Marketing Attribution Modeling Services
Every channel claims credit for the conversion. Most of them are wrong. We implement last-touch, Markov-chain, and Shapley-value attribution side by side in GA4, Mixpanel, and BigQuery so the next budget meeting is run on defensible numbers, not on whoever shouted loudest.
What is marketing attribution modeling?
Marketing attribution modeling is the discipline of assigning credit for a conversion across the multiple touchpoints a buyer encounters before they become a customer. A typical B2B buyer touches your brand five to twelve times before signing. Attribution decides how the credit for that signing gets split across all those touches, and the split decides where the next budget dollar goes. Pick the wrong model and you defund the channel that is actually driving pipeline.
Marketing attribution models, defined
An attribution model is a rule for distributing conversion credit across touchpoints. The rule can be rules-based (last-touch, first-touch, linear, time-decay, U-shaped) or data-driven (Markov chain, Shapley value, machine-learned probabilistic). The rules-based models are simple and transparent but ignore the structure of the actual buyer journey. The data-driven models account for which channels actually move buyers forward, but they cost more to implement and require enough data volume to be statistically honest.
Last-touch vs Markov vs Shapley
Three models, three different answers, three different best-fits. Most engagements run all three side by side rather than picking one.
| Dimension | Last-touch | Markov chain | Shapley value |
|---|---|---|---|
| How credit is assigned | All credit to the final touchpoint before conversion. | Credit proportional to each touchpoint’s removal effect on conversion probability. | Credit distributed fairly across coalitions of touchpoints using cooperative game theory. |
| Data needed | Just the last channel before conversion. Available in every analytics tool out of the box. | Full user-level journey path across channels. Typically 300+ conversions per month for stability. | Full user-level journey path. Computational cost rises with the number of channels. |
| Transparency | Trivially explainable. A non-technical exec gets it on first hearing. | Explainable with a transition diagram. Requires one meeting of education. | Defensible mathematically but harder to explain in plain English. Best paired with a worked example. |
| Bias to watch for | Systematically over-credits paid search and direct. Under-credits top-of-funnel channels. | Sensitive to channel definitions. Loose channel grouping distorts the transition matrix. | Computational simplifications at large channel counts can introduce subtle distortion. Document them. |
| Best fit | Baseline number. Pipeline reporting where every team already speaks the language. | Mid-funnel and assist credit. Channel-mix optimization conversations. | Fair allocation across channels for budget defense and board-level reporting. |
How to choose a model
The honest answer is that you do not choose one. You run all three and look at where they agree and where they disagree. Agreement across last-touch, Markov, and Shapley is a strong signal that the channel really does deserve the credit. Disagreement is a flag for further investigation, not a tie-breaker to be settled by picking a favorite model. The model choice for the public-facing number (the one that goes in the board deck) usually defaults to Markov for mid-stage companies and Shapley for late-stage companies whose CFO will audit the methodology.
What tools we set up
Google Analytics 4
Event taxonomy, conversion definitions, channel grouping, and the BigQuery export. We default to server-side GA4 where it improves coverage under cookie consent and iOS-tracking-prevention conditions.
Mixpanel
Product-side event capture for in-app conversions where marketing attribution needs to chain into product analytics. We define the funnel events and the user-identity stitching between anonymous and identified sessions.
BigQuery
The warehouse where everything lands. GA4 export, Mixpanel export, ad-platform spend (Meta, Google Ads, LinkedIn, TikTok), and CRM revenue all join on a documented schema. The attribution models run as SQL and Python jobs against this layer. Your BI tool of choice reads from the same place.
How our process works
Diagnostic, instrumentation, modeling, readout. Every step is in the audit trail. Every methodological choice is documented and reviewed before it ships.
Diagnostic
Map every conversion event, every channel, every source system, and every place credit currently gets assigned. The output is a written attribution-state document and a gap list.
Instrumentation
Stand up the data pipeline. GA4 enhanced ecommerce or custom events, Mixpanel for in-app, ad-platform spend ingest, CRM revenue join. Everything lands in BigQuery on a documented schema.
Modeling
Run last-touch, Markov-chain, and Shapley-value attribution against the same conversion set. Compare the recommendations side by side. Flag the divergences explicitly rather than picking a winner.
Readout
Quarterly report with channel-level credit, model-level disagreement, and a recommendation on next-dollar allocation. Dashboards stay live for the team. Every methodological choice is documented in the audit trail.
Engagement bands
Three shapes of engagement. Numbers are scoped on the discovery call and reflect how much instrumentation already exists. We do not publish public price lists because the answer depends on the stack.
Diagnostic
One-week fixed-scope audit. Source-system inventory, conversion-definition review, current-state attribution gap list, and a recommendation on which implementation band fits.
Implementation
Four to six weeks. Instrumentation, BigQuery pipeline, first modeled output across last-touch, Markov, and Shapley, plus the live dashboards and the experiment-readout template.
Ongoing
Monthly retainer covering model refresh, dashboard maintenance, quarterly recalibration, and a written recommendation on next-quarter channel mix.
