Setting Up a Marketing Attribution Model: From Last-Click to Data-Driven

Key Takeaways: Marketing attribution models map which touchpoints actually contribute to conversion, so budget decisions rest on data rather than gut feeling. Last-click attribution structurally undervalues awareness channels and overvalues channels close to conversion. This article describes five attribution models, shows how to migrate step by step from last-click to data-driven attribution, and demonstrates how Dutch subsidies reduce the implementation costs.

Last-click attribution costs companies an average of 23% of their marketing budget on the wrong channels

The vast majority of companies still rely on last-click attribution, the model where only the final touchpoint before conversion receives full credit. Research from Harvard Business Review shows that companies switching to multi-touch attribution achieve 15-30% more return on the same marketing budget.

The problem with last-click is fundamental: it ignores the entire customer journey. A customer who discovers your brand through a LinkedIn article, then reads a case study via organic search, and finally converts through a branded Google Ads query gives all credit to that single ad in the last-click model. The LinkedIn article and organic content receive zero credit despite being essential to the eventual conversion.

This leads to systematic errors in budget allocation. Companies overinvest in channels close to conversion, such as branded search and retargeting, and structurally underinvest in awareness channels that fill the pipeline. Google's own research concludes that advertisers switching from last-click to data-driven attribution achieve an average of 6% more conversions at equal budget.

The average B2B purchase process now encompasses 8 to 12 touchpoints spread over 3 to 6 months. In B2C e-commerce it is 4 to 6 over a shorter period. In both cases, assigning all value to a single touchpoint is a serious oversimplification with concrete financial consequences.

Five attribution models compared: from simple to data-driven

Five common attribution models exist, each with specific strengths and limitations. The right model depends on your sales process, number of touchpoints and available data.

The first-click model is the mirror image of last-click: all credit goes to the first touchpoint. This model overvalues awareness channels and undervalues everything after. It is useful if you specifically want to understand which channels attract new visitors, but as a standalone model it is just as limited as last-click.

The linear model distributes credit equally across all touchpoints. With five touchpoints, each receives 20% of the conversion value. This model is fairer than single-touch variants, but it makes no distinction between a casual ad impression and an in-depth product demonstration. Research from Forrester indicates that 38% of companies starting with multi-touch attribution choose the linear model as a first step.

The time-decay model gives more credit to touchpoints closer to conversion but does not entirely ignore earlier ones. It uses a decay curve, often with a half-life of seven days. This model aligns well with shorter sales cycles where recent interactions genuinely have more influence on the purchase decision.

The position-based model, also known as U-shaped, assigns 40% of credit to the first touchpoint, 40% to the last, and distributes the remaining 20% across intermediate touchpoints. This model acknowledges that both the first acquaintance and the conversion moment are especially important. For many B2B companies with a medium-length sales cycle, this is a practical starting point.

The data-driven model, also called algorithmic, uses machine learning to calculate which touchpoint combinations actually lead to conversion based on your own data. Google Analytics 4 offers a built-in variant using Shapley values from game theory. Companies with at least 300 conversions per month can deploy this model effectively, according to Google's own documentation.

The technical foundations: what data do you need

A reliable attribution model requires three data layers: user identification across sessions, complete touchpoint registration, and reliable conversion data with values. Without this foundation, any attribution model produces unreliable results.

User identification is the biggest challenge in the post-cookie era. First-party cookies, server-side tracking and CRM integration are the three pillars that together form stable user identification. Server-side tracking via tools like Google Tag Manager Server Container captures 15-25% additional data that client-side tracking misses due to adblockers and ITP browser restrictions. Analysis from Simo Ahava's tracking research shows that Safari users, who represent approximately 28% of web traffic in the Netherlands, lose their cookies after an average of 7 days due to ITP.

Touchpoint registration requires consistent UTM parameters across all campaigns and central storage of all interactions. Every ad click, email open, organic visit, social media interaction and offline contact must be recorded with a user ID, timestamp, channel, campaign and optionally a content identifier. In practice, 67% of companies have inconsistent UTM tagging, undermining the reliability of any attribution model.

Conversion data must contain not only the number of conversions but also the monetary value per conversion. Without values you cannot calculate which channels deliver the highest ROI. For e-commerce this is relatively straightforward: the order value is known. For B2B lead generation you must derive conversion values from the average deal size multiplied by the close rate per conversion type.

Online-offline data integration is crucial for companies where part of the sales process happens offline. A CRM connection that feeds offline conversions back to online touchpoints is technically feasible via match keys such as email address or phone number. Companies achieving online-offline integration report an average of 20% higher marketing ROI according to BCG, thanks to better insight into the actual customer journey.

Step-by-step implementation: from zero to data-driven in 12 weeks

Implementing a marketing attribution model is most effective in four phases: data audit, tracking implementation, model selection and activation. Total timeline is 8 to 12 weeks depending on the complexity of your marketing stack.

In the first two weeks you conduct a data audit. Map all active marketing channels, inventory which data is currently available, and identify the gaps. Check UTM consistency across all campaigns, verify that conversion tracking is correctly configured, and assess CRM data quality. This audit delivers a concrete overview of what is missing and what needs repair.

