ANALYTICS & REPORTING

Attribution caveats in plain English

Use before presenting channel numbers to someone who will treat them as exact.

attribution-caveats-plain-english.md

Download .md

You are explaining attribution limits to {{AUDIENCE}}, who are not analysts and will act on these numbers.

Setup:
Attribution model in use: {{MODEL}}
Tools reporting numbers: {{TOOLS}}
Lookback windows: {{LOOKBACK}}
Consent and tracking situation: {{CONSENT_SETUP}}
The numbers being presented: {{NUMBERS}}

Write:
1. One paragraph on what these numbers do measure, in plain English, no jargon.
2. A table: Caveat | What it does to the numbers (over-credits X, under-credits Y) | Rough size of the effect | How confident we are in that size.
3. The three channels most likely to be misjudged here, and in which direction.
4. Two decisions this data is good enough to support, and two decisions it is not.
5. One sentence someone can say in a meeting when a number is challenged.

Rules:
- Do not put a precise percentage on an uncertainty you cannot evidence. Use bands like "probably 10 to 30 percent" and say it is a band.
- Do not recommend switching attribution model as the answer unless you explain what that would and would not fix.
- Avoid the words holistic, leverage, and synergy. No em dashes.
- Under 600 words.

Fill in before running

Replace each placeholder with your own detail. The more specific you are, the less the model invents.

{{AUDIENCE}}
{{MODEL}}
{{TOOLS}}
{{LOOKBACK}}
{{CONSENT_SETUP}}
{{NUMBERS}}

Getting a better result

Name the actual tools - the mismatch between them is usually the thing people are arguing about.
Section 4 is what stops the deck being used for a decision it cannot carry.
Reuse the same caveat table every month so the limits stop being news.

Questions about this prompt

When do I run this rather than adding a footnote to the deck?

Before the numbers reach anyone who will act on them without asking how they were made. Footnotes get skipped. This produces a caveat table and one sentence somebody can actually say out loud when a number is challenged in the room, which is the job the footnote was never going to do.

What do I need in front of me?

The attribution model, the tools reporting by name, the lookback windows, the consent and tracking situation, the audience, and the numbers being presented. Name the tools rather than writing analytics. The disagreement people are actually having is usually between two named tools counting the same conversion under different rules.

What comes back?

A plain paragraph on what the numbers do measure, a caveat table with direction and rough size, the three channels most likely to be misjudged and which way, then two decisions the data supports against two it does not. That last pair is what stops the deck being used for a decision it cannot carry.

What is the mistake that costs me here?

Answering the caveats by switching attribution model. The prompt only allows that alongside an explanation of what it would and would not fix, and usually it moves credit around without adding information. Resist a precise figure on an uncertainty you cannot evidence too. Bands are honest, provided you say they are bands.