Marketing Attribution Platform Development

Rules-based models -- first-touch, last-touch, linear, time-decay, U-shaped, and W-shaped -- available alongside a data-driven Shapley value model that distributes conversion credit based on each channel's marginal contribution to the outcome. The Shapley value calculation draws from cooperative game theory: it considers every possible ordering of channels in the conversion journey and averages the marginal contribution each channel adds across all orderings, producing a credit distribution that reflects actual causal contribution rather than position in the funnel.

Attribution window configurable by channel and conversion type: a 28-day click plus 1-day view window for upper-funnel display, a 7-day click window for mid-funnel social, and a 1-day click window for bottom-funnel retargeting. These window configurations match the defaults used in Meta Ads Manager and Google Ads, so your platform-reported and model-reported numbers use the same measurement period. Comparison view showing how credit shifts across model types so the marketing team can see the tradeoffs before committing to a model for budget reporting. The comparison is particularly useful for channels like SEO and branded search, which last-touch over-credits and linear models under-credit relative to their actual role in the funnel.

Server-side event tracking to collect touchpoints without relying on browser cookies or client-side JavaScript that ad blockers can strip. The server-side layer receives events from a first-party endpoint on your domain, enriches them with IP-derived geo data and user agent parsing, and fans them out to your data warehouse and directly to ad platform conversion APIs in parallel -- so a single purchase event feeds both your attribution model and Meta CAPI simultaneously. UTM parameter standardisation and validation at ingestion time enforces a consistent taxonomy across campaigns: source, medium, campaign, content, and term values are validated against an allowlist, and malformed or missing UTMs are flagged before they corrupt attribution data rather than discovered in a reporting audit.

Hashed email matching stitches a user's paid social click to their email open and their organic search visit into a single journey using SHA-256 hashed email as the common identifier -- the same hashing standard used by Meta CAPI and Google Enhanced Conversions. Offline conversion import from in-store POS systems, phone sales records, and CRM closed-won events on a scheduled basis, so offline revenue is attributed to the digital touchpoints that preceded it. Ad platform conversion API integration -- Meta CAPI, Google Enhanced Conversions, TikTok Events API -- passes server-side conversion signals to platforms using matched customer data, improving platform-reported match rates and making the attribution system and platform dashboards draw from the same event feed rather than diverging data sources.

Anonymous session to known user merge at the point of email capture or login, so pre-identification touchpoints are included in the attribution journey rather than dropped. The merge uses a retroactive alias call -- the pattern used by Segment and RudderStack's user identity API -- that associates the anonymous ID history with the identified user record, so a user who visited via a paid social ad three days before signing up carries that touchpoint forward into their attribution journey. Cross-device journey assembly uses deterministic matching on login events where available, and probabilistic matching on IP address plus user agent fingerprint for anonymous sessions where the user has not identified across devices.

GDPR and CCPA compliance is built into the identity layer: personal data used for matching is stored in a privacy-safe manner with consent flags per user, and users who exercise their right to deletion have their identifying data purged while preserving the anonymised aggregate event data the attribution models depend on. Customer journey visualisation showing every touchpoint from first visit to conversion with timestamps, channel, campaign, and attributed credit per model. De-duplication of cross-channel touch events so a single visit that fires both a client-side pixel and a server-side event is counted once in the attribution model, using event deduplication keys that match the format used by Meta CAPI and Google Enhanced Conversions to prevent double-counting at the platform level as well.

Geo holdout test design and execution to measure whether a channel is driving incremental conversions or claiming credit for purchases that would have happened anyway. Holdout groups are defined at the geography or audience level: the test cell receives the normal campaign, the control cell is held out of media exposure, and the lift is the difference in conversion rate between the two cells adjusted for baseline variance. The holdout period is sized based on your historical conversion volume to reach statistical significance at 90% confidence before a result is acted on.

Conversion lift study integration with Meta and Google layers platform-native incrementality data alongside your own holdout results -- Meta's conversion lift studies use a randomised holdout at the user level, while geo holdouts capture cross-device and cross-channel halo effects that user-level holdouts miss. Test cell and control cell matching on historical behaviour reduces selection bias by ensuring the holdout group has a comparable pre-test conversion baseline. Incrementality-adjusted ROAS calculation by channel replaces reported conversions with incrementally attributed conversions, so a channel claiming 500 reported conversions that an incrementality test shows drove 200 incremental ones has its ROAS recalculated accordingly. LTV prediction by acquisition channel is layered on top: channels with high incrementality and high 90-day LTV receive higher budget priority than channels with high reported conversions but low incrementality and short LTV.

Bayesian regression-based MMM built on your historical spend and revenue data using PyMC or Prophet as the modelling framework, producing channel contribution estimates at a weekly or monthly aggregation level that is immune to cookie loss and ad blocker interference because it works on aggregate spend and outcome data rather than individual-level tracking. The Bayesian approach produces posterior distributions over channel contribution rather than point estimates, so the model communicates its uncertainty explicitly -- a media planner can see that paid search contribution is estimated at 28% with a credible interval of 22-34%, rather than a single number that implies false precision.

Seasonality and external factor controls are included as model regressors -- holiday calendar effects, economic indicators, and competitor spend proxies -- so the model separates channel contribution from baseline demand shifts that would otherwise inflate or deflate channel coefficients. Spend saturation curve by channel shows where marginal returns start to fall, using an Adstock transformation to model lagged and diminishing effects. Long-run contribution analysis for channels with delayed effects -- SEO, TV, and out-of-home -- that last-click and multi-touch models consistently undervalue because their effects extend weeks beyond the attribution window. Scenario planning tool for budget reallocation based on modelled channel response curves, showing projected revenue impact of shifting spend between channels at different total budget levels.

Attribution output converted into recommended budget allocation by channel, using marginal ROAS curves from the MMM to show where each additional dollar produces the most return. The optimiser solves the constrained budget allocation problem: given a total spend budget and channel-level marginal response curves, find the allocation that maximises projected revenue -- a calculation that produces meaningfully different recommendations from naive "spend more on the channel with the best ROAS" logic when saturation effects are factored in.

Campaign performance dashboard with attribution-adjusted metrics -- not platform-reported conversions but modelled contribution from your own data, pulled from BigQuery or Snowflake via the Google Ads API, Meta Marketing API, and your CRM. Impression share data from Google Ads API and Meta Marketing API is included alongside conversion data so the team can see whether a channel's conversion decline reflects lost impression share or genuine performance deterioration. Weekly automated report for marketing leadership with top-line channel contribution, budget vs. actual, incrementality results, and any attribution anomalies detected since the last report. Integration with Google Looker Studio or Tableau for self-serve analysis by the media planning team. Cross-channel ROAS, incremental cost per acquisition, and LTV by acquisition channel are the three metrics the reporting layer is designed to keep front and centre.