# Attribution brief: paid social prospecting budget (fictional example)

Acme Outdoor is a fictional online retailer on example.com. Every number, date and setting below is invented to show the method.

Date: 2026-10-17
Prepared for: head of growth
Status: draft

## Decision

- Decision: cut paid social prospecting from $40,000 to $20,000 a month from November, versus keeping it at $40,000.
- Who decides: head of growth, before the November budget lock.
- Cost of being wrong: a cut that removes demand shows up in new-customer revenue weeks later and takes a quarter to rebuild.
- Costly to reverse: yes. The cut is 50% of the channel's monthly spend, above the 20% threshold.

## Question class

Predict the effect of a budget change. Attribution can frame the hypothesis only.

## Outcome

- Outcome: first paid order per new customer, from the order ledger.
- Metric ID in `martech-context/metrics.md`: `new_customer_first_order_value_v1`, owner head of finance.
- Dedupe rule: one outcome per customer ID.
- Time basis: payment timestamp, UTC.
- Value basis: order value net of discounts and refunds as of 30 days after the order, before tax and shipping, USD only.
- Population: new customers in August 2026, internal and test orders excluded.
- Maturity: the latest order is 2026-08-31; plus the 30-day refund cutoff that is 2026-09-30, and plus GA4's 12 days 2026-09-12, both before the read date of 2026-10-17. Mature view. September orders were left out because orders after 2026-09-17 have not reached their refund cutoff.
- Ledger reconciliation: the outcome table holds 500 customers and $410,000, equal to the `metrics.md` ledger query for August 2026 new customers.

## Observed lag and history

- First observed touch to first order: median 9 days, 90th percentile 41 days. Computed on all touch history since 2025-11-01 with no lookback cap, from June to August 2026 outcomes (each at least 60 days, the longest candidate lookback, after the history start).
- Lag truncated? No: 2% of converters' first observed touch falls on 2025-11-01.
- Lookback chosen: 45 days, covering the 90th percentile.
- Platform views truncated by design: Meta prospecting ad sets use a 7-day click window, 34 days below the p90 lag; Google Ads purchase uses 30-day click-through, 11 days below. Cross-channel questions are answered from the warehouse views below.

## Platform settings read for this brief (2026-10-17)

- GA4: reporting model paid and organic data driven; acquisition key events 30 days, other key events 90 days; channels that can receive credit: paid and organic; `purchase` key event counted once per event; date basis event date; property time zone UTC; value from the event's value parameter, USD. Not used for the views below, which come from the warehouse. Doc: https://support.google.com/analytics/answer/16291112
- Google Ads: purchase action data driven, DDA status `AVAILABLE`; click-through 30 days, engaged-view 3 days, view-through 1 day; counting Every; transaction value from the tag (`always_use_default_value` false); primary and in the standard goal; account time zone UTC. Standard columns are by interaction date; this brief quotes only "Conversions (by conv. time)" for August. Docs: https://support.google.com/google-ads/answer/6259715 and https://support.google.com/google-ads/answer/3123169
- Meta: prospecting ad sets read as 7-day click and 1-day view, engage-through on with a 1-day window, incremental attribution off; the account switched to the link-click definition on 2026-04-14, so August is after it; ad account time zone UTC; date basis mixed, so web purchases sit on their conversion date; purchase value from the pixel and Conversions API, USD. The growth dashboard's connector requests windows `7d_click` and `1d_view`, the same as the ad set setting, with no currency conversion or time zone shift. Docs: https://www.facebook.com/business/news/click-attribution and https://developers.facebook.com/docs/marketing-api/insights/best-practices/
- LinkedIn: not in scope.

## Views

Computed in the warehouse from session touchpoints and the order ledger under contract `new_customer_views_v1`, as CTEs in one read-only query. 500 new customers, $410,000 of first-order value. Shares are of value, unattributed included. "Paid social" is prospecting campaigns only: Acme ran no paid social retargeting in August 2026, checked against campaign names in the naming registry.

- Last non-direct: paid brand search 31%, organic search 22%, email 9%, paid social 6%, referral 2%, unattributed 30%.
- First observed: organic search 27%, paid social 18%, paid brand search 12%, referral 8%, email 5%, unattributed 30%.
- Linear per interaction: organic search 23%, paid brand search 19%, paid social 14%, email 10%, referral 4%, unattributed 30%.
- Meta Ads Manager reports 140 purchases for prospecting ad sets in August under their ad set attribution setting, view-through and engage-through included. Truncated by design (7-day click against a 41-day p90 lag), not comparable with the views above and not additive with them.

Touch source checks passed:

- Of 2,140 users with two or more sessions from different sources, 1.2% have every touch equal to their first-acquisition source (pass, under 5%).
- No payment or own domain in the top ten last-touch referrers, and no touch with `utm_source` example.com.
- Zero touches with a click ID classed as direct.
- 3% of email clicks removed as scanner clicks (within 60 seconds of send).
- Tied first or last touches carry 0.8% of value, under 5%, so no tie-order rerun.

## Coverage

- 320 of 500 customers (64%) have at least one eligible touch.
- $287,000 of $410,000 (70%) by value.
- Unattributed: 180 customers, $123,000.

## Self-reported view

Not used. No "How did you hear about us?" field exists yet.

## Sensitivity

- Paid social credit, August period view (August new-customer value credited to paid social prospecting, divided by August prospecting spend of $40,000): $24,600 last non-direct (attributed ROAS 0.62), $57,400 linear (1.44), $73,800 first observed (1.85). Not a cohort match: with a 9-day median lag, some August orders came from July spend and some August clicks converted in September.
- Lookback 60 days instead of 45: paid social first observed share rises from 18% to 21%.
- The recommended action flips between views: last non-direct supports the cut, first observed and linear do not.

## Result

The views disagree on exactly the question being decided. Paid social appears early in paths and rarely last, the pattern of a channel that creates demand that brand search later captures, or of a channel that reaches people who would have found Acme anyway. Attribution cannot tell these apart. All ROAS figures above are attributed, none incremental.

## Evidence still needed

- A geo holdout: halve paid social prospecting in a matched set of regions for six weeks (42 days). Measure new-customer first orders by region from the ledger through the test end date plus 41 days (the p90 lag), so the readout comes about 12 weeks after launch. Or a Meta conversion lift study if the account is eligible. Hand off to incrementality-test-design, or use its fallback screen in `references/decision-map.md`.
- Feasibility screen:
  - Meta geo targeting can hold the regions apart, orders carry a shipping region, and no regional promotions are planned.
  - Pre-test history: daily first orders by shipping region exist since 2025-11-01, about 350 days, above the 126 days (3 times the 42-day test) that GeoX needs.
  - Response: the gross count of first orders from the ledger, not attributed conversions and not net revenue after refunds.
  - Daily spend by region can be exported from Meta for the same period.
  - Minimum detectable effect from historical regional orders: to be estimated before launch.
- Recommendation to the decision owner: defer the cut until the test reads out, or record in `decisions.md` that the cut goes ahead with the risk accepted. Optional: add an open-text "How did you hear about us?" question at checkout to see whether social is named.

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

- 2026-10-17: paid social prospecting budget decision; attribution views `new_customer_views_v1` (outcome `new_customer_first_order_value_v1`, August 2026 orders, mature on 2026-09-30, 45-day lookback, value net at 30 days) disagree on direction. Recommended: defer the cut to a geo holdout or record the accepted risk. Platform settings as read 2026-10-17. Review when the test reads out, about 12 weeks after launch.