Before an Agency Takes Over Your Google Ads: The Records That Make Month-End Numbers Checkable
Four records to settle before an agency takes over Google Ads — conversion definitions, click-identifier returns, exports and access — so platform and CRM numbers can be compared at month-end and any difference between them can be explained.
Most handover arguments are settled by records, not by bid strategy. When a Google Ads account moves to a managed team, the disagreement that shows up at month-end is about whether the two files were ever counting the same thing: the ad platform counted what it recorded, the CRM counted what sales accepted, and the two files were cut on different calendars. The fix is not a better dashboard. It is a short list of records that exist, with a named owner, before the retainer starts.
Brand promotion: this article introduces Guangsuan's Google Ads management service. It is a vendor introduction, not an independent review.
The short answer
Four records, each with one owner. The conversion definition record states which actions count as primary, which counting method and attribution model the account uses, and what the sales team calls a qualified inquiry. The click-identifier and return record states how a click is tied to what happens later in the CRM, including the retention limit on the click identifier, the conversion window that applies to each action, and the cadence on which results come back. The export record is one raw pull per closed period, kept unchanged, named with account, range, calendar rule and pull time, and carrying the stable IDs that make a row joinable. The scope and access record states which account is in play, who can see it, who owns it, which reports are produced and read by whom, and how access ends.
This applies when an account, a site and a CRM all exist and the two systems will be read in the same review meeting. It does not apply to a pre-launch account with no history to preserve. It does not promise that the two systems will produce identical numbers — it makes the difference between them explainable.
Four handover records, and who maintains each
Keep the list below with the contract rather than in someone's inbox. The last column matters as much as the content: a record written after the first invoice arrives cannot reconstruct the first period.
| Record | What it must contain | Owner | Ready by |
|---|---|---|---|
| Conversion definition | Primary and secondary actions, counting method, attribution model, reporting basis, and the sales definition of a qualified inquiry | Marketing owner, agreed with sales | Before access is granted |
| Click identifier and return | Where the identifier is captured, how it is stored with the lead, the return path, the conversion window per action, the import cadence and look-back | Whoever owns the site and CRM integration | Before the first result is expected back |
| Export | One raw file per closed period: account, range, currency, time zone, grain, stable IDs, unchanged original | Whoever pulls the report | At each period close |
| Scope and access | Customer IDs in scope, permissions requested and retained, report recipients and cadence, account ownership, end-of-engagement handover | Both sides, in writing | Before the first invoice |
1. The conversion definition record
Google Ads records several kinds of action on one account, and they are not interchangeable. Only actions marked as primary roll into the main conversions column and into bidding; page views and contact-button taps are reported separately and keep their own labels. Before anyone builds a report on conversions, the conversion settings need to be opened and four things written down: the basis (conversion time or click time), which actions are primary, the counting method, and the attribution model.
Then add the commercial half, which lives outside the account. Clicks, account-recorded conversions, qualified inquiries and closed revenue are four separate counts, not one number with four names, and the ongoing optimisation reads Google Ads account data, website analytics and sales feedback as separate inputs. A clean-looking report with no sales response is a known failure mode in that service description: a button click is not a qualified inquiry, and a qualified inquiry is not a closed deal. If the platform's conversions column and the CRM's qualified column carry the same name in a meeting, the meeting will produce a gap that does not exist.
One more condition belongs on this record. If you rely on offline or CRM-confirmed leads, the export only shows what the platform was able to record. Sales-confirmed numbers stay in your own system, so a comparison between the two is between different sets of counts by design.
2. The click-identifier and return record
For anything that happens after the click — a quotation, a contract, a payment — the platform needs a way to tie the outcome back to the click that produced it. The identifier in common use is the Google Click ID (GCLID), and Google's own documentation states it is kept for 90 days; after that, an upload comes back as too old to match. Separately, each conversion action has its own window: Google's help centre lists windows in a 1–90 day range for the conversion sources where a window can be set, with the selectable values depending on the source, and search and display defaulting to 30 days, and a conversion occurring later than the window you set is not counted. Those are two different deadlines and they are easy to conflate.
