AI Marketing Attribution

First-touch and last-touch attribution are folklore — they answer 'which channel was last' not 'which channel actually drove the conversion.' OctOpus builds data-driven attribution from your real journey data. Pick the model that matches your funnel shape; the agent fits and ships it.

Which model fits which funnel

Markov-chain attribution for short funnels (3-7 touchpoints) where the journey order matters. Shapley-value attribution for credit decomposition across many touchpoints (10+ ads, several channels). Uplift modelling when you have proper treatment/control exposure data and want incremental contribution, not assigned credit. OctOpus picks based on the journey-length distribution and exposure flag in your data.

What you provide

A journey table: one row per touchpoint, columns for user_id, timestamp, channel, campaign, creative, cost, and a final conversion flag (or revenue). OctOpus profiles the touchpoint distribution, fits the chosen model, and outputs revenue assigned to channel × campaign × creative with confidence intervals.

Calibration matters

An attribution model that ranks channels correctly but is 40% off on absolute revenue is a bad budget tool. OctOpus reports BOTH the rank correlation (Spearman) AND the absolute-dollar error (MAE) on a holdout cohort so you know which decisions to trust.

Key capabilities

Get started free
Drop a CSV. Get a deployed model in minutes.
Launch OctOpus →

Frequently asked questions

What's wrong with last-touch attribution?

It overpays for closing channels (branded search, retargeting) and underpays for top-of-funnel (display, social). Data-driven attribution learns each channel's actual incremental contribution from the journey data, which usually shifts 20-40% of revenue credit away from last-touch and towards upper-funnel awareness.

What's the difference between Markov and Shapley attribution?

Markov-chain attribution computes 'removal effect' — how much would conversions drop if a channel were removed? Good for sequential funnels. Shapley-value attribution divides credit fairly across all touchpoint combinations — game-theoretically optimal but expensive to compute. OctOpus picks based on funnel length.

Does it work with iOS 14+ / cookieless tracking?

Yes — attribution requires journey data, not third-party cookies. As long as you can join touchpoints to a user_id (your CRM ID, hashed email, or first-party device ID), the model works. The accuracy drops when journey coverage drops, which OctOpus flags during data profiling.

How often should I re-run the attribution model?

Monthly is typical — channel mix and campaign performance drift. OctOpus's notebook feature lets you save the attribution session and re-run on fresh data with one click. Set up a scheduled re-run if your data warehouse is on a stable refresh cadence.