Skip to content

Checks

Tracking audit

Checks each ad platform's conversion tracking against your CRM, so no budget decision rests on conversions that are double counted, unmatched or never sent back.

Other tools

Add this skill to Claude

Claude takes skills as a ZIP file. Your download has started; if it didn't, use the button below.

  1. 1In Claude, open Customize, then Skills.
  2. 2Choose +, then Upload a skill, and pick tracking-audit.zip.
  3. 3Ask for it by name, or let Claude use it when a task matches.

Skills need code execution turned on in Claude's settings. On a Team or Enterprise plan, an admin can share it with everyone.

Download the folder

Try it now, without installing

See the skill work on a made-up input before you add it. Nothing is installed, and you can swap in your own input.

Open in the Claude app

Opens a new chat in the Claude desktop app with the prompt filled in. Nothing is sent until you press enter.

Includes the whole SKILL.md and the sample input. Paste it into Claude, ChatGPT or any chat.

Influenced by
Enhanced conversions for leads, Conversions API documentation
Needs
The paid platforms you use, with their budgets · Each platform's conversion actions (source, what bidding uses, counting, window)
Works with
Claude, ChatGPT, Codex, Cursor and any agent that reads SKILL.md
Version
1.0.0 · Last checked · License MIT
Maintained by
Sabih AhmedMarkOps
SkillMarkOps

Tracking audit

Example output

Can budget rest on it?
PlatformFindingCall
Google AdsDemo form counted twiceFix first
LinkedIn AdsConversions match the CRMTrusted
Meta AdsNo settings sharedNot checked

Checks each ad platform's conversion tracking against your CRM, so no budget decision rests on conversions that are double counted, unmatched or never sent back.

The skill

The whole skill, in plain text.

Nothing is hidden. Read it, change it to fit your team, or take the whole folder.

Scripts
Optional Python scripts for the counting. It works without them and says which way it ran.
Network
Opens web pages or your tools if your agent can. Works from pasted data if not.

