Markentivo
Reporting
Numbers we can defend in a meeting.
A figure that only exists inside an ad platform is not enough to run a budget. We report spend, the results the back office will accept, and the gap between those two systems. The method is written down so a later reader can follow it.
What we measure
The standing objects are media spend, enquiries or orders, the cost of an accepted result, and the share of spend in revenue when revenue can be shared. Repeat purchases are read where the store or CRM can identify a returning customer. Vanity counts such as clicks without a defined next step stay in the appendix.
We look at the back office because platforms optimize to the events they can see. If those events include spam forms, short calls or view-through credit the finance team rejects, the bidding system will buy more of the same. The weekly note states which number is the decision number and which number is context.
Standing objects
- Media spend by channel
- Enquiries or orders
- Cost of an accepted result
- Spend as a share of revenue
- Repeat purchases, when visible
Tracking setup
Events are named for the action, not for the campaign. A purchase, a booked slot and a qualified form stay distinct. Deduplication is checked so a browser hit and a server hit do not both count. UTM rules are short: source, medium, campaign, and a content token only when two ads must be told apart.
Platforms and analytics will disagree. Different windows, view-through credit, delayed orders and filtered bots all move the totals. That disagreement is normal. The work is to keep it stable and explainable. A sudden 40-point swing is a tracking incident; a steady 10-point gap is a documented offset.
In the setup sheet
- Event names and triggers
- Deduplication rule
- UTM convention
- Known platform offsets
- Who can publish a new tag
Reporting artifacts
Four objects stay in circulation. The Friday summary is a short written note. The dashboard is live and uses the same objects as the note. The monthly PDF is the archive a finance lead can file. The change journal lists what was edited, by date, so a movement in cost can be traced to a bid rule, a feed fix or a paused ad.
We do not replace these with a slide deck that exists only for the call. If a chart cannot be produced from the dashboard, it does not appear in the PDF as a one-off drawing.
The set
- Weekly written summary
- Live dashboard
- Monthly PDF
- Change journal
Testing
Every test starts as a sentence: what we change, what we expect to move, and what would make us stop. The queue is visible. We do not run three headline tests and a bidding change on the same campaign in the same week. Sample size and a minimum number of days are set before launch so a noisy Tuesday does not end the test.
We stop when the stop rule is met, when delivery cannot reach the sample, or when a tracking fault makes the reading unusable. A result that sits inside the noise band is recorded as inconclusive. Inconclusive is a valid outcome; it is not rewritten as a win to keep the queue looking busy.
Stop rules
- Hypothesis written first
- One change at a time
- Minimum days and sample
- Inconclusive results stay labelled
Limits of attribution
Last-click and data-driven models both miss quiet work: sales follow-up, a returning buyer who typed the brand, a billboard, a referral that never carried a parameter. We will not promise accuracy to a percentage point. When a number is used for bidding, we say which model produced it and what it cannot see.
Where the decision is large enough, we prefer a geo split or a planned quiet period over another dashboard toggle. Those designs are slower and they produce a cleaner read. They are scoped as incrementality work on Full Funnel or as a written add-on.
What a model misses
- Offline close rates
- Brand demand that arrives direct
- View-through that finance rejects
- Cross-device paths without a login
Illustrative reading of a month
Sample layout. Figures are placeholders used to show report structure.
| Object | Google Search | Meta | Total |
|---|---|---|---|
| Media spend (CAD) | 12,400 | 8,100 | 20,500 |
| Platform-reported results | 186 | 240 | 426 |
| Back-office accepted results | 154 | 131 | 285 |
| Cost per accepted result (CAD) | 80.52 | 61.83 | 71.93 |
| Known tracking offset | View-through excluded | Short calls excluded | — |
Bring the numbers you already trust.
If finance has a definition of an order or a qualified enquiry, send that with the current monthly spend. We will say how we would baseline it and what would stay labelled as context.
Request a working session