mbfinotti/advertising-skills / skills/ad-conversion-tracking
ad-conversion-tracking
VerifiedVerify that conversion events are correctly configured, firing once, and deduplicated before a paid campaign launches, then turn the evidence into a GO / NO-GO decision. Covers tag debugging, test conversion tracing, browser/server event deduplication (including Conversions API setups), value and currency checks, conversion action status, consent handling, and test-data exclusion. Use whenever the user asks if their tracking works, mentions a pixel or tag check, conversions not showing up, duplicate conversions, or a pre-launch checklist - even if they never say 'conversion tracking'. Covers B2B lead gen and B2C ecommerce on any ad platform. Do NOT use to reconcile platform vs CRM numbers after launch - use mbfinotti/advertising-skills@ad-attribution-gap instead.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 29.1 KB |
| evals/evals.json | config | 20.9 KB |
| references/output-report-example.md | markdown | 2.0 KB |
evals/evals.json
View on GitHub{
"skill_name": "ad-conversion-tracking",
"evals": [
{
"id": 1,
"prompt": "We're launching a Google Ads campaign for Fernhollow Analytics on Monday - $38k budget for the first month. I just went through the whole setup: the conversion action 'Demo Request' is created and set as primary, Tag Assistant shows the tag green on the demo confirmation page, and I've got screenshots of every settings screen for the launch doc. The action was created last Tuesday and its status column says 'No recent conversions', but I figure that's just because we haven't launched yet. Can you sign off the tracking line on our launch checklist so we can go live?",
"expected_output": "A refusal to sign off from settings and debugger evidence, a NO-GO verdict pending a real test conversion traced end to end, the 'No recent conversions' status flagged as a blocker, and a report shape with per-check results and a named approver.",
"files": [],
"expectations": [
"Refuses to sign off tracking on the strength of settings screenshots plus a green tag-debugger result alone",
"Requires at least one real test conversion traced end to end before any GO can issue",
"The required trace includes inspecting the actual outgoing event payload, not just observing the tag fire in a debugger",
"The required trace includes the test event appearing in the platform UI with correct value and currency",
"Does not accept the 'No recent conversions' status on a primary action as a benign pre-launch artifact - treats an unhealthy primary-action status as an automatic NO-GO until it reads healthy",
"The verdict issued is NO-GO (or GO explicitly withheld), not a conditional or assumed pass",
"Includes a check that the container/tag works on the live site outside preview or debug mode (published, not only previewed)",
"Allows a few hours of reporting lag before concluding the test event is missing from the platform UI",
"Report distinguishes pass, fail, unknown, and not applicable per check, with unknown never counted as pass",
"The decision block names an approver and a date",
"Does not propose launching and monitoring as a substitute - verification completes before any spend starts"
]
},
{
"id": 2,
"prompt": "Kelpgrove Supply runs on Shopify with the Meta pixel, and our dev added the Conversions API three weeks ago. Since then Events Manager shows almost exactly double the purchases Shopify does. Our dev says the cleanest fix is to just switch off the CAPI purchase events, since the pixel already covers everything, and he can do that today. One thing I noticed while poking around: the pixel sends 'Purchase' and the server sends 'purchase'. Our holiday campaign launches in two weeks - what should we tell the dev to do?",
"expected_output": "Rejects disabling the server events, prescribes a shared per-conversion event ID (order ID) with exact event-name match, explains Meta's dedup key and window, orders the dedup confirmations correctly, and batches the fix into one release.",
"files": [],
"expectations": [
"Rejects switching off the Conversions API events as the fix for the double counting",
"States why disabling the server event is backwards: it silently discards the more resilient signal source",
"Prescribes a shared per-conversion event ID sent identically by both the browser and the server",
"Recommends the order ID as the shared event ID",
"Flags the 'Purchase' vs 'purchase' mismatch as a dedup breaker in its own right - event names must match exactly, including casing",
"Warns the event ID must be generated per conversion, never per page load, since a refresh would mint a new conversion",
"States the browser and server timestamps must land within the platform's matching window",
"States Meta deduplicates on event_id plus event name within a 48-hour window",
"Directs to the dedup rate in Events Manager as the first confirmation, before pulling any browser-vs-server count comparison",
"Cites a healthy dedup benchmark of roughly 90%+ of browser events deduplicated",
"Ships the fix as one batch with a single learning reset, rather than serial one-at-a-time changes",
"Explains that the double counting trains the bidding algorithm on false signal (over-counting exits learning on phantom conversions)"
]
},
{
"id": 3,
"prompt": "Pre-launch review at Brindlewick Outdoor Gear, an ecommerce brand running Meta and Google. Our event match quality score in Events Manager is 9.1/10, so I'm confident dedup is solid. The thing bugging me: Google Ads shows 412 purchases for the soft-launch window but GA4 shows 486 - about a 15% gap - and my CMO wants to hold the full launch until the two numbers match. Our backend order count for the same window is 502. How do I get these numbers to line up before we scale?",