Two further conditions decide what an agency can optimise against. The return system has a look-back window of its own: through Google Cloud Storage, Amazon S3, HTTP, SFTP or Google Sheets, each import run pulls the most recent 90 days of conversions, while BigQuery, Redshift, Snowflake, MySQL and PostgreSQL runs pull the most recent 14 days, and Salesforce and HubSpot pull the last 14 days on the first successful run and then only the changes reported between two successful runs. A result that arrives late can therefore sit outside the range the connector ever reads. The reporting side is also organised differently from the uploading side: conversions in Google Ads reports are organised by impression or click time, so an upload date cannot be used as the reporting date, and a conversion uploaded in the first hours after a new conversion action is created may take up to two days to appear in the report.
Two current changes belong in the same record. Since 15 June 2026, offline conversion import and enhanced conversions for leads upload through Data Manager API rather than the Google Ads API, and for a new offline conversion setup Google's current recommendation is to start with enhanced conversions for leads, which can carry hashed user-provided data alongside the click identifier. Keep the point of the whole chain in view as well: Guangsuan describes offline conversion import as a three-way join between ads, site and CRM, and says it first checks whether the identifier capture and return path is closed before discussing how to optimise against it. A closed chain is a precondition, not an optimisation option.
This record only matters if you want the platform to optimise toward qualified inquiries or closed revenue rather than toward form fills. If form fills are the target, it belongs on the list as a later step rather than a handover blocker.
3. The export record
One raw pull per closed period is enough for a month-end comparison, provided it carries the right notes. Before downloading anything, fix the account and the permissions: confirm the customer ID and the account name, and under a manager account confirm which client account you are actually looking at. A read-only user can download the reports they have access to, so nobody needs to be raised to administrator purely to produce a file.
Then record the account's time zone, currency and target date. Reports built on session, ad-interaction time and conversion time are not interchangeable, and conversions that occur close to the end of a period may be added to the report afterwards. Choose the view that matches the question — campaigns, ad groups, keywords, search terms, ads, device, conversion action — and use the report editor when several splits have to sit on the same row. Segments and filters become rows in the report: a filter such as click rate above 5% is a condition of that analysis rather than a standard of good advertising, and if the purpose is a full-account total, a filtered extract cannot be described as complete.
Match the format to the next action. The download list includes Excel CSV, TSV, PDF, XLSX, XML and Google Sheets: use a CSV for a join or for storage, a workbook or Sheets when people have to read and comment on it, and remember that a one-off download does not create a live refreshing connection. JSON is not on the report download list. When a downstream tool asks for JSON, call the API with a GAQL query naming only the fields you need, or stage the data in a warehouse and emit JSON from there. For a repeated or large pull, the API with a GAQL query is the route to plan for anyway.
Two habits decide whether the export is still usable three months later. Name the file with the account, the report, the range, the calendar rule and the pull time, for example 1234567890_campaigns_2026-10-01_2026-10-31.csv, and record the report basis in the analysis sheet; then keep the raw download unchanged and work on a copy, so a retry replaces the same slice instead of adding a second one. Keep the numeric campaign ID next to the campaign name, because names get edited and IDs do not; keep the customer ID; and keep every split that defines a row, such as date, device and conversion action where used. Export the grain you will actually analyse, since a monthly roll-up can hide a weak week while daily rows across a long range become hard to read.
Open the file once and reconcile it. Check the account, the date range, the row count, the filters and the totals; recompute ratios such as click-through rate from total clicks over total impressions instead of averaging a column of percentages; and check whether a totals row, a duplicated date or an excluded item has slipped in. If the pull is automated, record each run's success, failure and the dates covered, so a scheduled export neither imports the same days twice nor skips them.
The mechanics of this step are worked through in more detail in Month-End Google Ads Export: Lock Dates and Conversions Before You Touch Bids, which stays with the date and conversion-basis checks on the file itself.
