attribution

Files (6)

• SKILL.md
• evals.json · evals/
• attribution-models.md · references/
• by-business-type.md · references/
• first-party-tracking.md · references/
• measurement-paradigms.md · references/

SKILL.md (first lines)

# Attribution

You help users answer the hardest question in marketing: **which of my efforts actually caused this conversion and this revenue?** Attribution is where marketers lose the most money — to channels that look good in one dashboard and terrible in another, to "direct" and "branded search" that hide the real source, and to models that quietly encode an opinion as if it were fact.

This skill has two pillars. Know which one the user needs before you dive in:

- **(A) Interpretation** — choosing an attribution model, picking a measurement approach, and *reconciling the conflicting numbers* your tools report. This applies to everyone, even with zero engineering.
- **(B) Own your attribution (first-party)** — instrumenting and stitching attribution *yourself* when you control the site/app. This is the build track. Use it when the user says "I want to track this myself" or is hitting a conversion that lives on a domain they don't own.

Most requests start with (A). Reach for (B) only when they control the surface and want to build.

Product context: check for `.agents/product-marketing.md` and read it if present — business type, sales cycle, and primary conversion drive almost every recommendation here.

## Boundaries — what this skill does NOT own

State these up front so you don't rebuild neighboring skills:

- **General event tracking, tracking plans, UTM setup, GA4/GTM** → **analytics**. Attribution *assumes tracking exists*. The line: analytics = "what events and how to fire them"; attribution = "how touches join to conversions and survive to revenue."
- **Ad-platform pixels, CAPI, server-side conversion tracking** → **conversion-tracking**. Attribution consumes platform-reported numbers and corrects for their bias; it doesn't set up the pixels.
- **Pipeline stages, lead lifecycle, CRM revenue dashboards** → **revops**. Attribution feeds pipeline data; it doesn't define stages.
- **Showing up in / measuring AI search** → **ai-seo**. Attribution names AI traffic as a blind spot only.

---

## Pillar A — Interpretation

### 1. What attribution can and can't tell you

Set expectations before touching a number:

- **Attribution is directional, not truth.** It's a model of causality built from incomplete data (cookies expire, sessions fragment, offline touches vanish, people research on one device and buy on another). Treat it as a strong hint, never a verdict.
- **Every model is an opinion.** "First-touch" says the first ad gets all the credit; "last-touch" says the closing click does. Both are wrong in opposite directions. Choosing a model is choosing whose story to believe — say so out loud.
- **The attribution gap is normal, so separate outcomes from credit.** The sum of channel-reported conversions almost always exceeds real conversions, because every platform claims the same sale under its own attribution window and view-through rules. Count real conversions once, deduplicated against one source of truth, and explain platform credit separately. The two won't tie out, and a modeled allocation never replaces the conversion ledger.

When a user demands one true number, reframe: "We can get you a *defensible, consistent* number and a read on which channels are trending up. A single objective truth doesn't exist — here's why, and here's what we use to make decisions anyway."

### 2. Attribution models

The classic rule-based set plus data-driven — and when each one lies. **Last non-direct** is a last-touch variant (skip the junk drawer), not a separate school of thought.

| Model | Credit rule | Best for | How it lies |
|---|---|---|---|
| **First-touch** | 100% to the first known touch | Top-of-funnel / demand-gen valuation; short cycles | Ignores everything that closed the deal; over-credits awareness channels |
| **Last-touch** | 100% to the last touch before conversion | Direct-response, quick e-comm | Over-credits bottom-funnel + branded search/direct; ignores what created demand |
| **Last non-direct** | 100% to last touch, skipping "direct" | A cheap fix for direct pollution | Still single-touch; just moves the blind spot |
| **Linear** | Equal credit to every touch | Long, multi-touch journeys where every step matters | Treats a throwaway visit like a demo; flatters high-frequency channels |
| **Time-decay** | More credit to touches nearer conversion | Longer cycles where recency matters | Under-credits the top of funnel; still an assumption, not a measurement |
| **Position-based (U-shaped)** | 40% first, 40% last, 20% middle | B2B with clear "created" + "closed" moments | The 40/40/20 split is arbitrary; middle touches get shortchanged |
| **Data-driven (algorithmic/Shapley)** | Credit from modeled marginal contribution | High-volume accounts with enough conversions | A black box; needs volume; can't see offline/dark touches it was never fed |

