SALESTUDIA · DATA CONNECTIONS · UPDATED: 9 SEPTEMBER 2026
1. Google Ads Data Manager starts with a clear data brief
Google Ads Data Manager connects supported data sources to specific uses in Google Ads. For an SME, the key question is therefore: which approved data should reach which audience list or conversion action, and through which connection? Setup becomes worthwhile once the business has defined that relationship. A green connection status alone confirms neither the quality of the source records nor an attributed conversion or additional customers.
A connection plan makes the brief verifiable
The official overview of Google Ads Data Manager describes how to connect and activate your own customer data. In practice, this establishes a clear boundary: the tool can support a prepared data flow. It cannot decide which customer status is correct for your business, which order a revenue figure belongs to, or who should correct inaccurate master data.
This guide helps you build a practical working document: the Data Manager connection plan. It brings together the source, purpose, record selection, field mapping, destination, import evidence and responsible person. You can maintain it in a spreadsheet or your existing project system. What matters is that the same connection ID appears in the setup record, error log and handover.
Start with one data flow whose business purpose is clear. For example, approved CRM orders that your business has actually won should feed a specified conversion action. Write down which event each row represents and when that event occurred. If sales and marketing answer those two questions differently, the connection is not yet ready for a production import.
2. Always check the source and destination together
Support for a product does not mean support for every data flow
The supported data sources table distinguishes between the available uses. HubSpot is listed for Customer Match and conversion imports, while Salesforce is listed for conversion imports. The manually configured direct Shopify connection with selected customer lists serves Customer Match. Shopify also has a separate purchase measurement integration, which we examine later. For files and databases, too, check the source and destination combination actually offered.
Avoid recording only “CRM connected”. A useful description would be: “Production system, defined set of orders, completed events, conversion action in the advertiser account”. An audience list needs a separate business brief: who should be a member, why should they be included, and how should membership be updated?
Suitable when the available connector supports the required use and fields. Check the specific object and the available selection rules.
Useful when your data needs to be combined according to business rules before import. Your team then also takes responsibility for the export, its refresh schedule and a stable schema.
Greater technical complexity does not make a connection better. What matters is traceable data and a sustainable operating process. If a direct connector lacks required information, document the gap first. An additional export may close it, but it can also create a second, conflicting version of your data. Your chosen architecture must identify which system remains authoritative when values differ.
3. Establish access, data ownership and approval
Assign four responsibilities before the first run
Someone must own the source values, someone the connection, someone the Google Ads destinations and someone the approval for data use. In a small team, one person may hold several responsibilities. Still, record each responsibility separately. Otherwise, an incorrect order stage may end up with the administrator, even though only sales can resolve the meaning of that field.
For each source, record the production account, the required access and a backup contact. Keep credentials in your designated secure credential system; a reference to the relevant access arrangement is enough in the connection plan. Check in advance whether a staff change, a changed password or revoked permissions could interrupt the data flow.
You also need a precise business definition of the records to be transferred. “All contacts” merely describes a stored collection. “People with documented approval for this use and the approved customer status” describes a selection rule. That rule does not automatically complete the checks on consent and permitted use. Your own process must never silently treat an empty or unknown approval field as agreement.
If your stages are still maintained inconsistently, start by structuring CRM data and sales processes clearly. The connection plan then carries forward the agreed definitions. It should not reinvent them for every integration. Ask the business owner to use a few real, internally checked cases to demonstrate why one record belongs in the selection and another must be excluded.
4. Create your Data Manager connection plan
Each row describes a fully traceable data brief
The following template illustrates the structure using an explicitly fictional company: Rheinwerk Bürobedarf sells standard products through Shopify and manages bespoke business orders in HubSpot. The IDs shown are internal documentation fields, not mandatory Google field names. Give every production connection a permanent ID and add a version number when its configuration changes.
| ID and source | Purpose and selection | Field specification | Destination | Run evidence | Ownership |
|---|---|---|---|---|---|
| DM-01 · HubSpot | Business orders won and approved for transfer; specified Lifecycle Stage plus an additional condition | Version 1: event time, permitted identifiers, agreed value | Specified conversion action in the advertiser account | Source snapshot, run ID, result and unresolved errors | Sales: content; marketing: destination; technical team: operation |
| DM-02 · Shopify customer lists | Deliberately selected list approved for Customer Match | Version 1: available contact fields and selection rule | Named Customer Match list | List snapshot and import run; verify list use separately | Store owner and Ads owner |
| DM-03 · Shopify purchase measurement | Separate purchase integration through the Google & YouTube app | Document the mapping of Checkout completed | Purchase conversion mapped in the app | App setting and evidence from measurement checks | Store technical team: event; marketing: mapping |
| DM-04 · Approved export | Alternative to DM-01 for a documented connector gap | Same business event, separate technical schema | Supported import route checked in advance | Export version, replacement decision, comparison run | Data owner decides whether to replace the existing route |
DM-04 is a planned alternative, not a suggestion to import the same data through two routes in parallel. The register explicitly states whether a route is being prepared, is in production, is paused or has been replaced. Store a reference to the field specification and run log for each ID. A row with several destinations needs separate evidence for each, so a functioning part cannot conceal a failing one.
