Why do standard attribution models fail at firm scale?
Two reasons compound. Volume: data-driven models need conversion counts a typical firm never generates; below that, the models degrade to noisy versions of simple rules. Journey shape: legal clients research anonymously, ask friends, call from a different device than they browsed on, and convert by phone - so the observable digital path is a fragment. A model allocating credit across observed touchpoints allocates across the fragment and calls it the journey. The result reads precise and is structurally wrong.
What can web analytics honestly contribute?
Three things, all valuable and none of them full attribution: source of first contact for sessions that convert on-site, captured as click ids and campaign parameters and passed into intake; content performance - which practice-area pages, attorney profiles, and articles participate in converting paths; and funnel behavior - where inquiry forms shed people. What it cannot see: cross-device research, the referral that preceded the branded search, and most of the phone journey - which is why analytics feeds the model but must not be the model.
What is the matter-level model?
| Component | What it does | Source |
|---|---|---|
| Digital source capture | Click ids and campaign tags stored on the inquiry, never re-keyed | Web forms and call tracking |
| Asked source | "How did you hear about us" recorded verbatim at intake - the only visibility into referrals and offline | Intake conversation |
| Adjudicated source | One field where the two are reconciled per matter by a stated rule, at matter level - thirty judgments a quarter, not thousands | Intake or marketing owner |
| Outcome join | Matter opened, retained, and fees realized, linked back to the inquiry | Practice management and billing |
This model's power is its honesty: digital capture is evidence, the asked answer is testimony, and the adjudicated field records the reconciliation instead of hiding it in a modeling black box. At law-firm volumes, human adjudication per matter is not a compromise - it is more accurate than any statistical model the caseload can feed.
Branded search deserves its own caveat: a client sent by a referral who then searches the firm's name shows up as "organic search". Counting branded search as an acquisition channel systematically steals credit from referrals and reputation - split branded from non-branded before judging any channel.
How should the model change spending?
Quarterly, by channel, on retained matters and realized fees against spend - with confidence proportional to counts. Non-branded search and paid channels get judged on adjudicated matters; content gets judged on participation in converting paths; referral investment gets judged on the asked-source data that only intake discipline produces. And when two channels are statistically indistinguishable at your volume, the honest verdict is "indistinguishable" - budgets split on qualitative grounds beat budgets split on decorated noise.