Multi-touch attribution

Multi-touch attribution you can defend in a client meeting

We build attribution on your own warehouse data rather than accepting each platform's self-report. That means one stitched view of the customer journey, several models compared side by side, and a clear statement of what the numbers can and cannot prove.

Journey stitchingModel comparisonIncrementality tests
Quick answer

Multi-touch attribution built on your own data: cross-platform touchpoint stitching, honest model comparison, and incrementality testing to validate the result.

Every platform claims the same conversion

Add up the conversions each platform reports and you will find you sold roughly twice what you actually sold. Each one uses its own window, its own view-through rules, and its own last-click bias, and each is measuring in isolation. Meanwhile the honest constraint is getting harder: cookie loss and walled gardens mean a full deterministic journey is no longer available for a large share of traffic. Attribution work that ignores that produces confident numbers that are simply wrong.

What we implement

  • Touchpoint collection and identity stitching across paid, organic, email, and direct
  • Deduplication of conversions claimed by multiple platforms, so totals reconcile to actual orders
  • Multiple models run in parallel: last-click, position-based, time-decay, and data-driven where volume supports it
  • Model comparison reporting that shows how budget decisions change under each, instead of one blessed answer
  • Geo holdout and incrementality test design to validate attribution against real lift
  • A documented statement of coverage: what share of conversions are deterministically tracked and what is modeled

How we work

  1. Assess tracking coverage honestly before modeling anything on top of it

  2. Stitch journeys in the warehouse and reconcile total conversions against your order system

  3. Run several models side by side and show where they disagree and why

  4. Validate with a holdout test, because only an experiment establishes incrementality

Typical stack

dbtSnowflakeBigQueryServer-side trackingPythonLooker Studio

Frequently asked questions

Partially, and honesty about that is the whole job. Deterministic cross-site journeys have shrunk, so MTA is now most reliable for logged-in, first-party, and post-click paths. For upper-funnel and cross-device influence, marketing mix modeling and geo experiments answer the question better. Most teams need both, used for different decisions.

MTA works bottom-up from individual tracked journeys and is good at granular, in-platform decisions. MMM works top-down from aggregate spend and outcomes, needs no user-level tracking, and is better for channel-level budget allocation including offline. They answer different questions and should agree directionally; where they do not, that gap is itself informative.

That is the wrong question to answer in the abstract, so we show you several. The useful output is not a single model but seeing which conclusions hold across all of them. Decisions that flip depending on the model are the ones that need an experiment, not a better model.

Yes, and we usually recommend it. Attribution assigns credit within a set of assumptions; only a holdout tells you what would have happened without the spend. We design geo or audience holdouts, size them for statistical power, and read them out.

Go deeper

Get an honest attribution read

We will assess your tracking coverage first and tell you what attribution can credibly prove with the data you have.

Start a project

Proof from our work

Related solutions