The benefit becomes clear at the first problem. Instead of “The CRM connection is not working”, the report reads “DM-01, version 2: the new source column was not available as agreed in Tuesday’s run”. The owner, affected scope and starting point for investigation are immediately apparent.
5. Choose the right CRM object and event
A contact, a status change and an order are different records
Using HubSpot as a Data Manager source involves specific filtering rules for conversion imports: the selection must use Lifecycle Stage, with additional conditions joined to it using AND. Google describes a 14-day lookback for the first successful conversion run, followed by reported changes since the last successful run. A newly configured connection therefore does not provide a freely configurable import of the entire CRM history.
The Salesforce connection handles Lead, Opportunity and Order through separate object connections. Google requires the appropriate API permissions and necessary field history tracking; sandbox connections are not supported. Check these conditions with your Salesforce owner before scheduling a business acceptance test.
At Rheinwerk, the existence of a contact does not mean a business order has been won. The team therefore records which status change triggers the event and which timestamp belongs to it. Substituting the date of the latest general contact update would be problematic: correcting a phone number must not turn an old order into a new sale.
Check three internally selected cases: the first time a record reaches the target status, a later change to master data only, and a sale that is subsequently withdrawn. Describe the required behaviour for each. If the available connector cannot represent that behaviour using the existing fields, mark the connection as unresolved. Renaming the conversion action does not solve the data problem.
6. Separate Shopify customer lists from purchase measurement
Two data flows need two entries in the connection plan
The Shopify documentation for Data Manager describes two different routes. The manual direct connection lets you select customer lists for Customer Match. Check the initial selection, because all available lists may be selected by default. This workflow is not a general order import.
Google also describes an automatic purchase integration through the Google & YouTube app: the Checkout completed event sends purchase data to Google Ads from the server, with corresponding tag events deduplicated. Changing the conversion mapping can take up to twelve hours. Disabling the purchase event in the app affects both tag and server-side purchase measurement. This documentation does not establish support for other Shopify events.
Check activation in the specific store: configured app measurement, updated tags, acceptance of the required terms and the corresponding notification. An existing account connection alone does not confirm that these conditions are met.
For Rheinwerk, this means DM-02 documents the selected customer list and its purpose. DM-03 separately documents the purchase conversion, app mapping and time of the check. “Shopify is connected” would be too vague for either task. Even a complete customer list does not prove that the current purchase measurement is mapped correctly.
Before feeding in additional order data from an export, record the purchase data flows already in place. Establish the event identity, destination action and expected behaviour on retries. Do not assume that the documented deduplication for the app integration also applies to arbitrary imports of your own.
When troubleshooting, change only the affected setting. Record the mapped conversion action before and after the change. Reconfiguring the app, introducing an export and changing the destination definition at the same time creates several possible causes of error and makes reliable acceptance testing harder.
Product data for Merchant Center is another separate area of work. The checklist for preparing Shopify for Google Shopping covers that store preparation. Successful product data delivery confirms neither your customer lists nor the processing of an individual purchase event.
7. Define the destination before mapping fields
Available data and permitted account use are separate questions
A list identifies the people who belong to a defined group. A conversion action identifies the event being measured. The same CRM record may contain business information relevant to both purposes. Selection rules, evidence and destinations still need separate checks. Adding a customer contact to a list does not, by itself, turn that contact into a conversion event.
Record the exact name of every destination and the relevant Google Ads account. In particular, distinguish the manager account from the account in which the use is configured. For a conversion, add its business meaning: a completed sale, a qualified enquiry or another agreed event. “New lead” remains ambiguous after connection if it has no definition.
Whether an account may use Customer Match to the required extent needs an additional check. An available connector, a list that can be imported and sufficient technical access do not replace that check. Document the use available in the account and the applicable conditions. If the intended use is unavailable, changing the filename will not solve the problem.
For conversion actions, you must also establish how they fit into the goal configuration. Document which actions are approved for primary or secondary use and which campaign goals are affected. The connection plan can then refer to that decision. Successfully transferring data must not inadvertently substitute for a separate business decision about optimisation.