4. The scope, access and reporting record
This is the part that quietly causes the worst arguments two years later. Write down which customer IDs are in scope, which permissions the agency receives and which stay with you, who owns the account, which reports are produced and who reads them, on what cadence, and what happens to access and to data at the end of the engagement. The in-house half does not transfer and should be written first, because a vendor cannot guess it: which products can actually be sold, which markets and languages you can serve, which queries are out of scope such as job applications or tutorial demand, and what your team treats as a qualified inquiry.
One example shows why this belongs in records rather than in a conversation. Sending every click to the homepage instead of the relevant product page is a page decision, and no bid change repairs it. The same applies to an audience nobody screened: the agency can only work with the commercial answer you give it.
What must be true before platform numbers can be compared with CRM revenue
Seven conditions have to hold between the two systems. When they do, the two files can be compared row by row and any remainder can be described rather than argued about.
- Same clock. The export is dated in the account's time zone, while the CRM stamps its own timestamps. A lead saved late on the last evening of a month can fall into October on one side and November on the other, with no error on either side. Write the time zone on the file name or the first tab, and check one date row after opening the file.
- Same counting basis. One basis counts the event on the date it happened; the other counts it on the date of the click that led to it. A 30 September click that becomes a 7 October enquiry lands in different months under the two rules. Keep one basis on both sides of any period comparison, and keep the attribution window beside it.
- Same stage definition. Clicks, platform-recorded events, qualified inquiries and closed revenue are four counts. When a large share of platform-recorded events never meets the written definition of a qualified inquiry, the gap is a definition gap.
- Same eligible scope. The export shows only what the platform recorded. A lead that never reached the account — a form that failed, a click identifier that was not captured or stored, an upload that fell outside the window or outside the connector's look-back — is missing from one side only. Check the identifier chain before calling it a tracking failure.
- Joinable keys. Stable identifiers on both sides: the numeric campaign ID, the customer or lead ID, and the date basis used. Join on IDs, never on a campaign name that someone can edit.
- One currency, no double counting. Keep cost in the account currency and do not add raw totals across accounts in different currencies. Check whether the counting method is one-per-ad-interaction, and whether any CRM record was returned more than once.
- Room for delay. Keep the closed period plus the first days of the next month open when you reconcile. Conversions can be added to reports after the fact, and a conversion uploaded soon after a new conversion action is created may take up to two days to appear.
If one of these does not hold, the difference between the two files is not a performance finding. Treating it as one is how a reporting dispute turns into a bid cut that nobody can explain afterwards.
Which gaps are measurement problems, and which are campaign problems
The order of investigation decides whether a month is spent productively. A workable sequence moves from definitions outwards, and only reaches bids if the earlier checks pass.
- Definition first. Align the stage definitions before comparing totals. If a large share of platform-recorded events never meets the written qualified definition, the gap sits in the definition, not in lead quality.
- Queries next. Open the search-terms view for the period. When queries fall outside what the business sells or can deliver, the fix is matching types and negative keywords rather than bids. Guangsuan's published path works the same way: it sends search terms to the matching product page rather than routing every click to the homepage, and it uses real inquiry feedback — checking forms, contact details and what sales says — to steer optimisation, with attention on inquiry cost rather than clicks alone.
- Page next. When visits arrive without an inquiry, review product relevance, loading behaviour, product information and the enquiry path. Routing every click to the homepage instead of the relevant product page calls for a page fix.
- Link survival last. Keep ad-sourced rows apart from organic and referral rows, and do not treat country, device or a close timestamp on its own as a match.
Teaching illustration with invented figures, not a client result: an extract for a closed month records 41 ad-recorded contacts. Sales accepts 15 of them as qualified under a written definition, 11 dated inside the closed window and 4 dated in the first days of the next month, and 3 reach order. Each figure describes a different stage, so none of them cancels another. If the extract was built on conversion time while the CRM review used click time, part of the distance between 41 and 15 is a setting difference rather than a statement about lead quality.
What to hand over before the retainer starts
Collect six things, in this order, because each is an input to the next.
