Why inherited conversion actions mislead bidding
Google documents conversion tracking as the way an account learns what happened after an ad interaction: a purchase, a sign-up, a call or an app install. The measurement hub describes the unit, the conversion actionA Google Ads definition of one action to count, such as a purchase or lead, with its source, category, counting method, value and conversion window., with its source, category, counting method, value, window, attribution model and a primary or secondary flag. Primary actions fill the Conversions column and are what bidding optimises towards; secondary actions appear in All conversions and are observed but do not drive bids. Standard goals group actions by category, and custom or campaign-specific goals let a campaign use a different set.
Each layer of history adds actions. A previous agency installs its own tags; a developer hard-codes one on a page that later moves; someone links GA4Google's event-based analytics platform. and imports its key events alongside the native tag; a test action is created and never removed. Three failures follow. Double counting, when two actions record the same event and both are primary. The wrong lesson, when a page view or a click on a button is primary and the strategy learns to buy cheap readers. And false precision, when a cost per acquisition target is met only because the Conversions column is inflated.
The Smart Bidding strategies guide states the consequence plainly: the model treats every included action as equally worth chasing unless values distinguish them. The account's own audit tutorial flags these problems; this tutorial fixes them. When the same results also steer Meta or another platform, the guide to conversion events and values across ad platforms maps every platform's events to one written definition of each result before any action is made primary.
Before you start
- Admin access to the Google Ads account, and read access to the tag container, the site's GA4 property and the account's change history, so each action can be traced to where it fires.
- A closed reporting period, such as last month, old enough for its conversion windows to have closed, with conversions per action from both the Conversions and All conversions columns.
- The business's own figures for the same period: orders from the shop's system, leads and booked jobs from the CRM, logged calls. Without them, you cannot tell a duplicate from a real conversion.
- A list of the outcomes the business is paid for and what each is worth. The measurement plan framework is a good shape for it.
- A quiet window for the change: not the week of a sale, a launch or a bid strategy switch.
- Agreement from whoever reads the reports that reported conversions will fall and targets will be reset.
The steps
Export every action with its settings
From the conversions summary, list every conversion action, including removed and hidden ones if the view allows, into a sheet that will become the register. For each, record the name, source (website, app, phone call, import), category, status, counting setting, value setting, the click-through, engaged-view and view-through windows, attribution model, primary or secondary flag and the goals it belongs to. Add the conversions it recorded in the closed period in both columns. Then find where each one fires: the tag in Tag Manager, the page, the GA4 event behind an import. An action whose source cannot be found is the first candidate for suspicion.
Map each action to a business outcome
Beside each action, write the outcome it represents: a purchase, a qualified lead, an enquiry, a call, a micro-step such as adding to basket, or nothing at all. Several actions may map to the same outcome, which is the signal for the next step. Actions that map to nothing, such as a page view promoted to a goal years ago, are candidates for secondary status or removal. Keep the mapping in business language, because it is what the people who read reports will check: "booked collection" rather than the tag's internal name.
Find duplicates and overlapping sources
Group the actions by outcome and compare their counts for the closed period. Two actions with near-identical counts for the same outcome usually record the same event twice, most often a native tag and a GA4 import, as the comparison of Google Ads and GA4 figures describes, or an old hard-coded tag beside a Tag Manager one. Check counts against the business's records too: an action recording more purchases than the shop took orders points at a tag firing on reloads or on a button click rather than on success. For purchases, a transaction ID passed with each order lets Google Ads ignore a repeat of the same order, which the setup tutorial uses to stop a reloaded confirmation page counting twice. The deduplication entry covers the general problem.
Decide which actions are primary
For each kind of campaign, choose the one or two actions that represent the outcome it exists to produce, and mark the rest secondary. A shop's purchase action, a service firm's qualified enquiry and a call above a minimum length are typical primaries. Where a native tag and a GA4 import record the same event, keep one as primary, normally the native tag, and the other as secondary for comparison, as the setup tutorial advises. Micro-steps stay secondary: they are useful for diagnosis and harmful as a bidding signal, because a Key EventA GA4 event marked as important to the business, the property's measure of success. that happens often and cheaply is exactly what a conversion-maximising strategy will buy more of.
Set counting for each action
Counting decides whether one ad interaction can produce several conversions of the same action. The setup tutorial's rule holds: one per interaction for leads, sign-ups and enquiries, because a person who submits twice is still one lead, and every conversion for purchases, because each order matters. Inherited accounts often have every set on a lead action, which turns repeat submissions into extra conversions and, as the CPA in Google Ads entry shows, changes the denominator of cost per conversion. Correct counting on secondary actions as well, so the comparison columns mean the same thing.
Set values on primary actions
A value-based strategy needs values on its primary actions, and even a count-based one benefits when reports show worth as well as number. For purchases, pass the transaction value dynamically from the site. For leads, set a fixed value that reflects relative worth, so that an enquiry for a large job and a newsletter sign-up are not equal. The conversion value entry covers the three ways to set a value and why a value that ignores margin or lead quality teaches bidding to chase the wrong customers. Leave secondary actions at a nominal value or none, so they do not distort the value columns.