**Platform availability note:** Google Ads and GA4 retired first-click, linear, time-decay, and position-based as selectable reporting models (2023). Those UIs offer **data-driven** and **last-click** (plus GA4's paid-channels last-click variant). Teach the full table as *concepts* and as models you can compute on your own event path / CRM / warehouse — not as Google dropdown options.

**Rules of thumb:**
- Never report a single model in isolation for a long sales cycle. Show **first-touch and last-touch side by side** — the truth lives between them, and the gap between them *is* the insight.
- Data-driven attribution needs volume. Google Ads historically gated DDA behind ~3,000 ad interactions and ~300 conversions in 30 days; hard minimums are gone and DDA is the default, but Google still recommends ~**200 conversions and ~2,000 ad interactions** in 30 days for quality. Below that, DDA often collapses toward last-click priors — noise dressed as science. On Google/GA4 when thin: prefer last-click (or accept that DDA ≈ last-click). Elsewhere (first-party / CRM / warehouse): compute position-based or first+last side by side.
- The model matters far less than being **consistent** and pairing it with an out-of-model sanity check (Pillar A §4, self-reported).

For the model math, worked examples of one journey scored six rule-based ways, and Shapley explained plainly, see `references/attribution-models.md`.

### 3. The three measurement paradigms

Models split credit *within* your tracked data. Paradigms are how you get at *causality* — increasingly rigorous, increasingly expensive:

| Paradigm | What it is | Answers | Needs | Watch out |
|---|---|---|---|---|
| **MTA** (multi-touch attribution) | Stitch user-level touches, apply a model | "Which touchpoints appear on converting journeys?" | Clean cross-device user-level tracking | Cookie loss + privacy gut user-level data; click-paths systematically over-weight search/direct and under-weight impression channels |
| **MMM** (media/marketing mix modeling) | Top-down regression of spend vs. outcomes over time | "What's each channel's aggregate contribution, including offline/brand?" | 2–3 yrs of weekly data, spend variation | Correlational; slow to react; needs real budget swings to learn |
| **Incrementality** (geo holdout, PSA, ghost ads, on/off) | Controlled experiment: exposed vs. withheld | "Did this channel *cause* lift I wouldn't have gotten anyway?" | Ability to withhold; enough volume for significance | The gold standard, but you can only test a few things at a time |

**How to choose:** small budget / short cycle → good UTM + last-non-direct + a self-reported survey beats a fancy model. Mid budget, several channels → MTA for day-to-day + periodic incrementality tests on your biggest line items. Large budget, offline + brand spend → MMM for the portfolio + incrementality to validate MMM's coefficients. Incrementality is the tiebreaker whenever two channels both claim the same conversions.

Decision table by budget × sales cycle × channel count, and how to *read* a geo-holdout / PSA test (not a stats tutorial), in `references/measurement-paradigms.md`.

### 4. Self-reported attribution

The most underused signal, and often the most honest for long cycles and dark social. A post-conversion "How did you hear about us?" survey catches what tracking structurally cannot: podcasts, word of mouth, Slack communities, a founder's tweet, "a friend told me."

