Sales were falling. The problem was not the campaign.

The online store’s orders started to fall and, for several weeks, stayed close to zero. Over the same period ad spend had not gone down: it had gone up.

In cases like this the first reaction is to act on the campaign: change the ads, cut the budget, try another campaign type. Here that would have been a mistake.

Before touching the campaign I checked the data the campaign based its decisions on. In Google Ads, recorded purchases had almost disappeared for weeks. Not because nobody was buying: the tag that reports each sale to Google had stopped.

For automated bidding this is the worst condition. Without recorded sales it does not know which clicks bring orders and keeps spending blind. The more budget it gets, the more traffic it buys without being able to assess it.

The account history showed three more problems:

• over time several purchase conversion actions had been created and removed, and none of them had settled;
• after a first fix, two actions counted the same orders: Google saw far more purchases than the real ones;
• almost all the automatic changes proposed by Google were active. One of them lowered the return target every time the recorded return fell, that is, precisely because of the broken data.

Five corrections, all on measurement:

1. a new purchase tag, verified order by order;
2. removal of the action that counted the same purchases twice;
3. automated bidding realigned on purchases only, with no secondary signals;
4. the automatic changes that acted on the wrong data switched off;
5. server-side tracking: orders reach Google directly from the store’s back office, no longer from the visitor’s browser, and the data does not depend on cookie or browser blocking.

Meanwhile the budget was reduced, not increased: until the data was reliable, every extra euro would have bought more traffic that could not be assessed.

With the data restored, the campaign started optimising on real information again. In the following weeks the store’s orders picked up and, within two months, were back above the levels before the drop.

The recorded return changed too. In the month of the fix the campaign returned in sales less than a third of what it spent. Two months later, more than double, with lower spend.

When sales fall, check the data first, then the campaign. The campaign is the first suspect, but often not the cause. Changing ads, products or campaign type on broken data means judging every change with the wrong yardstick: even a right move would look useless.

Automated bidding amplifies the data it receives, right or wrong. With correct data, automated bidding is the most efficient tool there is. With missing or inflated data, it buys the wrong traffic with the same efficiency. Increasing the budget in these conditions does not speed up results: it speeds up spend that cannot be assessed.

The warning sign is the gap between the platform and the back office. Purchases recorded by Google and the store’s orders never match order for order. But they must move together. When the platform shows almost zero sales and the store keeps receiving them, or the other way round, the problem is in the measurement, not in the market.

Google’s automatic changes are not neutral. They are calculated on the account data. If that data is wrong, the changes follow it: in this case they lowered the return target precisely because the recorded return was distorted. Every change to the account should have a person who decided it and a written reason.

Tracking is not installed once and for all. Conversion actions created and removed several times point to tracking redone several times with nobody in charge. Updates to the site, the theme, the plugins or the cookie banner can stop it without any warning. Server-side tracking reduces these risks, but it still needs regular checks.

Four checks to run on your account

1. Weekly comparison. Purchases recorded in Google Ads against orders in the back office, same week. If the gap changes a lot from one week to the next, the data is not reliable.
2. A single purchase action. Among the conversions used by automated bidding there must be only one purchase action. Two actions counting the same order double the signal.
3. The right value. The average value of a recorded purchase must be close to the store’s real average order value. If it is much lower or much higher, automated bidding optimises on a wrong number.
4. Auto-apply recommendations. In Google Ads, Recommendations → Auto-apply: check which changes Google can make on its own and whether anyone ever chose them.