Real client results
Snapshot of recent engagements. Anonymized to industry and traffic band. Full case studies under NDA on the discovery call.
B2B SaaS, scaled ARR
Markov re-ranked LinkedIn ahead of paid search.
Last-touch had LinkedIn at the bottom of the channel ranking. Markov showed it was the most common second-to-last touch before high-intent conversions. Budget reallocated with executive sign-off.
Multi-location healthcare
Referral channel finally got honest standing.
Word-of-mouth referrals had been credited as direct traffic for years. CRM join surfaced the true volume. Referral program funded for the first time.
DTC ecommerce
Platform ROAS reconciled to a single ground-truth number.
Meta, Google, and TikTok each claimed double-digit ROAS against the same order set. Shapley reallocated credit to the actual channel mix. CFO finally signed off on Q4 spend.
Who this is for
SaaS
Multi-touch B2B buying journeys with self-serve trials and sales-assisted closes. Attribution is the only way to honestly compare the SDR org against the demand-gen budget.
B2B services
Long sales cycles where the buyer touches your content, your sales rep, your webinar, your case study, and your G2 page before signing. Last-touch gives the rep all the credit. Markov tells the truth.
Healthcare
Patient acquisition across organic search, paid local, referral, and direct. HIPAA-aware pipelines that respect consent boundaries while still producing usable attribution.
Professional services
Legal, accounting, consulting. Referrals matter but are systematically under-credited. We give referral channels honest standing in the model alongside paid and organic.
Multi-channel ecommerce
DTC brands running Meta, TikTok, Google Shopping, email, and influencer at the same time. Each platform reports its own ROAS, all of them are wrong, and the marketing leader needs one defensible number.
What you get
- Documented attribution methodology.
- GA4 + Mixpanel instrumentation review.
- BigQuery data pipeline with documented schema.
- Last-touch, Markov, and Shapley side by side.
- Live dashboards your team owns after handoff.
- Quarterly recalibration readouts on retainer.
What this is not
- Not a paid-media management engagement.
- Not enterprise marketing-mix modeling.
- Not a guarantee of a specific ROAS lift.
- Not a CDP or reverse-ETL replacement.
- Not a cookie-stuffing or fingerprinting workaround for privacy constraints. We respect consent boundaries.
Marketing attribution modeling, answered
What is marketing attribution modeling?
Marketing attribution modeling is the practice of assigning credit for a conversion across the multiple touchpoints a buyer encounters on the way to becoming a customer. The model is the rule for distributing that credit. Different models produce different recommendations for where to spend the next dollar, so the choice of model is itself a strategic decision.
Which attribution model should I use?
Start with last-touch as a baseline because every analytics platform reports it by default and your team already speaks the language. Layer Markov-chain attribution on top once you have enough conversions per month (typically three hundred or more) for the model to be statistically meaningful. Shapley-value attribution is the right call when the executive team needs a defensible, fair allocation across channels for budget meetings. Most engagements run all three side by side so you can see where the recommendations agree and where they diverge.
What tools do you set up?
Google Analytics 4 for event capture and the default channel groupings, Mixpanel for product-side event streams when the conversion happens inside an application, and BigQuery as the warehouse where the raw GA4 export, Mixpanel export, ad-platform spend, and CRM revenue all land together for modeling. We do not push a specific BI tool. Looker Studio, Hex, Mode, and Metabase all work against the same warehouse layer.
How long does the engagement take?
Diagnostic and current-state audit takes one week. Instrumentation and data pipeline takes two to four weeks depending on how clean the source systems already are. First modeled output (last-touch vs Markov vs Shapley side by side) lands in week five or six. Ongoing monitoring is a separate scope.
How does pricing work?
Three bands. The diagnostic engagement is a fixed-scope week to map data sources, conversion definitions, and current model gaps. The implementation engagement covers pipeline build and the first modeled output. The retainer covers ongoing model refresh, dashboard maintenance, and quarterly recalibration. Numbers are scoped on the discovery call. We do not publish public price lists because every stack is shaped by how much instrumentation already exists.
What is NOT included?
We do not run paid media, do not own ad-platform optimization, do not build media-mix-modeling (MMM) econometric models at the scale a Fortune 500 marketing-science team would, and do not promise a specific ROAS lift. We promise honest credit, defensible methodology, and a documented modeling stack your team can run after we leave.
Do you support privacy-safe attribution?
Yes. We default to first-party event collection, server-side GA4 where it improves coverage, and consent-aware pipelines that respect GDPR, CPRA, and HIPAA constraints. We do not stand up cookie-stuffing or fingerprinting schemes. If your privacy posture rules out a specific touchpoint, that touchpoint is excluded from the model and we flag the resulting uncertainty in the readout.
Want an honest attribution diagnostic on your stack?
Twenty minutes on a call. We ask about your conversion definitions, your source systems, and the channel mix the team currently believes. We tell you what is most likely wrong and what an implementation engagement would have to fix.