- **When it beats tracking:** long consideration cycles, high word-of-mouth, brand/community-led, or heavy dark-social (see §5). If a big slice of your journeys are "direct," you have a self-reported-shaped hole.
- **Ask at the moment of conversion** (signup, first purchase, demo request) — highest recall, before memory fades.
- **Wording:** open-ended ("How did you first hear about us?") captures dark social; a short pick-list is easier to quantify but pre-biases the answer. Best practice: pick-list of your known channels **plus a free-text "other/tell us more."**
- **Treat it as a triangulation input, not gospel** — recall is fuzzy and people credit the *memorable* touch, not the first. **Discount survey share when turning it into credit** (e.g. 22% recall of YouTube is evidence of awareness, not a claim that YouTube deserves 22% of conversions). Pair with incrementality / platform deltas before writing an allocation rule.
- On the build side, this is a form field written to your CRM/analytics as a person property — see Pillar B and `references/first-party-tracking.md`.

### 5. Reconciling conflicting sources

The request behind most attribution work: **"Google says 50, Meta says 40, GA says 60, my CRM says 35 — who's right?"** Nobody is. Here's the framework.

**Why each source systematically lies:**

| Source | Biased toward | Because |
|---|---|---|
| **Ad platforms** (Google/Meta/LinkedIn) | Over-counts *itself* | Claims view-through + click conversions in its own window; every platform counts the same sale; motivated to look good |
| **GA / web analytics** | Last non-direct click | Loses cross-device, loses cookie-blocked users, dumps the unknown into direct |
| **CRM** | Whatever the rep typed / the form captured | Human entry, lead-source overwrites, offline deals with no digital trail |
| **Self-reported survey** | The *memorable* touch | Recall bias; under-counts boring-but-real touches like retargeting |

**How to triangulate:**
1. **Pick one source of truth for the conversion count** — usually your CRM or backend (the system where money is real). Everything else explains *where those came from*, they don't get to redefine *how many*.
2. **Never sum across platforms.** If Google and Meta both claim a conversion, you have one conversion with two claimants, not two conversions. De-dupe against the source-of-truth total.
3. **Read directional agreement, not absolute match.** If every source says paid search is up and organic is down this quarter, that trend is trustworthy even though no two numbers match.
4. **Use self-reported as the tiebreaker** when platforms fight over the same conversions, and **incrementality** when the stakes justify a test. Do **not** turn a platform-vs-GA gap into a multiplier rule (Meta claims 2× GA ≠ "double Meta") — investigate windows and view-through, then ground any override in incrementality.
5. **Expect and budget for the gap.** Report "platforms claim N; we can verify M; the delta is over-claiming + view-through + untracked — here's our best allocation."
6. **Operationalize triangulation as rules when you can** — once you have a holdout or a trusted survey band for a channel, apply a documented **channel-level override** on reporting (shift credit from Direct/branded into the under-credited channel for the test window) rather than pretending raw MTA is truth. Details in `references/measurement-paradigms.md`.

The output is an honest allocation with confidence levels, not a false reconciliation to the decimal.

### 6. The blind spots

Where conversions hide, making real channels look weak:

- **Direct** — the junk drawer. Bookmarks and typed URLs, yes, but also stripped referrers, app-to-web, dark social, and any touch your tracking dropped. A large direct share can include genuine direct visits, repeat users, and measurement loss. Investigate its composition before treating it as a tracking defect.
- **Branded search** — people who discovered you elsewhere and Googled your name. Last-touch hands the credit to paid/organic *branded* search; the real driver was whatever made them search. Segment branded vs. non-branded or you'll defund the top of funnel.
- **Dark social** — sharing that carries no referrer: DMs, Slack/Discord, podcasts, newsletters, screenshots. Structurally invisible to tracking; self-reported is the only way to see it (§4).
- **AI traffic** — assistants and AI search increasingly influence buyers, then send them via branded search or direct, so the AI touch is invisible in analytics. Name it and hand deeper work to **ai-seo**.

When direct and branded search dominate, the most common explanation is a top of funnel that's working while attribution hides it. Confirm before concluding: rule out existing customers, lost referrers, and offline discovery, then check self-reported attribution or run an incrementality test. It's the most common misread in marketing in both directions, so say which one the evidence supports.

### 7. Business-type fork

From coreyhaines31/marketingskills, MIT.

More from coreyhaines31/marketingskills