Why this matters
The attribution settings guide is explicit that the window decides which conversions delivery learns from, not only which appear in reports. A team that flips the setting after every weak month retrains its campaigns each time and breaks its own history. And no window helps if events are thin: the low event match quality page shows how a purchase Meta cannot connect to a person counts under no setting at all.
The brand, its path to purchase and its events
Blushwood Jewellery is an invented online jeweller with a bespoke engagement ring service run through video consultations. Customers browse for weeks before buying, and the team has changed the attribution setting on its purchase campaigns several times this year, each time after a disappointing month.
Blushwood Jewellery: lag, windows and economics
| Time from most recent Meta ad click to order (shop records) | Share of orders |
|---|---|
| Same or next day | 31% |
| 2 to 7 days | 29% |
| 8 to 28 days | 26% |
| Over 28 days | 14% |
Median time from first visit to order: 23 days. Last month's purchase campaigns, spend £18,000:
| Setting | Purchases | Revenue |
|---|---|---|
| A: 7-day click and 1-day view (current, the default) | 214 | £74,900 |
| B: 7-day click only | 151 | £52,850 |
| C: 1-day click only | 82 | £28,700 |
Contribution margin after returns: 55%. Owner's target ROAS: 3.0, window never recorded. History: A until March, C from May, A again from July.
Blushwood Jewellery: event diagnostics and the bespoke service
- Purchase sent by pixel and server integration with a shared event ID; rating low.
- Server coverage: email 98%, phone 0%, browser ID 9%, click ID 4%, IP and user agent 100%.
- Diagnostics warn of badly formatted email; a test payload shows a mixed-case address hashed as typed.
- Checkout collects a phone number on every order. About a third of sessions decline advertising cookies.
- Bespoke: a form fires a Lead event that one campaign optimises for. Last quarter, 240 forms, 96 consultations attended, 38 deposits paid at an average commission of £2,400, all recorded only in the CRM.
What the setting changes
The click part of the window sets how slow a buyer can be and still teach delivery something; a short click window trains the system towards people who buy almost at once. The view part adds people who saw an ad without clicking, which gives a modest account more examples but is weak evidence among people already browsing. Every change is a significant edit that restarts learning. Purchases beyond the longest click window are invisible to every setting; that gap is closed by measurement outside Meta and by better matching, not by hunting for a longer window.
Task
- Work out the return under each setting, the view-through share and the break-even return.
- Decide which setting the purchase campaigns should optimise under, explaining the effect on delivery and on reports separately.
- Write the reporting rule: which window the owner's target is judged against and what every report shows.
- Diagnose the low match rating and order the fixes, with how each is verified.
- Plan which bespoke outcomes the CRM sends back, with which identifiers, and what the consultation campaign optimises for.
- Write the rule for any future change to the setting.
Worked answer
Blushwood's returns by window
- ROAS: A ≈ 4.16, B ≈ 2.94, C ≈ 1.59. Break-even: 1 ÷ 0.55 ≈ 1.82.
- View-through share under A: (214 − 151) ÷ 214 ≈ 29%.
- Click-path orders inside 7 days of the last click: 60%; inside 1 day: 31%.
- Judged under B, the month sits at 2.94 against the 3.0 target: just short of target, comfortably above break-even.
The setting. Keep A for delivery and stop changing it. Its week-long click window covers most click-path buyers; C trained delivery towards the fast minority and starved the considered purchases that make the business. The view component adds learning examples. B is defensible, but the case for it is honest reporting, which the rule below delivers without another learning reset.
The reporting rule. Judge the target under B, the click-only view, because a click is the stronger evidence. Every report shows A and B side by side with the window in the heading. The view-through credit sits mostly in remarketing, among people already browsing, so it is flagged for the holdout work in the next module rather than counted as earned ROASRevenue generated per unit of ad spend..
Event matching, in order. Normalise email before hashing, as the Conversions API guide documents, with a unit test that hashes a known value. Then read the browser and click identifiers from cookies at session start, store them on the order and send them with the server event. Then add the phone number checkout already collects, with its country code, once the privacy notice covers it. Verify each by test event, cleared diagnostics and rising coverage over the following weeks. Leave visitors who declined alone: their events should not exist. Record all of it in a parameter contract for the Conversions APIServer-side event sharing with Meta. integration.
Bespoke outcomes. As the guide to connecting a CRM to ads and analytics sets out, the form captures hashed email and phone, the click and browser identifiers and the consent state on the CRM record. The CRM sends consultation attended and deposit paid, the latter with its value. Attended consultations are frequent and early enough to optimise for, whereas form fills reward people who never turn up. Deposits are too rare to optimise for but put commission value into reporting.
Future changes. Proposed in writing, tested on one ad set, logged with a date and followed by a fresh baseline.
Common mistakes
- Choosing the window that flatters the month. It retrains delivery and makes earlier reports incomparable.
- Reading a long path as a case for views. The lag that matters runs from the last click.
- Optimising a considered purchase on the shortest click window. Delivery learns to find impulse buyers.
- Reading coverage as usable data. A mis-hashed address counts as present and matches nobody.
- Sending form fills from the CRM. The pixel already sees them; send the stages it cannot.
Next, test the same judgements in the signals and attribution checkpoint, or revisit the intermediate Meta Ads level. How credit is shared across touchpoints in general sits under AttributionAssigning conversion credit to touchpoints..
