Revenue Attributor

Revenue Attributor

Purpose

Connect ad spend to on-site conversion signals using Humblytics attribution, and rank campaigns by spend efficiency (sessions and ad_conversions per dollar). True revenue/ROAS, revenue_conversions, and trial_count are only populated when a Stripe/ChartMogul revenue connector is attached to the property — without one, those fields are 0. For true ROAS, join Humblytics spend + on-site conversion data to Stripe/ChartMogul revenue separately. This skill moves marketing teams from raw click counts to spend efficiency, and to revenue accountability once a revenue connector is in place.

When to Use

• Planning next month's ad budget and need data to reallocate
• Diagnosing why spend is up but revenue is flat
• Comparing paid vs organic vs direct revenue contribution
• Auditing campaign-level ROAS across Meta, Google, TikTok, LinkedIn
• Building a quarterly ad performance review for leadership
• Deciding whether to kill, scale, or hold a specific campaign

Connecting Humblytics

Live data comes from the Humblytics MCP (server humblytics). This skill calls mcp__humblytics__* tools — get_ads_attribution, list_meta_connections, get_meta_daily_insights, list_google_ads_connections, and friends — so there are no base URLs, curl calls, or API keys to wire up here. Connect the MCP once (see the repo README); the API key lives in the connection headers, set a single time.

• Never paste API keys into chat — they persist in transcripts and logs. The key belongs in the MCP connection config, never in CLAUDE.md, .cursorrules, or any file committed to git.
• Property: for a single-property key the MCP auto-resolves the property, so nothing to pass. For a multi-property key, call list_properties and pass the chosen propertyId.
• Stripe/ChartMogul requirement: the Humblytics property must have a Stripe/ChartMogul revenue connector attached for revenue attribution to populate. No separate Stripe key is needed here — Humblytics handles ingestion internally. Without a revenue connector, revenue, roas, revenue_conversions, and trial_count come back 0.

Three paths to ad data (Meta Ads + Google Ads)

Path A — Humblytics connectors (preferred, read-only)

If the property has Meta Ads or Google Ads connected at Connectors in the Humblytics dashboard, this skill pulls campaign metadata, daily insights, and full-funnel attribution directly through the Humblytics MCP — no Meta App Review, no Google Ads developer token. This is the default path. The connectors are read-only — they let the agent see campaign data and revenue, but not pause campaigns or change budgets.

• Detect availability via list_meta_connections and list_google_ads_connections. Empty connections arrays mean nothing is connected; fall back to Path A2 below.

Path A2 — User-provided CSV (fallback)

If no connectors are set up, the skill asks the user to paste ad spend from Meta Ads Manager or Google Ads Editor. Pair the CSV columns with Humblytics-attributed revenue from get_ads_attribution (which still works for revenue even without a spend connector — it just leaves spend fields zero).

Path B — Meta CLI (only when the agent needs to manage campaigns)

For pausing laggards, shifting budget, or other write actions, point the user at Meta's official meta ads CLI (released April 29, 2026). It's a published, supported tool that creates resources in PAUSED status by default. Scope the access token to a single ad account, store it in .env (never CLAUDE.md), and review every campaign before flipping it active. This skill does not call the Meta CLI itself — it just hands off when management actions are required.

Path C — Don't roll your own Meta app

Do not give the agent direct Meta Marketing API access through a system user on an unapproved developer app. Routing production API traffic through a draft or unpublished Meta App — regardless of how the access token was issued — is how ad accounts, including long-standing ones with seven-figure spend, are getting permanently banned. Meta is actively enforcing against unapproved-app API traffic. Use Path A or Path B instead. The only safe DIY route is a Meta Developer App with the Marketing API product and full App Review completed for the permissions you need (e.g. ads_read) — budget weeks for review.

Before You Start

1. Confirm a revenue connector is attached — Attribution requires Stripe/ChartMogul revenue events; without one, you only get click data
2. Confirm UTM hygiene — Campaigns without UTM parameters can't be attributed to source
3. Time range — Default to last 30 days; 60-90 days for monthly comparison; 12 months for strategic planning
4. Attribution model — Default to last-touch; switch to first-touch or linear if specified
5. Property — The MCP auto-resolves the property for a single-property key; for a multi-property key, call list_properties and pass the chosen propertyId

Core Workflow

Step 1: Pull Revenue + Spend Data

Primary call — full-funnel attribution merging ad spend with sessions and revenue:

• get_ads_attribution with startDate and endDate (both YYYY-MM-DD)

This returns per-campaign rows with campaign, utm_campaign, platforms, target_urls, top_landing_page, spend, impressions, clicks, sessions, ad_conversions, revenue_conversions, revenue, roas, trial_count, true_cpa, cost_per_trial, click_to_session_rate, avg_hours_to_convert, plus top-level totals, unmatched_ad_campaigns (UTM hygiene gaps), unmatched_utm_campaigns (organic/email traffic), breakdowns, date_range, and connections. NOTE revenue and roas are 0/null on properties without a revenue connector. Empty campaigns[] with non-empty unmatched_* is the strongest signal that UTM tagging needs work — the data is flowing, it just isn't joining.

Spend supplement (Path A — connectors connected):

• Meta: list_meta_connections → pick a connection → get_meta_daily_insights with that connection id, its accountId, and since/until (YYYY-MM-DD)
• Google: list_google_ads_connections → pick a connection → list_google_ads_campaigns with that connection id and customerId

Path A2 fallback (no connectors): ask the user to paste a CSV from Ads Manager. The skill joins those rows to the get_ads_attribution revenue side.