"expected_output": "Corrects the match-quality misread (completeness, not dedup proof), classifies the 15% platform-vs-analytics gap as normal and structural, clears the analytics-vs-backend gap against its threshold, refuses to delay launch chasing a zero gap, and sets up the post-launch ratio watch.",
"files": [],
"expectations": [
"Rejects reading the 9.1/10 match quality score as proof of deduplication - identifies it as a completeness signal, not an accuracy signal",
"States that duplicate conversions can still score high on match quality, so the score alone never confirms dedup",
"Points to the dedup rate in the event manager as the check that actually answers whether dedup works",
"Does not spend the pre-launch window trying to improve the match quality score",
"Classifies the ~15% platform-vs-analytics gap as inside the normal, structural 10-20% band",
"Refuses to hold the launch until the platform and analytics numbers match - names chasing a zero gap as wasted launch-window effort",
"Evaluates the analytics-vs-backend gap (486 vs 502, roughly 3%) against the under-25% threshold and clears it",
"States that an analytics-vs-backend gap above roughly 35% is what would warrant investigation before launch, and this is nowhere near it",
"Recommends tracking the platform-vs-backend ratio weekly after launch, with a stable ratio read as healthy even when nonzero",
"Flags a ratio swinging week to week as the real alarm condition, meaning the pipeline is dropping events",
"Marks ongoing cross-platform reconciliation after launch as a separate task outside this pre-launch verification's scope"
]
},
{
"id": 4,
"prompt": "I run marketing at Corvasse, a B2B SaaS - $15k ACV, sales cycle averages 45 days. Our Google Ads conversion actions are Form Fill, Demo Booked, Newsletter Signup, Pricing Page View, and Closed Won (imported from Salesforce, about 7 per month). All five are set as primary, counting is set to 'Every' across the board, and the click-through window is the default 30 days. I want Smart Bidding to optimize straight to Closed Won since that's what actually matters. Before we double the budget next month, can you check whether this setup makes sense?",
"expected_output": "Demotes the noise actions to secondary, switches lead counting to once, extends the attribution window past the sales cycle, rejects optimizing to a 7-per-month event under the volume threshold, and routes optimization to an upstream event with offline outcome import keyed on the click identifier.",
"files": [],
"expectations": [
"Demotes Newsletter Signup and Pricing Page View from primary - micro and page-view actions go secondary, excluded from bidding",
"Explains that primary designation must map to business outcomes because bidding optimizes only against primary actions",
"Changes lead counting from 'Every' to count-once per click for the lead-gen actions",
"Flags the 30-day click window as shorter than the 45-day sales cycle, so late-closing conversions are simply never counted",
"Recommends extending the attribution window to cover the sales cycle",
"Rejects optimizing Smart Bidding to Closed Won at roughly 7 conversions per month on volume grounds",
"Applies a threshold of roughly 30 conversions in 30 days before optimizing to a funnel stage",
"Recommends optimizing to a higher-volume upstream event and importing downstream outcomes back into the platform",
"Ties offline outcome import to the click identifier being captured and stored on the CRM record",
"Verifies with a test lead carrying a unique QA value, confirmed to land on the CRM record",
"Requires deleting the QA test lead from the CRM as part of test-data cleanup",
"Notes that starved conversion volume traps the campaign below the platform's learning threshold (or cites Google's 15-conversions-per-30-days Target CPA minimum)"
]
},
{
"id": 5,
"prompt": "Agency audit for our client Duneport Legal Software before their campaign relaunch. The launch date is contractually fixed, nine days out, and their only developer is unavailable until next quarter. Traffic is US-only. Here's what we found: (a) the lead form conversion tag fires on every page load, not just submissions; (b) the lead action counts 'every' instead of one per click; (c) an internal traffic filter is defined but sitting in 'testing' state; (d) browser and server events are both sent with no shared event ID, so leads double-count; (e) the cookie banner blocks the tag when a visitor declines; (f) a tracking redirect strips gclid before the landing page. What order do we fix these in?",
"expected_output": "An efficiency-ordered fix list (settings toggles, then trigger work), the dev-dependent shared-event-ID row deleted and named in the report with its cost and acceptor, the consent row deleted as not applicable, the redirect row risk-accepted past the hard date, and one batched release.",
"files": [],
"expectations": [
"Orders the fixes by efficiency - impact bought per unit of fix effort - not by raw severity and not purely by cheapness",
"Puts the settings-class toggles first: the count-once switch and the filter flip (near-zero effort)",
"Flips the internal-traffic filter from testing to active, stating that tagged traffic is not excluded traffic",
"Notes exclusion filters are not retroactive, so the filter must be active before launch, not after",