8. Write a field specification with concrete examples
Similar column names are not enough to justify a mapping
The data preparation requirements distinguish between destination schemas. Event time is essential for conversion imports; the identifiers available depend on the measurement method. Ambiguous dates can be misinterpreted. If a timezone is missing, the intended processing must supply it unambiguously. This table is an internal field specification for Rheinwerk’s prepared conversion export, not a universal upload template.
| Source field | Example | Meaning | Use | Check | If incorrect |
|---|---|---|---|---|---|
| abschluss_zeit | 2026-09-08T14:30:00+02:00 | Actual time of the agreed completed sale | Conversion date and time | Compare with the CRM event | Do not invent a replacement time |
| google_click_id | Existing, unchanged GCLID | Stored click identifier | GCLID, where supported for the chosen route | Origin and unchanged value | Mark as missing; never generate one |
| kontakt_email | Internally approved source value | Permitted contact identifier | Email Address for a suitable method | Association and approval | Have the record reviewed |
| auftragswert | 1250.00 | Consistently defined value | Conversion value, if used | Confirm the unit and value definition | Do not silently replace it with zero |
| export_freigegeben | true | Internal approval for this data brief | Selection rule, not a measurement event | Exclude unknown values | Resolve with the business owner |
| quellreferenz | RW-2026-081 | Internal traceability | Audit log; no automatic destination mapping | Preserve a stable reference | Do not approve the import batch |
Add the explicitly agreed currency to the value specification and check how the chosen destination processes it. Avoid mixing amounts without their units. Technically converting a text field into a number does not establish whether the amount is gross, net, estimated or actually booked.
For the separate measurement setup and handling of user data, the plan refers to enhanced conversions for web and lead data. Here, the focus is the agreed transfer specification: which source value is sent where, and for what reason? Have the mapping reviewed by someone who understands its business meaning, rather than only the visible field names.
9. Make filters and transformations traceable
A technically valid transformation can be wrong for the business
Write down each filter condition in plain language and compare the selected records with the source. For Rheinwerk, one rule might be: the agreed closing status has been reached and the export is approved for this purpose. Also check a case deliberately excluded from the selection. Reviewing only examples that should pass often leaves an overly broad selection undetected.
The available transformation actions include type conversions, multiplication, and date and time processing. Multiple actions run from top to bottom, and not every source offers every action. A failed numeric conversion may appear as an error in the run. Record the order and expected result in the field specification.
A practical test needs only a few carefully chosen cases: a normal amount, an empty value and an amount in a different decimal format. For timestamps, use one value with an explicit timezone and one without. Define the expected results beforehand. A preview provides evidence only when it agrees with those expectations.
Stop setup if a transformation would have to supply missing business information. A conversion may change a known unit. It must not create the time of a sale, undocumented approval or invented revenue.
Save the filter and transformation together as a version of the connection plan. A later filter change alters the records sent even when all field mappings stay the same. To compare results, you must therefore be able to see which selection was active at which time.
10. Set up the connection in controlled steps
Compare the summary with the connection plan
Google requires administrator access in Google Ads for setup. The documented connection workflow starts in Tools and Data Manager, then moves through the source and intended use to record selection, field mapping, optional transformations and a final review. The specific selection steps depend on the connector. First check that you are working in the correct account.
- Open the brief: Have the approved connection plan entry beside you. Check its ID, source system, object and destination.
- Connect the source: Choose the intended product and the available use. Authorise the designated access.
- Select the records: Check the table, object or list. Apply only the agreed selection and its associated filters.
- Map the fields: Compare every relevant source field with the specification. Check transformations against your prepared examples.
- Review the summary: Compare the name, schedule, record selection and destination. Save this configuration in your own log.
- Review the first run: Monitor the result and document discrepancies before marking the connection as live.
The first run transfers real data. Use only data approved for that purpose. The fictional IDs in this article illustrate documentation; they are not artificial customer or conversion records to upload. If you cannot define a permitted test batch, first agree on an appropriate verification method with the data owner.
Clicking the final button completes setup, but it does not complete acceptance testing. Record the next review date and the responsible person in the same entry. This keeps the connection verifiable even if the run is processed later.
11. Align source refreshes with the import schedule
The right sequence matters more than an early start
The options for managing existing connections include scheduling and reviewing runs and uses, depending on the configuration. The most frequent schedule for these connector imports is daily. Check the timing of automatic Shopify purchase measurement and an API integration separately. The source, filename or table reference must remain stable. Before changing shared credentials, check which other connections depend on them.
Rheinwerk agrees on an internal sequence: sales finishes its daily CRM updates, the approved export is then made available, and the import follows. The exact times are operational decisions. An import that starts before the export is ready can read the same old data again without any technical error. Record source freshness and run time separately.
The preparation requirements linked in section 8 specify lookback periods by source: conversion imports from GCS, Amazon S3, HTTP, SFTP and Google Sheets use 90 days for each run; BigQuery, Redshift, Snowflake, MySQL and PostgreSQL use 14 days for each run. The CRM rule in section 5 works differently. These extraction windows are neither attribution windows nor audience membership durations.
Schedule a change so that complete source snapshots from before and after it remain comparable. Treat a renamed file or replaced object as a technical change requiring another check. After a failed run, first document the last successful source snapshot. This establishes which data may be missing and the conditions a retry must meet.
12. Keep evidence for every relevant run
Transfer, processing and use need separate evidence
The following table is an internal review sheet, not a promise of identical columns in every interface. Collect the available evidence from the source, connection details and destination system. Anything that is not visible or not yet complete remains explicitly unresolved. Avoid combining processes with different denominators into one success rate.
| Review stage | Evidence | What is counted | Time reference | Supported conclusion | Next check |
|---|---|---|---|---|---|
| Source ready | Approved source snapshot | Rows before selection | Export or object snapshot | Planned data is present | Compare the filtered selection |
| Records selected | Internal selection check | Rows meeting the conditions | At the recorded source snapshot | Selection matches the business rules | Monitor ingestion |
| Run processed | Run details and error evidence | Processed records and records flagged for issues | Specific run | Technical status of this stage is known | Assign unresolved cases |
| Usable at the destination | Destination diagnostics or report | Relevant destination metric | Processing status at the destination | Evidence of use at the destination | Assess matching or attribution separately |
| Business result assessed | Reconciliation with CRM and business figures | Agreed event cohort | Appropriate observation period | Result within the defined scope | No unverified causal claim |
An accepted import request does not mean every row has finished processing. Processed customer data does not automatically mean matched users, and matched users do not guarantee attributed conversions. State what each number counts alongside it. Without that definition, the number must not support an acceptance decision.
For each run, retain at least the connection ID, configuration version, source snapshot, visible result, remaining errors and person responsible for resolving them. This lets you later distinguish a decline caused by fewer source events from one caused by a changed selection or technical error. A screenshot without a time reference and configuration version is often insufficient.
13. Classify errors before importing again
A retry makes sense only after investigating the cause
The Data Manager diagnostics distinguish, among other states, “Needs attention” when imports continue and “Urgent” for critical problems that prevent import. These diagnostics are being rolled out gradually and may not yet appear in your account. Typical checks cover credentials, source format, availability and the field history tracking required for Salesforce.
First determine whether the error affects the entire connection or only certain records. For an access issue, the owner needs the affected connection and the last successful run. For a format error, they also need a sample row that may be shared internally and the expected format. Do not attach complete customer files to routine error reports when a technical description is sufficient.
Before retrying, record which data has already been processed. Check the documented behaviour of this route for retries and event identifiers. A new import file with another name or changed event times is not a neutral retry. It can change which business events the data represents and make subsequent investigation harder.
Set this internal stop rule: no repeat import while the previous result is unknown. First establish the status, fix the cause and define the retry scope. If the responsible team member is unavailable, keep the issue open with an owner and a next review date. Starting a run manually without a documented reason is not a reliable fix.
14. Reconcile a complete example in your own register
The example separates data quality from advertising impact
The following figures are entirely fictional and are not benchmarks. Rheinwerk’s prepared export contains 120 records. Twelve lack sufficient internal approval for this data brief, and another 18 do not meet the agreed event definition. The approved selection therefore contains 90 records. The source owner confirms this calculation before import.
During the internal format check, four of the 90 records are excluded because their event times are unresolved. That leaves 86 in the first approved batch. The four cases receive internal references, an error reason and an owner. The team does not add estimated times merely to make the file look complete.
After import, marketing copies the processing result actually shown into the run log. This technical stage is complete only when that evidence verifiably confirms the checked batch. The calculation 120 minus 12 minus 18 minus 4 remains an internal record reconciliation. It does not claim that Google displays exactly these review columns or uses the same error classification.
Matching, attribution and advertising results then require a new review with their own time references. The team must not infer 86 Google Ads conversions or 86 additional customers from 86 technically approved records. If later reports show different figures, compare the definitions, cohort and processing status. Do not retroactively alter source data simply to make it match an advertising report.
15. Use an API only when technical operations are covered
Automation also expands responsibility
The Data Manager API is a separate programmatic integration route for supported audience and conversion use cases in Google products. Google positions it for organisations with technical resources or a suitable data partner. Configuring a connector in the interface and operating a custom API integration are different tasks.
Choose the API when a concrete need justifies the additional responsibilities: for example, an existing data processing operation with clear ownership and a required, supported destination. Check the current capabilities and access requirements for that exact use case first. A general product description does not guarantee every combination of event, destination and account.
Extend the connection plan to include technical run references, responsibility for reviewing processing responses, retry rules and monitoring. Acceptance of a request alone is not sufficient for sign-off. The technical team must be able to explain how it identifies final results and partial failures, and which data a retry would affect.
If only one external person understands the transfer process and nobody internally can recognise a failed run, the operational handover is incomplete. Agree on access to the necessary evidence and a clear procedure for handling interruptions.
16. Sign off each connection against defined criteria
The map view complements the plan
The Data Manager map view visualises supported sources and data flows. According to the documentation, it currently omits audiences, campaigns and conversion status, as well as some data sources. A missing entry is therefore insufficient evidence that a data flow does not exist. Also consult the relevant detail pages and your own connection plan.
| Criterion | Required evidence | Owner | Reason to stop | Review again |
|---|---|---|---|---|
| Source and purpose | Approved connection plan | Business owner | Unclear event definition | After business clarification |
| Fields and selection | Versioned specification and checked cases | Source owner and technical team | Unknown timezone or record scope | After correcting the data |
| Destination and parallel routes | Account, destination ID and inventory of existing routes | Ads owner | Unresolved duplicate transfer | After deciding the data route |
| Operation and result | Run evidence, unresolved cases and backup cover | Technical owner | No verifiable processing result | Once evidence is complete |
Mark each criterion as “supported by evidence”, “unresolved” or “not applicable, with a reason”. A successful source test cannot compensate for an unresolved destination mapping. Record the sign-off date and the exact configuration version. A later change to selection, fields or destination triggers the relevant checks again.
17. Frequently asked questions about Google Ads Data Manager
Does Data Manager replace our CRM?
No. The CRM remains the system for your agreed customer and sales information. The connection plan specifies which approved parts are shared for a particular purpose. Unclear customer stages or conflicting order values must be resolved by the responsible business owner.
Can I use every supported source for Customer Match and conversions?
No. Check the specific source and destination combination and the available fields. A product appearing in the selection does not confirm every use case. Also document the use permitted in the account. Data availability and permission to use it are separate requirements.
Is the Shopify connection only for customer lists?
The manual direct connection with selected lists serves Customer Match. The documentation also describes the separate purchase integration through the Google & YouTube app. Maintain separate records for both routes and check their activation and destinations in the specific store.
Should I transfer all available data just in case?
Start with a clearly defined and approved data brief. Every additional set of records needs a clear purpose, suitable selection rules and an owner. A larger export does not make an unclear event definition more reliable and creates more work when errors occur.
Why is data missing despite a successful run?
Check the source snapshot first, then filters, lookback period, field processing and destination. A successful intermediate stage does not answer all of these questions. Compare only figures with the same definition and time period. Unknown processing results remain unresolved until suitable evidence is available.
Can I simply restart an import that failed?
First establish which data has already been processed and why the error occurred. Define the retry scope using the documented behaviour of the chosen route. Do not change event times or identifiers merely to force processing again.
Does a small business need the Data Manager API immediately?
Only when the specific data brief requires a programmatic route and its operation is covered. An available connector that fits the business requirements may be enough. An API also needs firm arrangements for processing evidence, partial failures, retries and technical backup cover.
Does a working connection prove that Google Ads generates more revenue?
No. It initially verifies only the data flow that was checked. Matching, attribution and additional business value require separate analysis. Keep technical run metrics separate from business results and state explicitly which conclusion the available evidence actually supports.
18. Approve data flows only after a verifiable handover
Your next step is a connection plan backed by complete evidence
Choose a source and one clearly defined purpose. Check the supported data route, selected records and every relevant field. Align source refreshes with the import and keep run evidence through to a verified processing result. This produces a connection your team can understand, check and update deliberately when something changes.
The key handover question is: can a backup owner explain which data most recently went where, and which cases remain unresolved? When the account, destination, configuration version and evidence can be found, the operation is traceable. If any of these elements is missing, add it before expanding the setup.
Want a structured review of your data connections? SaleStudia helps you review sources, conversion goals and existing data flows together, then develop a practical plan for verification and setup.
Get a professional review of Google Ads and your data connections