Weeks three through six focus on tracking implementation. Implement server-side tracking to minimise data loss. Standardise UTM parameters with a naming convention that all team members follow. Set up a central data warehouse, whether BigQuery, Snowflake or a simpler alternative, where all touchpoint and conversion data converges. The investment for a basic server-side tracking setup is 3,000 to 8,000 euros, a fraction of the marketing budget you will spend more effectively thanks to correct data.

In weeks seven and eight you select and configure the attribution model. Do not start with the most complex model immediately. Begin with a position-based or linear model as reference and compare the outcomes with your current last-click data. This difference analysis reveals which channels are structurally over- or undervalued. With sufficient data volume, at least 300 conversions per month, you can then activate a data-driven model.

Weeks nine through twelve focus on activation: translating attribution insights into budget decisions. Do not shift your entire budget at once. Instead, test with 10-20% shifts based on the new insights. Measure the effect over four to six weeks and scale successful shifts. Companies following this phased approach report an average of 12-18% improvement in marketing ROI within the first quarter, according to Attribution Academy.

Common mistakes that undermine your attribution model

The three most common attribution mistakes are ignoring view-through conversions, combining unequal conversion types, and failing to correct for external factors. Each of these errors can structurally distort your attribution data.

View-through conversions, where a user sees an ad but does not click and converts later anyway, are ignored by most attribution models. Research from Nielsen shows that display campaigns realise up to 60% of their impact through view-through. By excluding these conversions you structurally underestimate the effect of display, video and social media advertising. The solution is setting a realistic view-through window, typically 1 to 7 days, and including these conversions with a weighted factor in your model.

Combining micro- and macro-conversions in a single model clouds your insights. A newsletter signup and a quote request are fundamentally different conversions with different values and different journey patterns. Separate your attribution analysis by conversion type and assign realistic values based on downstream conversion rates. A newsletter signup that leads to an assignment worth 15,000 euros in 3% of cases has an attribution value of 450 euros.

Failing to correct for external factors such as seasonal influences, PR exposure or pricing changes leads to false attribution. A spike in branded search after a press article gets attributed by the model to SEO efforts, while the actual cause is PR. The solution is maintaining an external-factors log and applying corrections for anomalies in attribution data. Companies that do this structurally report 15-25% more accurate attribution outcomes.

Tools and costs: from free to enterprise

Tool selection for marketing attribution ranges from free built-in solutions to enterprise platforms costing 50,000 euros per year. The right choice depends on your data volume, number of channels and desired granularity.

Google Analytics 4 offers free data-driven attribution based on Shapley values. For companies with fewer than 25 million events per month and a limited number of channels, this is a solid starting point. The limitation is that GA4 only fully integrates Google touchpoints and offline conversions must be manually imported.

Mid-market solutions like Ruler Analytics, Dreamdata or HubSpot Attribution cost 500 to 3,000 euros per month and offer multi-touch attribution with CRM integration. These tools suit B2B companies with 50 to 500 conversions per month that want to combine online and offline touchpoints. Implementation time is typically two to four weeks.

Enterprise platforms like Adobe Analytics, Google Analytics 360 or Rockerbox cost 50,000 to 200,000 euros per year and offer advanced cross-device tracking, offline integration and custom modelling. This investment is justified with an annual marketing budget above 500,000 euros, where even small improvements in attribution accuracy yield significant absolute savings.

A custom-built attribution model in Python or R, based on Markov chains or Shapley values, costs 15,000 to 40,000 euros in development and offers maximum flexibility. This is the route for companies with unique data requirements or a complex multi-channel landscape. The WBSO subsidy can cover up to 40% of development costs when the model is technically innovative, which is regularly the case for custom Markov-chain implementations.

Subsidies and financing for attribution implementation

Dutch companies can recoup a significant portion of implementation costs through subsidies. The WBSO and the AIP scheme are the two most relevant programmes for marketing attribution projects.

The WBSO subsidy reimburses up to 40% of wage costs and outsourced R&D for technically innovative projects. Developing a custom attribution model with machine learning, building a server-side tracking infrastructure or creating a cross-device identity resolution system regularly qualifies as WBSO-eligible innovation. For a project scope of 50,000 euros in development costs, this yields a subsidy of approximately 16,000 to 20,000 euros.

The AIP scheme focuses specifically on AI applications in business and can be relevant when your attribution model uses machine learning algorithms. Combining WBSO for the technical development and AIP for the AI component maximises your subsidy potential.

The payback period for a professional attribution model is typically three to six months, excluding subsidies. With subsidies the net investment drops such that break-even is often reached within six to eight weeks. With a monthly media budget of 20,000 euros and a 15% improvement in allocation efficiency, you save 3,000 euros per month, or 36,000 euros per year. That amply justifies even without subsidy an investment of 15,000 to 40,000 euros in a solid attribution model.

Conclusion: start today, not tomorrow

Marketing attribution is not a luxury for large companies with million-euro budgets but a necessity for any organisation that wants to make data-driven decisions about marketing spend. The step from last-click to multi-touch attribution delivers an average of 15-30% more return on the same marketing budget, an improvement directly visible on the income statement.

Start with a data audit of your current tracking, standardise UTM parameters, and activate the data-driven model in Google Analytics 4. Measure the difference with your current last-click data and use those insights to gradually shift budget towards channels that genuinely contribute to growth. With available subsidies via the WBSO and the AIP scheme, the financial barrier is lower than ever.