What is Google actually learning from?

I got involved in a Google Ads conversion tracking problem.

The client runs a taxi service. They've got a fairly typical setup. Someone clicks on an ad, books a taxi through a widget on the site and then pays by card.

The booking ends up in Hubspot. And the widget reports it as a conversion via Google Tag Manager.

The problem was that the booking count and value in Google Ads was very different to what's in Hubspot.

Matching lead count in a CRM to conversions in Google Ads is always difficult. In this case there are a few more challenges:

• Patchy mobile internet means that sometimes successful bookings don't get recorded as conversions.

• Cancellations or changes don't update the conversion count or value.

• The widget validates the card and confirms the booking but the customer's bank still has to complete the transaction. Most of the time that's fine, but occasionally a card is declined a few minutes later. Again this doesn't change the conversion count or value.

In this past these discrepancies weren't a big deal, because Google used keywords as its main signal. Today with smart bidding conversion signals like leads or bookings take priority over whether someone typed in "taxi near me" or "airport taxi service".

That makes the quality of the conversion signal as important as anything else in the ad account. You don't want Google learning from incomplete or incorrect signals.

I think this is where browser-based conversion tracking starts to show its limitations. Google learns from what the browser sent at that moment. That's not always what ultimately happened.

You want Google Ads optimising for conversion signals that are:

• Right. They should reflect what actually happened in the business, not what a browser thought happened.
• Reliable. A temporary network problem shouldn't permanently change what Google learns about your business.
• Auditable. You should be able to see exactly what data Google is optimising from and verify that it matches reality

Why I've standardised on offline conversions

Using offline conversions makes it possible to tick these boxes.

• You can trigger conversions from actual business events in your CRM, booking system or lead management system.
• If there's a temporary problem, you can upload them again.
• And, depending on how you set things up, you can see exactly what data Google is learning from and verify that it matches what happened in the business.

For example, here's a screenshot (from a different client) which shows which conversion was uploaded and what was data was sent to Google. If something looks wrong, I have somewhere to start investigating.

Traditionally, offline conversions were used for down-funnel events like qualified leads, booked appointments and sales.

I take a slightly different approach. I use them for everything, including first-contact events like form submissions, phone calls and WhatsApp messages.

Having one system accountable for all conversions makes monitoring easier and saves me looking for answers in three different places if something goes wrong.

Going back to the taxi company. Google and HubSpot showing different numbers is annoying. But the real problem is that Google was learning from a version of events that isn't always true.

I think this is becoming more of a factor in ad profitability than most people realise.

We put a lot of effort into monitoring things like budget pacing, landing page uptime and so on. I think conversion signals deserve the same degree of care.

Related articles

Offline conversion tracking with Contact Form 7

The Google Ads WhatsApp conversion tracking guide

Offline Conversion Tracking with Fluent Forms

The problem with call duration conversions

My lead count doesn't match my conversion count

Google Ads conversion tracking for WhatsApp orders

Topics you'll find on this site