Study method
How a study is actually run
The method is a sequence of sittings and file rules, not a secret model. You should be able to explain every credit choice to a finance lead who did not attend the briefing.
1. The question, in one sentence
We will not start a full-channel study on “tell us what worked.” A usable question names channels and a fear: “Did the PLUS-highway boards steal credit from search in the fortnight after they went up?” or “Are app installs being counted as if they were paying users?” The sentence is written on the first page of the method note and is not quietly replaced later.
2. File list, not a treasure hunt
You receive a plain list: media plan by week, invoices, outdoor flight logs, television station logs where you have them, marketplace order extracts without names, store traffic if the doors already count people, and — for app campaigns — install dates separate from first-purchase dates. If a file cannot be produced, that channel is marked unresolved before any credit is assigned.
3. Reconciliation week
Spend is lined against activity. Duplicate marketplace orders, missing panel days, and app installs with no campaign name go into the conflict log. This week is dull and it is where most studies are won or lost. A pretty table built on unreconciled files is how last-click quietly returns.
4. Rules in sentences
Before numbers, we write: the lookback window; how we treat voucher codes; whether first-touch or last-click is the primary table; what even-split alternative we will also show; how festival lag is handled for Hari Raya or Chinese New Year; how app installs are kept apart from in-app purchases. Brand initials the page. Then we apply the rules.
5. Two tables, one readout
The primary table uses the initialled rules. A second page shows the ranking if one sensitive choice (usually the lag window) is moved. The readout is scheduled so brand, agency, and finance hear the same hour. We do not send the agency a private preview that finance never sees.
6. What remains unresolved
Channels without invoices, door counts wrecked by weather, or app files that mix unpaid and paid installs stay on an unresolved line. Leaving them off would make the ranking tidier and less true.