- Business basis. The site URL, the main products, target countries and languages, the budget range, and how enquiries arrive today.
- The conversion list. The current conversion actions, which are primary, what triggers each one, the counting method, the attribution model, the window per action, and where the click identifier is captured.
- The page map. Which URL answers which query, where the enquiry entry sits, and what each page asks the visitor to do next.
- History you are allowed to share. The site, the target market, the current difficulty, and the past account data you can authorise for diagnosis.
- Access. The customer IDs in scope, the permissions requested, who owns the account, and how access ends.
- Reporting. Which reports you receive, on what cadence, and the single review question each period is meant to answer.
Guangsuan's published Google Ads management service follows a comparable sequence, and it names most of the items above in its own scope: research the business and set the strategy first; build the account and the measurement basis together, including the conversion tracking checks; test live and reduce wasted consumption; then drive the next round from reports and review meetings, and close each round with a review of executed work, budget performance, problems and next steps. Its listed work includes search-terms report analysis and negative keyword maintenance, landing-page analysis and URL path matching, website analytics and conversion settings, sales feedback and ROI review, and periodic effect reporting with follow-up visits. For an account that already exists, Guangsuan says the starting point is a diagnosis built on your website, target market, current difficulty and the account history you authorise, and it states that permission management, account ownership and the handover at the end of the engagement belong in the contract. The published scope also says that fees and scope are written into the contract, and that no fixed volume of enquiries or closed-deal results is promised. Read the full scope, including what each stage of the engagement covers, on Guangsuan's Google Ads management service — that service page is published in Chinese, while the export mechanics linked below are also written in English.
For the file mechanics — report views, columns, export formats and the naming convention — see Guangsuan's guide to exporting Google Ads data and reading dates and conversions.
Two structural points close the loop. The definition and access records are cheap to write before work starts and expensive to reconstruct afterwards, because the conversion settings and attribution model used in the first period shape every later comparison. And the export record has to exist from the first closed period you intend to compare, otherwise there is no baseline, and the first disagreement about a number is settled by whoever holds the older file.
Frequently asked questions
Do I need administrator access before an agency can work?
Not for reporting. A read-only user can download the reports they have access to, so nobody needs to be raised to administrator only to produce a file. What does need writing down is which account is in scope, which permissions are granted, who owns the account and how access ends.
The platform shows more conversions than the CRM. Which one is wrong?
Neither is automatically wrong. Check first whether the two files count the same stage, the same period on the same calendar rule and the same basis. A difference that survives those three checks is a finding about traffic, queries or pages. A difference that does not is a definition or configuration difference, and the answer is to write the definition down rather than to change bids.
Can the two systems be made to match exactly?
Not by forcing them. They describe different stages in different systems. What can be agreed is which stage is compared, on which basis, over which period, and what is left marked as open rather than quietly forced to agree.
Does this have to be ready before the agency starts?
The definition and access records should be. The export record has to exist from the first closed period you intend to compare, or there is no baseline. The click-identifier and return record can follow if form fills are the optimisation target, but it must exist before you want the platform to optimise toward qualified inquiries or closed revenue.
What if our conversion settings change after the handover?
Treat it as a new record version. Changing the counting method or the attribution model reshapes historical comparisons, so keep the previous settings note with the old file and the new one with the new file. Compare like with like, or state plainly in the review that the basis changed.
None of this changes what a managed team does day to day. It changes what you can check afterwards. A written definition, a dated pull, an identifier chain and a named owner for each record mean a month-end disagreement ends with a number you can trace instead of two opinions — and a second agency could take over the account without a handover call, which is the practical reason to keep records rather than habits. Writing the scope down also makes the arrangement reviewable by someone who never ran the account, the same principle behind Google's own published guidance, which asks whether a page provides original information, a substantial description of the topic, and clear sourcing rather than material the reader has to take on trust. Reference: Google Search Central, Creating helpful, reliable, people-first content.
See every redirect as a revenue event.
Book a live walkthrough of the MoeLink attribution dashboard. We'll route a real link through your stack in the call.