# Attribution brief: <short title>

Date: <YYYY-MM-DD>
Prepared for: <decision owner role>
Status: draft / approved on <date> by <role>

## Decision

- Decision in one sentence (action, size, date, alternative):
- Who decides, and when:
- Cost of being wrong, and how reversible the action is:
- Costly to reverse? Change as a share of the channel's monthly spend (20% or more counts), or the owner's stated tolerance and whether the downside exceeds it:

## Question class

Pick one:

- Operate a campaign inside one platform
- Describe observed paths
- Predict the effect of a budget change (attribution frames the hypothesis only)
- Discover influences that leave no click (self-reported view)

## Outcome

- Outcome event and grain (order, customer, contact, opportunity, account):
- Dedupe rule:
- Time basis and reporting timezone:
- Value basis, refund cutoff and currency handling:
- Maturity: latest outcome date plus refund cutoff (plus 12 days for GA4 attributed data, 41 days for iOS postback values) versus the read date. Mature, or provisional until <date>:
- Metric ID in `martech-context/metrics.md`:
- Population and exclusions (new vs returning, internal, test):
- B2B only: stage dates taken from (dated stage-entry property or stage history, never current lifecycle stage); pipeline snapshot date; weighted pipeline labelled expected value, if shown:
- Ledger reconciliation: outcome table count and value per currency against the `metrics.md` ledger query:

## Observed lag and history

- First observed touch to outcome lag, median and 90th percentile, from uncapped history and mature cohorts only, date range used:
- Lag truncated (over 10% of converters' first observed touch on the history start date)? yes / no:
- Earliest date of touchpoint data:
- Lookback chosen and why:
- B2B only: anchor date (opportunity create for pipeline created, close for won revenue):
- Platform views truncated by design (platform window below the p90 lag), with the gap in days:

## Platform settings read for this brief

One line per platform, each with read date and doc URL. Every line also states date basis, counting method, time zone and value basis. Full records in the inventory file.

- GA4 (reporting model, lookbacks, channels that can receive credit, conversion-level settings):
- Google Ads (per action: model, DDA status, click, engaged-view and view-through windows, primary or custom goal use, interaction-date or conversion-time columns):
- Meta (attribution setting, engage-through on or off and window, click definition period, incremental attribution in use or not, date basis mixed, each connector's requested windows or ad set default, time zone shift and transforms):
- LinkedIn (counting setting, click and view windows per conversion rule):
- CRM (source fields used and how each is populated, attribution report model or influence models, zero-contact deals or zero-contact-role opportunities):
- MMP, SKAdNetwork or AdAttributionKit, if apps are in scope (postback-only share, app attribution windows and cooldowns, postback maturity date):

## Views

- Primary view and why it fits the question:
- Reference views computed on the same outcome population (at least one):
- Contract and model version, if a multi-touch view is used:

## Coverage

- Outcomes with at least one eligible touch, count share:
- Same, value share:
- Unattributed count and value:

## Self-reported view (when used)

- Question wording, form and where it is asked:
- Raw answers stored per submission (field and table):
- Codebook version used to code answers:
- Response rate for the period and outcome population:
- Shares by code, each with the response rate beside it, shown beside (never added to) the software views. Coded counts only; no raw answers or names:

## Sensitivity

- Assumptions changed one at a time (lookback, new vs returning, brand vs non-brand, identity rule, tie order if tied touches carry over 5% of value):
- Does the recommended action change under any of them? yes / no, which:

## Result

- What the views show, with every number labelled by platform or view, model, window and date basis:
- Each ROAS labelled attributed or incremental, and period view or cohort view:
- Attributed figures and incremental figures kept separate:

## Evidence still needed

- Incrementality test, MMM, self-reported view, or none (and why none is acceptable):
- Hand-off skill:

## Proposed record for `martech-context/decisions.md`

- <date>: <view chosen> for <decision>, outcome <metric ID>, lookback <n> days, model version <name>. Platform settings as read on <date>. Next review <date>.