"Places the trigger correction (tag firing on page load instead of the completed action) in the hour-of-tag-manager-work class, ahead of every dev-dependent row",
"Acknowledges the missing shared event ID as critical impact yet does not rank it first, because it costs a few days of dev coordination",
"With no developer available before launch, removes the shared-event-ID row from this launch's fix list entirely rather than parking it at the bottom",
"Every deleted row is named in the report with the measurement it costs and who accepted the loss - never silently dropped",
"Deletes the consent-path row as not applicable given US-only traffic touching no consent-regulated region",
"Defers the gclid-stripping redirect past the hard launch date as an explicitly risk-accepted item, in writing",
"Everything that makes the cut ships in one batch with a single learning reset - no serial fixes",
"Names what the efficiency order starves (the server-side/click-identifier rebuild) and states a condition that would promote it, such as a setup meant to be trusted for quarters or offline outcome import depending on it",
"Surfaces the unfixed double counting to the client as a named launch risk requiring explicit sign-off, not glossed over"
]
},
{
"id": 6,
"prompt": "We just built the site for Miravelle Skincare - brand new domain, GA4 property and Meta pixel installed four days ago, basically zero traffic so far. Our first ever ad campaign starts next week, targeting mostly France and Germany. I ran a test: opened the site, accepted the cookie banner, bought a €12 sample using our discount code, and watched the purchase tag fire in Tag Manager's preview window. Events Manager still shows nothing though. Can I mark tracking as verified and call it done?",
"expected_output": "A provisional PASS labelled as such (new property, no traffic), conditioned on debug evidence plus a correct payload, with mandatory re-verification on first live conversions, the consent-reject path tested and promoted pre-launch for majority-EU traffic, and refire plus live-site checks completed.",
"files": [],
"expectations": [
"Recognizes the brand-new-property case: the platform UI cannot be expected to show the conversion within the session, since events take hours to appear and a fresh conversion action starts in a waiting status",
"Issues a provisional PASS explicitly labelled as provisional - never a plain GO or 'verified'",
"Conditions the provisional pass on debug-mode evidence plus a correct, inspected payload",
"Mandates re-verification the moment the first live conversions land",
"Requires re-running the test conversion on the consent-reject path, not only the accept path",
"States the reject path is where tracking signal silently dies",
"With consent-regulated traffic the majority of the audience, treats the consent path as a pre-launch fix rather than a risk-accepted deferral",
"Notes that changing what fires on the consent-reject path needs privacy or legal sign-off",
"Requires testing on the live published site outside preview mode, since a container can work in preview and never have been published",
"Requires refreshing the confirmation page after the test purchase to check no second event fires",
"Payload inspection covers the event name, the value (12), the currency (EUR), a per-conversion event ID, and the click identifier",
"The test journey starts by landing with the click identifier in the URL, walking the exact path a paid click takes"
]
}
],
"trigger_queries": [
{ "query": "Can you check if our Meta pixel is set up right before we turn ads on Monday?", "should_trigger": true },
{ "query": "our google ads conversions doubled overnight after we added server side tagging, wtf", "should_trigger": true },
{ "query": "I need a pre-launch checklist for tracking before we start spending on LinkedIn ads", "should_trigger": true },
{ "query": "conversions not showing up in google ads even though people are clearly buying", "should_trigger": true },
{ "query": "how do I verify the purchase event actually fires once on the thank you page", "should_trigger": true },
{ "query": "we're about to put $30k a month behind this campaign - how do I know the tracking can be trusted?", "should_trigger": true },
{ "query": "is my GTM conversion tag firing correctly?", "should_trigger": true },
{ "query": "double counting purchases with pixel + conversions api, help", "should_trigger": true },
{ "query": "tag assistant shows everything green, does that mean tracking works?", "should_trigger": true },
{ "query": "how do I do a test conversion without messing up our data", "should_trigger": true },
{ "query": "do I need to check anything before flipping the campaign live tomorrow?", "should_trigger": true },
{ "query": "the conversion action says 'no recent conversions' - can we still launch?", "should_trigger": true },
{ "query": "should the browser and the server send the same event id?", "should_trigger": true },
{ "query": "our consent banner might be blocking the conversion tag, how do I check?", "should_trigger": true },
{ "query": "how do I exclude our office traffic from conversions before launch", "should_trigger": true },
{ "query": "verify value and currency are being passed on the purchase event", "should_trigger": true },
{ "query": "boss wants a go/no-go on tracking before the campaign spends a cent", "should_trigger": true },
{ "query": "walk me through debugging the facebook pixel on our checkout", "should_trigger": true },
{ "query": "TikTok events manager shows two events per purchase, is that deduped?", "should_trigger": true },