3 test cases

tracking-audit/SKILL.md

In the folder

  • tracking-audit/SKILL.md
  • tracking-audit/references/platform-checks.md
  • tracking-audit/scripts/reconcile.py
---name: tracking-auditdescription: >-  Checks the conversion tracking behind each ad platform a B2B SaaS company pays for (conversion actions, what  bidding optimizes for, enhanced conversions for leads, click IDs in the CRM, consent, offline imports and CRM  stage mapping), reconciles platform conversions with CRM leads, and returns a pass or fail table per platform with  severities and the fix order. Use it before any budget, bid or channel decision rests on platform conversion data,  after a site or form change, or when platform and CRM figures disagree. Not for UTM naming (use utm-check), for  designing which CRM stages to send back (use crm-feedback-loop) or for pipeline by channel (use pipeline-report).license: MITcompatibility: >-  Works in Claude, ChatGPT, Codex, Cursor and any agent that reads SKILL.md. Reads a connected ad account or CRM  read-only if the agent has one, otherwise exports, screenshots or pasted rows. Python 3 is optional, for the  reconciliation.metadata:  author: SHAKEUP  version: "1.0.0"  stage: check  plugin: markops  homepage: https://www.shakeuphq.com/skills/tracking-audit--- # Tracking audit Finds out whether each ad platform's conversion data can carry a budget decision, and what to fix first when it can't. ## Purpose - One job: check the measurement chain from ad click to CRM stage, per paid platform, and say which platform's data is safe to act on.- Stage: check. It ends in a verdict with severities.- It never changes an account, a tag or a CRM setting. It reads, checks and writes the fix list.- It never passes a check without evidence. A check with no input is `not checked`, never `pass`. ## Use when / not | Situation | Use || --- | --- || A budget, bid or channel decision is about to rest on platform conversions | this skill || Platform leads and CRM leads disagree | this skill || A form, site, tag manager or CRM change went live | this skill || UTMs land in the wrong GA4 channel or in Unassigned | `utm-check` || Planning which CRM stages to send back to each platform | `crm-feedback-loop`, after this skill || Pipeline by channel for the board | `pipeline-report` | ## Inputs | Input | Required | Ways to provide, in order | If missing || --- | --- | --- | --- || Paid platforms in use | yes | `## Channels` in the profile; the reader's request | ask once || Conversion actions per platform: name, source (tag, import, API), what bidding uses, counting, window, status | yes, per platform | a connected ad account, read-only; an export or screenshot of the conversions list; pasted rows | that platform is `NOT CHECKED` || Platform lead conversions for one recent period, click-through only | no | connected account; export; pasted figures | skip the reconciliation for that platform || CRM leads created in the same period, with source, medium, click ID columns and whether an email exists | no | a connected CRM; a CSV export; pasted rows | skip the reconciliation and the click ID check; coverage `partial` || Offline import status: method, last upload, upload frequency, errors in the last 7 days | no | the platform's upload history or diagnostics; the reader's answer | D checks `not checked` || Diagnostics: enhanced conversions, server event status, match quality | no | each platform's diagnostics screen | B4 `not checked` || Consent setup: banner, consent mode, which regions see it | no | browse the site; a tag debugger output; the reader's answer | C checks `not checked`; open item if the profile's regions need consent || CRM stages: names, entry rule, date fields | no | CRM settings; a pasted list | E checks `not checked` | ## Profile read From `company-profile.md`: `## Company` (Regions, Sales cycle, Business model), `## Channels` (paid platforms, status, monthly budget), `## Targets` (target cost per lead, the rates a budget decision uses), `## Accounts and data` (CRM, analytics, which ad accounts are connected or export only), and every `## Learnings` entry for `tracking-audit` with status `noted`, `fact` or `accepted`. No profile: ask in one message for the paid platforms with their monthly budgets, the CRM, and the regions the company sells in. Say that `company-profile` saves this next time. ## Steps 1. **Apply learnings.** Read them first. An accepted learning that changes a check or a threshold overrides the default here; name it in the output when it fires. 2. **Scope the platforms.** List every paid platform with status `active` or `testing`. Order them by monthly budget, largest first: the biggest budget decision gets checked first. A platform with no budget figure goes last and is named in open items. 3. **Gather the inputs per platform** in the fallback order. Read connected accounts only; never save or change a setting. Record where each input came from. 4. **Run the checks.** Use the tables below on every platform in scope. Platform screens, field names and limits are in `references/platform-checks.md`: open it at this step. Give each check 1 of 4 results: `pass` (evidence shown), `fail` (with its severity), `not checked` (the input is missing; say which input would check it) or `n/a` (it doesn't apply, such as upload checks when nothing is uploaded).    **A. Conversion definitions**    | ID | Passes when | Severity if not | Why |   | --- | --- | --- | --- |   | A1 | At least 1 conversion action is a CRM lead or a later stage (a form that creates a CRM record, or an imported stage) | critical | Bidding has nothing to learn that the CRM agrees with |   | A2 | What bidding uses is a lead or later event, never a page view, click, scroll, chat open or video view | critical | The platform buys the cheap event, not the lead |   | A3 | Each real lead is counted once: no tag and GA4 import for the same form both used for bidding; pixel and server events share an event ID | critical | Double counting halves the reported cost per lead |   | A4 | Lead actions count once per click or person, not every submit | warning | Repeat submits inflate leads |   | A5 | The conversion window is at least the reader's median days from click to the stage imported, inside the platform's limit | warning | Stages after the window are dropped |   | A6 | Every live campaign is attached to the conversion it is measured on | warning | Unattached campaigns show 0 and bid blind |   | A7 | Conversion values, if used, come from the reader's data and say how; no placeholder value such as 1 | warning | Value bidding chases a made-up figure |    **B. Identifiers and matching**    | ID | Passes when | Severity if not | Why |   | --- | --- | --- | --- |   | B1 | 80% or more of the platform's CRM leads carry a match key (its click ID or an email) | critical under 50%, warning from 50% to 79% | Imports match only rows with a key; under half, bidding learns from whichever forms happen to keep the ID |   | B2 | Google: enhanced conversions for leads is on, the tag collects hashed email or phone on the lead form, and the import sends the same hashed data | warning; critical when there is neither a GCLID import nor enhanced conversions for leads | Leads without a stored GCLID can't be matched |   | B3 | Server events send hashed email plus the platform's click ID where stored, and share an event ID with the pixel | warning | Unmatched or duplicated events |   | B4 | Each platform's diagnostics show no open errors | warning | Failures there are silent everywhere else |    **C. Consent**    | ID | Passes when | Severity if not | Why |   | --- | --- | --- | --- |   | C1 | Where the profile's regions include the EEA, the UK or Switzerland: a consent banner decides before ad tags set cookies or send data, and Google consent mode receives the choice | critical | Data is sent for people who refused, or lost for people who accepted |   | C2 | Uploads to Google carry the ad user data consent value | warning; critical for EEA traffic | Google's API says uploads without it may not be attributable |   | C3 | No personal data in URLs (an email in a thank-you page address); user data sent to platforms is SHA-256 hashed where the platform expects hashes | critical | Personal data reaches every tool that reads the URL |    **D. Offline imports**    | ID | Passes when | Severity if not | Why |   | --- | --- | --- | --- |   | D1 | Each platform that bids receives at least 1 stage past the lead (MQL, SQL, opportunity or closed-won) | warning | It bids for leads, not pipeline: hand to `crm-feedback-loop` |   | D2 | Uploads run at least daily, and the gap between uploads never exceeds the platform's event age limit | critical when the gap exceeds the limit; warning when uploads are less than daily | Late events are rejected (Meta and OpenAI: 7 days) |   | D3 | Upload errors in the last 7 days: none | warning above 0; critical above 10% of rows | Above 1 in 10, the signal is too damaged to read conversion data |   | D4 | Each imported event carries the date the stage happened, not the upload date | critical | Wrong dates move conversions across days and out of windows |    **E. CRM stage mapping**    | ID | Passes when | Severity if not | Why |   | --- | --- | --- | --- |   | E1 | Each lifecycle stage has 1 written entry rule (a field value or an event) | warning | Stages mean different things to sales and marketing |   | E2 | Each stage records the date it was entered | critical | No stage dates: no imports and no timing |   | E3 | Stages sent to platforms map 1 to 1 to conversion actions with matching names | warning | Reports and bidding read different events |   | E4 | New leads with a blank or "offline" source are fewer than the leads of the smallest paid platform in scope | warning | The unknown bucket is bigger than a channel you are judging |   | E5 | One person is one record before leads are counted or uploaded | warning | Duplicates inflate leads and uploads | 5. **Reconcile platform conversions with the CRM.** Run the script with the period's CRM leads and each platform's click-through lead conversions:    ```   python3 scripts/reconcile.py --crm leads.csv --count google=48 --count linkedin=21 --start 2026-09-01 --end 2026-09-30   ```    Without Python, count by hand for each platform: the CRM leads created in the period; those carrying the platform's click ID, or a paid source and medium for it; the share of those with a click ID or an email. Then apply the same rules: platform conversions above the CRM's total leads plus 5% is critical (ceiling); conversions with no CRM lead tied to the platform is critical (orphan); a ratio of platform conversions to the platform's CRM leads below 0.7 or above 1.3 is a warning (the band is wide because windows, time zones and dedup rules differ). Never add platform conversions together to count leads: each platform credits itself for shared leads. Say which way you ran it. 6. **Grade each platform.