Check windows and attribution per action
Each action carries its own conversion windows and attribution model, and inherited accounts often have a mix: one action on data-driven attribution with a long window, its duplicate on last click with a short one. Bring the primary actions for the same kind of campaign onto consistent settings, so their figures can be compared and summed. How to choose the model and set windows from the account's own conversion lag is the subject of the conversion windows and attribution guide; here, record what each action uses, note any that look wrong and change them in the staged step below rather than one by one.
Align goals with campaigns
Actions reach bidding through goals. Standard goals follow the category, so every purchase action rolls into the account's purchase goal, and campaigns use the account default unless given their own. Check each campaign: a lead campaign should not be optimising towards a purchase goal it cannot influence, and a campaign built for one job, such as calls, may need a campaign-specific goal. Where a custom goal combines actions, confirm each member is one you marked primary for that purpose. Write the goal for every campaign in the register.
Stage the changes
Google's Smart Bidding documentation lists changing conversion actions among the changes that can start a learning period. Making the clean-up in dribs over several weeks puts the account through learning again and again. Prepare every change in the register first, then apply the ones that affect primary actions and goals together, campaign group by campaign group, in the quiet window. Changes to secondary actions can go first, because they do not reach bidding. Do not switch bid strategies in the same week; the results dropped after a bid strategy change diagnosis shows why two simultaneous changes make the result unreadable. Reset targets against the cleaned figures and note the date and reason in the change history. Later changes to a single action, after the clean-up, can be gated with the conversion goal change checklist.
Record the action register and compare with the business
Finish the register: one row per action with its outcome, source, primary or secondary status, counting, value, windows, model, goals and the reason for each decision. Then, after the windows for the first cleaned period have closed, compare the primary conversions with the business's records for the same days. They will not match exactly, but they should be the same order of magnitude and move together; the tutorial on reconciling with sales records turns that comparison into a monthly routine. The conversion rate calculator gives a quick check that the cleaned conversion rate is plausible for the traffic.
A cleaned account can be checked against a short list. The tracking verification checklist covers whether each tag fires correctly, and the Google Ads account audit scorecard scores the result alongside the rest of the account.
| Check | A clean account | A cluttered account |
|---|---|---|
| Primary actions per kind of campaign | One or two, each a business outcome | Many, including page views and clicks |
| Duplicates | One primary action per event; any import is secondary | Native tag and import both primary |
| Counting | One for leads, every for purchases | Every on leads |
| Values | On primary actions, reflecting worth | Missing, or equal on unequal actions |
| Windows and model | Consistent across primary actions of one kind | A different mix on each |
| Goals | Chosen per campaign and recorded | Account defaults nobody has looked at |
| Comparison with the business | Same order of magnitude, moving together | Conversions far above orders or leads |
An action register before and after (illustrative)
Lintwhistle Laundry clears out five primary actions
Lintwhistle Laundry, an invented laundry and dry-cleaning service with collection and delivery, inherits an account whose Conversions column shows 1,480 conversions a month against £3,600 of spend, a cost per conversion of £2.43. Its booking system records 365 paid collections in the same closed month.
| Action | Source | Before | After |
|---|---|---|---|
| Collection booked | Tag Manager tag | Primary, every, 410 (fired on reloads) | Primary, every, booking ID passed, 348 |
| Collection booked (GA4) | GA4 key event import | Primary, every, 398 | Secondary, kept for comparison |
| Booking complete (old) | Hard-coded tag on a retired page | Primary, every, 162 | Removed after one month at zero |
| Contact form | Tag Manager, button-click trigger | Primary, every, 240 | Secondary, one, trigger moved to success, 96 |
| Price list view | Page view | Primary, every, 270 | Secondary |
The before column adds up to 1,480 primary conversions, roughly four for every real booking. After the staged change the Conversions column shows 348 bookings against 365 in the booking system, a gap the owners put down to consent refusals and bookings made by phone, and the cost per conversion is £10.34. The target CPA of £4, which the campaigns had appeared to beat, is replaced with £12 from the value of a new customer, and the Search campaigns stop buying price-list readers. The register now explains every action, and the next comparison with the booking system is booked for the following month.
Common mistakes
- Cleaning up without the business's figures. A duplicate looks like a real conversion until it is compared with orders or leads.
- Deleting before demoting. Secondary first, removal only after the action is shown to be redundant.
- Leaving a GA4 import primary beside the native tag. The same sale counted twice.
- Fixing one action a week. Each change to primary actions can restart learning; batch them.
- Changing the bid strategy in the same week. The effect of the clean-up cannot be separated from it.
- Keeping old targets. Targets met on inflated counts are wrong once the counts are honest.
- Unequal actions with equal values. Bidding then chases whichever is cheapest.
- No register. The next person inherits the same mystery.