Context tools for source/page-level breakdowns:

• get_traffic_breakdown — UTM source/medium/campaign + device + location dimensions
• get_forms_breakdown — Conversion events (signups, purchases)
• get_pages_breakdown — Page performance (there's no page_group filter; pull all pages and filter client-side if needed)

Step 2: Build the Attribution Table

For each source → campaign → ad level:

Step 3: ROAS Ranking + Waste Detection

Precondition — only tier by ROAS when totals.revenue != 0. If revenue is 0 across all campaigns (no Stripe/ChartMogul revenue connector attached), do NOT tier or recommend kills by ROAS. Instead, report spend plus on-site conversion signals (sessions, ad_conversions) and explicitly flag that revenue attribution is unavailable until a revenue connector is connected. Only apply the tier table below once totals.revenue != 0.

Sort campaigns into four tiers:

Waste detection rules:

• Campaign has > $500 spend and < 1× ROAS → hard flag
• Ad set has > 100 clicks and 0 conversions → landing page or audience mismatch
• Campaign clicks but no UTMs → invisible to attribution, fix tracking first

Step 4: Channel Mix Analysis

Compare revenue contribution across channels:

• Paid share: % of revenue from paid ads
• Organic share: % from SEO/direct
• Referral share: % from partner/affiliate sources
• Blended CAC: Total ad spend / total acquired customers

Red flags:

• Paid share > 80% → business is dependent on ad spend; diversify
• Paid share < 10% with high spend → attribution is broken, fix UTMs
• One campaign > 50% of revenue → concentration risk

Step 5: Reallocation Recommendation

Build a specific reallocation plan:

CURRENT ALLOCATION (monthly):
- Google Ads: $8,000 (2.1× ROAS)
- Meta Ads:   $12,000 (0.9× ROAS)
- LinkedIn:   $3,000 (4.2× ROAS)

RECOMMENDED REALLOCATION:
- Google Ads: $9,000 (+$1,000) — scale branded search
- Meta Ads:   $6,000 (-$6,000) — kill "lookalike-v3", keep retargeting only
- LinkedIn:   $7,000 (+$4,000) — scale the 4.2× ROAS campaign
- Reserve:    $4,000 — test new channel (TikTok or YouTube)

NET CHANGE: $0 (same total budget)
EXPECTED ROAS LIFT: 1.6× → 2.4× (projected)

Step 6: Output Format

PERIOD: [start] to [end]
TOTAL AD SPEND: $X
TOTAL ATTRIBUTED REVENUE: $Y
BLENDED ROAS: Z×

TOP 3 PERFORMERS:
1. [Campaign] — [ROAS]× — [Recommendation]
2. [Campaign] — [ROAS]× — [Recommendation]
3. [Campaign] — [ROAS]× — [Recommendation]

BOTTOM 3 (KILL CANDIDATES):
1. [Campaign] — [ROAS]× — $[spend] wasted
2. [Campaign] — [ROAS]× — $[spend] wasted
3. [Campaign] — [ROAS]× — $[spend] wasted

CHANNEL MIX:
- Paid: X%   Organic: Y%   Direct: Z%   Referral: W%

REALLOCATION PLAN:
[Specific budget moves with dollar amounts]

PROJECTED IMPACT:
[Expected ROAS lift and revenue gain]

TRACKING GAPS:
[Campaigns missing UTMs, ad sets with zero attribution, Stripe integration issues]

Attribution Principles

1. Revenue over clicks. A campaign with 10× the clicks of another means nothing if it doesn't drive revenue.
2. Absolute wins matter, not just ROAS. A 10× ROAS campaign at $200/mo spend matters less than a 2× ROAS campaign at $20K/mo.
3. Kill slowly, scale cautiously. A low ROAS campaign might be the top-funnel driver. Check assisted conversions before killing.
4. UTM hygiene is everything. Without clean UTMs, attribution is fiction. Fix tracking before fixing spend.
5. Blended CAC is the truth. Per-campaign ROAS is useful, but blended CAC tells you whether the business model works.

Creative Inspection (Path A only)

When get_ads_attribution flags a high-spend Meta campaign, drill into the underlying creative before recommending action. get_ads_attribution does NOT return Meta's internal campaign IDs, so you must first resolve the connection and campaign IDs from list_meta_connections — the ads call requires a real Meta campaignId, not the utm_campaign string (an unknown/fake id returns a Graph API "object does not exist" error). Verified live flow (2026-05-29):

1. list_meta_connections → take the connection id (and adAccounts[].id).
2. list_meta_campaigns (with the connection id) → list campaigns per account with their real Meta id, name, status. (Equivalently list_meta_account_campaigns with the connection id and accountId.) Match the flagged utm_campaign/name to its Meta campaign id.
3. list_meta_campaign_ads (with the connection id and campaignId) — list ads in that campaign with creative (thumbnailUrl, imageUrl, title, body), status, adsetId, destination URL.
4. get_meta_ad_creative (with the connection id and adId) — full creative metadata (name, title, body, etc.) for one ad.

Also available when you need connection health or a single account's campaigns over a window: get_meta_connection_status (connection id) for status, and list_meta_account_campaigns (connection id, accountId, since, until) for the per-account campaign list.

Look for: destination URL mismatch (ad copy promises X, lands on Y), broken links, low-quality stock visuals, or one ad in the campaign hogging the spend with poor performance. This is the read-only diagnostic step. To pause the ad, hand off to Meta CLI (Path B).

Related Skills

• funnel-reporter — End-to-end funnel metrics for the periods you're analyzing
• ad-expert — Creative and targeting fixes once you've identified waste
• cro-optimizer — If ads are working but landing pages aren't converting, pair with CRO