next.

Step 3

Node Size Shows Influence

Larger nodes have a higher attribution score, meaning more conversions flow through them. Quickly spot the most impactful touchpoints at a glance.

Step 4

Edge Thickness Shows Frequency

Thicker edges represent paths taken by more customers. These are your most common conversion journeys, the patterns worth optimising.

Data You Can Trust

Every number in NodeNarrative is backed by first-party data, not sampled, not modelled, not estimated from third parties.

95%+

Attribution Accuracy

Validated against known conversion paths using first-party data collected directly from your store and ad platforms.

7

Attribution Models

Run simultaneously and compare where they agree. High agreement means high confidence in those insights.

<100ms

API Response (p95)

Real-time attribution powered by optimised graph queries and edge caching. No waiting for overnight batch jobs.

0%

Sampled Data

First-party data only: never modelled, estimated, or sampled from third parties. Every data point is real.

How We Measure Accuracy

Our 95%+ accuracy claim isn't marketing, it's a measured outcome. We validate attribution results against known conversion paths: journeys where the full sequence of interactions is captured from first touch to purchase.

Accuracy is calculated by comparing model-predicted attribution to observed outcomes across controlled datasets. Where models disagree, we highlight the divergence so you can investigate, because knowing where certainty ends is just as valuable as the certainty itself.

This is possible because we use first-party data collected directly from your store and ad platforms. No sampling, no estimation, no third-party data modelling. Every data point is real.

Privacy-Preserving Attribution

Most attribution tools treat privacy as a constraint that reduces data quality. NodeNarrative's graph architecture makes privacy and accuracy complementary, not competing.

Consent-aware attribution models

Every identifier in the knowledge graph carries its own consent status. When a customer grants email consent but declines cookie tracking, the attribution model uses the email touchpoints and excludes the cookie-based ones. The graph enforces this at the identifier level, so models never accidentally use unconsented data paths. This is more accurate than the all-or-nothing approach competitors use, where declining any single consent removes the entire customer from attribution.

Erasure without breaking the model

When a customer exercises their GDPR right to be forgotten, NodeNarrative cascades the deletion through every connected touchpoint in the graph. The attribution models automatically recalculate without the erased data. Because the graph maintains structural integrity after deletion, historical model outputs remain valid for the remaining customers. No manual reprocessing, no orphaned records, no model degradation.

Cookieless attribution, same accuracy

NodeNarrative's cookieless tracking mode uses first-party server-side event collection and deterministic identity resolution through hashed identifiers. Unlike competitors who claim "cookieless" but rely on browser fingerprinting (legally grey under the ePrivacy Directive), our approach achieves the same attribution accuracy through the knowledge graph's ability to resolve identities across sessions without client-side tracking mechanisms.

Methodology Questions

Common questions about our attribution models and approach.

How is this different from GA4?
GA4 is a general-purpose web analytics tool with 3 attribution models and a last-click default. It uses sampled data and has no graph analysis. NodeNarrative is purpose-built for marketing attribution with 7 models (including U-Shaped, Time-Decay, Shapley values, and Markov Chain), knowledge graph-based journey analytics, and independent first-party data, with no sampling and no black boxes.
What data do you need to get started?
Shopify store data (orders, sessions, and customer events) via our one-click integration, plus read access to your ad platforms (Google Ads and Meta Ads at launch). No manual CSV imports, no developer setup required. Most teams connect their data sources in under an hour.
How long until I see results?
First attribution data appears within 24 hours of connecting your store. Our progressive model activation means heuristic models (first-click, last-click, linear) work from day one, while advanced models (Shapley values, Markov Chain) activate automatically as the graph accumulates enough journey data for statistical significance, typically 2–4 weeks. You can also import years of historical data at onboarding (3 years on Starter, up to 7 on Professional, unlimited on Enterprise) using a dedicated import allowance that’s completely separate from your monthly event quota. More historical data means more accurate attribution models from day one.
Are the attribution models customisable?
Attribution windows and channel groupings are fully configurable. The mathematical models themselves are standardised, peer-reviewed implementations. We don’t modify the underlying algorithms, because that’s what makes them trustworthy. You tune the inputs, not the maths.
What happens when models disagree?
Divergence is highlighted, not hidden. When all 7 models agree on a channel’s contribution, you can be confident in that insight. When they diverge, it’s a signal to investigate further. This is a feature, not a bug: no single model captures the full picture, which is why we give you all seven.
Is this the same as marketing mix modelling (MMM)?
No. MMM uses aggregate, top-down data (spend vs revenue) for strategic budget allocation. NodeNarrative provides individual journey-level, bottom-up attribution. However, our Shapley value model offers MMM-like fairness at the individual journey level: the best of both worlds.
How does consent affect attribution accuracy?
Every identifier in the knowledge graph carries its own consent status. When a customer grants consent for email tracking but not cookie tracking, the graph respects that boundary at the identifier level. Attribution models only use consented data paths, so your results reflect real, privacy-compliant customer journeys. This is more accurate than competitors who either ignore consent distinctions or exclude entire customers when any single identifier is unconsented.
Can NodeNarrative attribute without cookies?
Yes. Our cookieless tracking mode delivers full attribution accuracy without placing any cookies or using fingerprinting techniques. We use first-party server-side event collection and deterministic identity resolution through hashed identifiers. Unlike competitors who claim "cookieless" but rely on browser fingerprinting (which is legally grey under the ePrivacy Directive), NodeNarrative achieves cookieless attribution through graph-based identity resolution.

See It in Action

Start your 14-day free trial and see what knowledge graph-based attribution reveals about your marketing.

Start Free Trial Request a Demo