**    | Result | Condition (first match wins) |   | --- | --- |   | `NOT CHECKED` | No conversion settings for the platform |   | `FIX FIRST` | 1 or more critical findings |   | `CHECK` | Warnings only |   | `TRUSTED` | Every check passed; `not checked` items listed | 7. **Write the fix list.** One line per failed check: the fix, the screen or field where it happens, and the role that does it (from `## Channels` owners if named). Order: critical before warning, then by platform budget. Never write a fix that changes a target or a rate. 8. **Say what budget decisions can rest on.** Per platform: `safe` when it is TRUSTED, or CHECK with no warning in A, and B1 and the reconciliation were both run and passed; otherwise `not yet`, naming the check that blocks it or the input that would check it. A check that wasn't run never makes a platform safe. ## Check and verdict | Verdict | Condition (first match wins) || --- | --- || `BLOCKED` | No paid platform named, or every platform in scope is `NOT CHECKED` || `FAIL` | Any platform with budget is `FIX FIRST` || `PASS WITH NOTES` | Warnings only, or a platform is `NOT CHECKED` while others pass || `PASS` | Every platform `TRUSTED` | Named stop: `BLOCKED` asks for the missing input (the platform list, or the conversion settings of the largest platform). No other stops. ## Learn Append a learning to the profile when: - the reader overrides a severity or a check ("our demo and pricing forms are 2 real leads, so we count every submit"): write the reason as a `proposal`;- a fix is confirmed later (upload errors gone, match keys up): write a `fact` with the new figure and its source;- a platform limit in `references/platform-checks.md` turns out to have changed: write a `fact` naming the page and the date read. Example: `- 2026-10-24 | tracking-audit | fact | LinkedIn match keys rose from 41% to 88% after li_fat_id was added to the demo form | Why: CRM export, 2026-10 | Status: noted` ## Output ```Verdict: <PASS | PASS WITH NOTES | FAIL | BLOCKED> · Coverage: <complete | partial (what's missing) | blocked> · Reconciliation: <script | by hand | not run> Budget decisions:- <platform>: <safe | not yet (check ID and why)> | Platform | Check | Result | Severity | Evidence | Fix | Owner |(platforms by budget, largest first; inside each, critical first; "not checked" rows say what input would check them) Reconciliation: <platform>: <conversions> platform vs <n> CRM leads tied to it (ratio <r>), match keys <n%>; CRM leads in period <n>Per platform: <platform>: <TRUSTED | CHECK | FIX FIRST | NOT CHECKED>Fix order: <numbered list>Learnings applied: <list, or none>``` Then the handoff block. ## Hand off ```handoffskill: tracking-auditversion: 1.0.0verdict: FAILcoverage: partial (Meta not checked: no access)for: <company> paid platforms, <period>outputs:  platforms: {google: FIX FIRST, linkedin: CHECK, meta: NOT CHECKED}  critical: 2  match_key_rate: {google: 0.62, linkedin: 0.91}  imports_past_lead: {google: none, linkedin: none}  stage_dates: yesnext: [utm-check, crm-feedback-loop]open: [Meta conversion settings]``` `crm-feedback-loop` reads `match_key_rate`, `imports_past_lead` and `stage_dates`. `pipeline-report` reads the verdict to say how far platform figures can be trusted beside CRM figures. ## Failure modes | Problem | Do this || --- | --- || Only screenshots, no exports | Check A, C and D from the screenshots; mark B1 and the reconciliation `not checked` || The CRM export holds emails | Count whether an email exists; never repeat an address in the output || The reader asks whether their setup is legal | Say the skill checks the setup against the platforms' documented rules, not the law, and name the question for their legal adviser || Platform and CRM use different time zones | Widen the period by 1 day on each side for the reconciliation and say so || The reader asks for a "normal" match rate or conversion rate | Say there is none in this skill; compare against their own last period instead || A platform's documented limit differs from `references/platform-checks.md` | Use the platform's current page, cite it with today's date, and write a `fact` learning || Google Ads auto-tagging is off | B1 for Google fails: no GCLID reaches the site; the fix is to turn it on | ## Influenced by - Google's documentation on enhanced conversions for leads, an upgraded offline import that matches hashed form data to signed-in users: https://support.google.com/google-ads/answer/15713840?hl=en- LinkedIn's Conversions API documentation, for match keys, the 90-day event limit and campaign association: https://learn.microsoft.com/en-us/linkedin/marketing/integrations/ads-reporting/conversions-api- SHAKEUP's MarkOps practice: tracking is checked per platform, ordered by budget, and every check needs evidence before it can pass.