{ "query": "the form thank-you page seems to log a lead every time someone refreshes it", "should_trigger": true },
{ "query": "we just installed gtag, how do we make sure a lead only gets counted once?", "should_trigger": true },
{ "query": "what's a healthy dedup rate for browser vs server events?", "should_trigger": true },
{ "query": "someone told me our smart bidding is learning from junk conversions, how do I check that", "should_trigger": true },
{ "query": "audit our conversion setup before the Black Friday spend goes live", "should_trigger": true },
{ "query": "how do I trace a single purchase from the ad click all the way into the ads platform?", "should_trigger": true },
{ "query": "need to confirm gclid actually makes it into our CRM before we scale spend", "should_trigger": true },
{ "query": "I think our SPA fires the signup event again on every route change", "should_trigger": true },
{ "query": "quick sanity check on the client's conversion tags before their launch?", "should_trigger": true },
{ "query": "event match quality is 9 out of 10, are we good to launch?", "should_trigger": true },
{ "query": "can you make sure our test orders aren't polluting the conversion data", "should_trigger": true },
{ "query": "we launch Friday, what could break our measurement?", "should_trigger": true },
{ "query": "google ads reports 340 conversions this month but hubspot shows 220 closed deals - which one is right?", "should_trigger": false },
{ "query": "our CPA doubled over the last month and I can't figure out why", "should_trigger": false },
{ "query": "which of our campaigns should get more budget next quarter?", "should_trigger": false },
{ "query": "help me pick between meta and linkedin for our new product's ads", "should_trigger": false },
{ "query": "our ads get plenty of clicks but the landing page doesn't convert", "should_trigger": false },
{ "query": "is our CTR dropping because the creative is worn out?", "should_trigger": false },
{ "query": "write me 10 headline variants for a search ad", "should_trigger": false },
{ "query": "should we switch from target CPA to target ROAS on this campaign?", "should_trigger": false },
{ "query": "are we on pace to spend the monthly ad budget or are we behind?", "should_trigger": false },
{ "query": "how big does the seed list behind my lookalike audience need to be?", "should_trigger": false },
{ "query": "design a retargeting sequence for our cart abandoners", "should_trigger": false },
{ "query": "too many ad sets stuck in learning limited - should we consolidate campaigns?", "should_trigger": false },
{ "query": "what negative keywords should I add from this search terms report?", "should_trigger": false },
{ "query": "is a 3.2 ROAS good for an ecommerce store our size?", "should_trigger": false },
{ "query": "set up GA4 ecommerce reports and a funnel exploration dashboard for the team", "should_trigger": false },
{ "query": "how do I A/B test two ad creatives properly?", "should_trigger": false },
{ "query": "map the buying committee we should target for our enterprise deal ads", "should_trigger": false },
{ "query": "reconcile shopify order revenue against what facebook reported for last quarter", "should_trigger": false },
{ "query": "did iOS privacy changes just permanently break our attribution numbers?", "should_trigger": false },
{ "query": "our pricing page converts at 1.2%, how do we improve it?", "should_trigger": false },
{ "query": "track feature adoption events in Mixpanel for the product team", "should_trigger": false },
{ "query": "write a UTM naming convention for all our campaigns", "should_trigger": false },
{ "query": "should we use last click or data driven attribution?", "should_trigger": false },
{ "query": "help me set up server-side GTM infrastructure on our subdomain", "should_trigger": false },
{ "query": "estimate the incremental lift from our brand campaign with a geo holdout", "should_trigger": false },
{ "query": "how do I get more conversions out of the same ad budget?", "should_trigger": false },
{ "query": "the client wants a weekly paid media performance report template", "should_trigger": false },
{ "query": "should we kill this ad or is the dip just seasonal?", "should_trigger": false },
{ "query": "plan how we scale from $10k to $50k a month without tanking CPA", "should_trigger": false },
{ "query": "which ad format works best for app installs on tiktok?", "should_trigger": false },
{ "query": "do we legally need a consent banner for our newsletter signup form?", "should_trigger": false }
]
}
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| finishing-a-development-branchobra/superpowersUse when implementation is complete, all tests pass, and you need to decide how to integrate the work | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| systematic-debuggingobra/superpowersUse when encountering any bug, test failure, or unexpected behavior, before proposing fixes | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| test-driven-developmentobra/superpowersUse when implementing any feature or bugfix, before writing implementation code | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| skill-creatoranthropics/skillsCreate new skills, modify and improve existing skills, and measure skill performance. Use when users want to create a skill from scratch, edit, or optimize an existing skill, run evals to test a skill, benchmark skill performance with variance analysis, or optimize a skill's description for better triggering accuracy. | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |