Once the account skeleton is built and campaigns are running, many teams hit a wall that has nothing to do with bidding, the data. Conversion numbers do not match across reports, GA4 and Google Ads each tell a different story, and the CRM shows a batch of closed deals while Google Ads recorded only half. The subtler problem sits in Smart Bidding. It learns from the conversion signals you feed it; define them wrong and the system will send budget in the wrong direction, however clever it is. So before touching bids, answer one question. Is your conversion data trustworthy.

Structure first, data second, the order this series set. How to build the account was covered last time; this piece does one thing, building the evidence chain from business events to Google Ads conversion goals. The chain has five links, event definitions, data-layer instrumentation, Google tag or GTM, GA4, and the conversion goals in Google Ads. The same principle runs through all of it, define first, wire second, and only then let automated bidding in.

Step one, define conversion events from business outcomes

The first brick of the data foundation is not laid in code; it is laid in business language. Write down what counts as a conversion, and every setting downstream has something to stand on. The tool is an event dictionary. List the business actions one by one, purchases, qualified leads, quote requests, phone calls, sign-ups. Define six things per event, the trigger condition, the uniqueness key for deduplication (order ID, email, or phone), the value (fixed or dynamic), primary or secondary, the source system, and the owner.

The primary-versus-secondary line is the most important column in the dictionary. Google's documentation is direct, primary conversions are used for bidding, and secondary conversions are for observation only; misconfiguring them can prevent Smart Bidding from optimizing effectively. So there is one test, is this action worth letting the system spend budget to optimize. Purchases and qualified leads, the direct outcomes, go primary; add-to-cart and sign-ups, the middle steps, start as secondary and get watched. Setting page views as a primary conversion is the classic beginner contamination, the system learns browsers as buyers.

Step two, instrument once, trigger once

With the dictionary done, the data layer is next. The goal is one unique, stable representation per business event, one name, fixed parameters, firing only when it should. The common trap is triple counting, a page load fires once, a button click fires again, the funnel end fires a third time, and without deduplication, conversion counts double.

In practice, map business events to data-layer events in one table before touching any tool. The table carries at least the event name, the trigger moment, the deduplication key, and the value parameter. That table becomes the shared source of truth for GTM, GA4, and Google Ads; whichever configuration disagrees with it is wrong. The full design of the event taxonomy, including how to treat micro-conversions, gets its own piece in this series.

Step three, wire it up, Google tag versus GTM

At the tool layer, three deployment routes are common. Small sites record conversions directly with the Google tag, simple and direct. Content and marketing sites usually run Google Tag Manager to govern tags centrally, which turns changing tracking into changing configuration. Teams already deep in GA4 can take the route of importing GA4 key events into Google Ads. The routes can coexist, but a team should pick one main route so a single action never gets recorded twice.

With GTM, one conversion action corresponds to one tag and one trigger, with parameters riding along. One default in Google's documentation is easy to miss, when no value is defined, Tag Manager sends the conversion value as 0. E-commerce teams should either pass a dynamic value or knowingly accept the zeros, never let silent zeros flow into reporting.

Step four, GA4, mark key events for attribution

GA4 has no separate conversion concept; important actions are marked as key events, and that mark decides whether they can flow into advertising. Mark every action the event dictionary flagged for measurement as a key event, then return to Google Ads and create conversions from GA4 key events.

Google's documentation highlights a benefit of this route. Conversions created from GA4 key events keep counts consistent between the two products, with the same conversion managed in native interfaces on both sides, which shrinks the reconciliation gap. For teams already heavy on GA4, this route deserves to be the main one.

Step five, finish each conversion action in Google Ads

Once the wiring is done, return to the conversion page in Google Ads and walk each action through its settings panel. The items include the name, the value, the count (every conversion or one per user), the click-through conversion window, the engaged-view window, the view-through window, the attribution setting, and the enhanced conversions toggle. Every item is a business decision, not a fill-in-the-default exercise.

The count and the windows deserve the longest pause. A form a user submits twice should probably count once for lead businesses; purchases in a repurchase business should count every time. Windows should match the sales cycle. Impulse purchases can run short; B2B evaluations run long, and a window that is too short hides the real conversions in the back half from the system. The full configuration reference for these settings continues in the conversion goals piece.

Verify, do not stop at "tag published"

The sign that configuration is complete is not that the tag went live; it is that the data can be verified. Verification runs in three steps.

First, walk a real path. Enter from a test ad click, complete a test order or a test lead submission, and watch the debug tools the whole way, whether events fire, whether parameters carry, whether the deduplication key repeats.

Second, reconcile the triangle. Put the Google Ads conversion count, the GA4 key event count, and the orders or leads in the business system side by side. Differences are normal, time zones, attribution windows, and counting conventions all differ; the goal is that every difference can be explained, not that three numbers are forced to agree.

Third, watch the recording status. In the first days after launch, check the conversion actions' status column regularly, note why each of the states, recording, not recording, delayed, appears, and write down the latency baseline so future reporting holds no surprises.

Consent is the boundary condition of the data foundation. In markets that require user consent, configure tags through the platform's consent mode so they adjust behavior to each visitor's choice. Google's own framing is that consent mode lets tags respect choices while still giving conversion modeling a basis. The principle here is to implement consent properly; no team should treat it as something to route around.

Offline conversions are the other boundary, and B2B and offline-heavy businesses cannot skip it. Save the GCLID produced by ad clicks, and when a deal closes, send the lead back with its GCLID. Google's documentation notes that enhanced conversions for leads, which add hashed first-party data on top of the GCLID, showed a median 10% increase in conversions compared with standard offline imports. Note one 2026 change as well, since June 15, offline conversion imports and enhanced conversions for leads have migrated to the Data Manager API, the legacy Google Ads API path is blocked, and new integrations should start on the new channel.

Common pitfalls

Making micro-conversions primary. Page views and hover events look busy, but what they tell the system is that a visitor came, not that a customer did; once primary conversions fill up with such signals, Smart Bidding aims at the wrong thing.

Shipping without tests and reconciliation. A published tag is the starting gun, not the finish line; a data foundation with no test order and no triangle reconciliation is a building on sand.

Editing numbers when the two sides disagree. Google Ads and GA4 differ by design; the right move is to explain every difference line by line, not to hand-tune one side until it looks good.

Treating consent as an obstacle to route around. Privacy compliance is a hard market boundary; consent mode, data minimization, and lawful basis are part of the data foundation, not technicalities to bypass.