How it thinks

Inputs in, a verdict out, and a stop only where you decide.

What we built

We check tracking per platform, in budget order, and reconcile every platform against the CRM before anyone reads its cost per lead. A check without evidence stays not checked, so a partial audit can't pass for a clean one.

It reads your company profile if you have one, or asks for only what it needs. Then it works through its steps, checks its own work and ends in a verdict.

The steps

  1. 01

    Read your learnings first

    Exceptions your team accepted, such as counting every submit on 2 different forms, change the defaults before it starts.

  2. 02

    Order platforms by budget

    The largest budget gets checked first, because the biggest decision rests on it.

  3. 03

    Run 23 checks per platform

    Conversion definitions, click IDs and matching, consent, offline imports and CRM stage mapping, each with the severity it carries and why.

  4. 04

    Reconcile with your CRM

    Its script compares platform conversions with the CRM leads tied to each platform. A platform claiming more leads than your CRM created fails at once.

  5. 05

    Grade each platform

    Trusted, check, fix first or not checked. A check with no evidence is never a pass.

  6. 06

    Say what budget can rest on

    Per platform, safe or not yet, with the check that blocks it and the fix order for your team.

What you need

  • Required

    The paid platforms you use, with their budgets

    The Channels section of your company profile → Your request

  • Required

    Each platform's conversion actions (source, what bidding uses, counting, window)

    A connected ad account, read-only → An export or screenshot of the conversions list → Rows you paste

  • Optional

    CRM leads for the same period, with source, medium and click ID columns

    A connected CRM, read-only → A CSV export → Rows you paste

  • Optional

    Offline import status and platform diagnostics

    Each platform's upload history and diagnostics screens → Your answer

  • Optional

    Your consent setup

    It reads your site if your agent can browse → A tag debugger output or your answer

What it can return

  • PASS

    Every platform is trusted.

  • PASS WITH NOTES

    Warnings only, or a platform couldn't be checked while the others pass.

  • FAIL

    A platform with budget has a critical finding, such as double counting or no consent banner where you need one.

  • BLOCKED

    No platforms named, or no conversion settings for any of them.

Where it stops for you

  • When no paid platform or conversion settings can be found after one question

Example output

What you get back.

A run for a made-up company. Your output depends on your inputs.

Here's a run for a made-up B2B SaaS company that sells to the US and the UK. It wanted to move budget from LinkedIn Ads to Google Ads because Google's cost per lead looked lower.

Verdict

FAIL

Coverage: complete

Reconciliation: script

PlatformCheckResultSeverityFix
Google AdsEach lead counted once2 primary actions for the same demo formCriticalKeep the tag as primary and set the GA4 import to secondary
Google AdsReconciliation46 conversions against 22 CRM leads, ratio 2.1WarningRecheck after the double count is fixed
Google AdsStage past the leadNo import, no enhanced conversions for leadsCriticalTurn on enhanced conversions for leads, then plan imports
BothConsentNo banner, though the company sells in the UKCriticalAdd a banner and Google consent mode before ad tags fire
LinkedIn AdsReconciliation19 conversions against 17 CRM leads, ratio 1.1PassNone
Budget decisions

Google Ads, not yet. Its reported cost per lead is about half the real one, because each demo is counted twice. LinkedIn Ads, not yet, until consent is fixed.

The handoff block the next skill reads
skill: tracking-audit
version: 1.0.0
verdict: FAIL
coverage: complete
for: paid platforms, September 2026
outputs:
  platforms: {google: FIX FIRST, linkedin: FIX FIRST}
  critical: 3
  imports_past_lead: {google: none, linkedin: none}
  stage_dates: yes
next: [utm-check, crm-feedback-loop]
open: []

What good looks like

  • Every platform with budget is graded, largest first
  • No check passed without evidence; gaps say what would close them
  • Platform conversions reconciled with CRM leads, never added together
  • A fix list with the screen, the role and the order
  • A plain safe or not yet for each budget decision

Install

Use it in the tool you already have.

Pick your tool. Each tab has the shortest path we know of.

  1. 01
  2. 02
    In Claude, open Customize, then Skills.
  3. 03
    Choose +, then Upload a skill, and pick the ZIP. Code execution must be on in your settings.
  4. 04
    Ask for it by name, or let Claude use it when a task matches.

Questions

Before you add it.

Does it change anything in our ad accounts?

No. It reads settings, ideally through a read-only connection, and writes a fix list with the screen and the role for each fix. Your team makes the changes.

Why does it compare platform conversions with the CRM?

Each platform counts conversions its own way and credits itself for shared leads. The CRM is the record both sides agree on, so a platform far above it is counting something else.

What if we only have screenshots?

It checks what the screenshots show and marks the click ID check and the reconciliation as not checked. The output says exactly which input would close each gap.

Is it a legal review of our consent setup?

No. It checks the setup against the platforms' documented rules and flags gaps. Legal questions go to your legal adviser, and the output names them.