Edition 1 · 8 September 2026
Of everything advertisers count, how much could ever be traced to revenue?
We opened 38 real Google Ads accounts covering $1,695,957 of spend and counted what their conversions actually are. Free, no email required, and the working is published below.
Conversions in a revenue-linked category
977 of 285,052 conversions · 5 of 38 accounts
Purchase, store sale, qualified lead or converted lead. The categories Google reserves for an outcome that represents money or a closed sale.
Conversions carrying a real value
13,556 of 285,052 conversions · 13 of 38 accounts
A value that varies with the sale rather than a flat number applied to every conversion. Anything under five per conversion is treated as a placeholder.
Conversions that are both
966 of 285,052 conversions · 5 of 38 accounts
A revenue-linked category carrying a value that could plausibly be revenue. This is the closest thing in the data to a provable sale.
The finding
Accounts recording an imported qualified or converted lead.
Not one of the 38 accounts recorded a single conversion in the categories Google reserves for a lead a business later confirmed was real. Every account in this portfolio stops counting at the enquiry.
A conversion can carry a money value, and that value is what automated bidding optimises towards. In this portfolio it is nearly always a placeholder.
| How the value is set | Conversions | Share |
|---|---|---|
| A flat value of exactly 1 | 224,263 | 78.7% |
| No value at all | 46,502 | 16.3% |
| A value that varies with the sale | 13,556 | 4.8% |
| A flat value under 5 | 730 | 0.3% |
A flat value of one is the default a great many tag setups apply without anyone choosing it. It tells the bidding system that every conversion is worth exactly the same as every other, which is the one thing that is never true.
| Category | Tier | Conversions | Share |
|---|---|---|---|
| Contact | Intent | 198,438 | 69.6% |
| Book appointment | Intent | 32,679 | 11.5% |
| Default / other | Unclassified | 21,531 | 7.6% |
| Outbound click | Proxy | 11,643 | 4.1% |
| Submit lead form | Intent | 10,657 | 3.7% |
| Phone call lead | Intent | 4,690 | 1.6% |
| Store visit | Proxy | 1,385 | 0.5% |
| Page view | Proxy | 1,008 | 0.4% |
| Purchase | Revenue | 977 | 0.3% |
| Sign-up | Intent | 808 | 0.3% |
| Get directions | Proxy | 710 | 0.2% |
| Engagement | Proxy | 359 | 0.1% |
| Begin checkout | Proxy | 107 | 0.04% |
| Add to cart | Proxy | 35 | 0.01% |
| Unknown | Unclassified | 25 | 0.01% |
Account counts per category are deliberately not published, because at this sample size a small category would identify an advertiser.
The gap is not that the tracking is broken. It is that the counting stops too early.
Nearly seven in ten conversions in this portfolio sit in a single category, Contact, and almost all of them carry a flat value of one. Those are real signals of interest. They are not sales, and nothing downstream ever tells the advertising platform which of them became one.
The consequence is not a reporting inconvenience. Automated bidding learns from the conversions it is given. Feed it undifferentiated enquiries at an identical value and it will faithfully buy more enquiries, including the ones nobody wants, and the reported cost per conversion will improve while the business does not.
The single most valuable change available to almost every account here costs nothing in media. Decide which counted events actually represent money, attach the value they were really worth, and send the confirmed ones back. That is the work the missing join describes, and on this evidence essentially nobody has done it.
- Source
- The Google Ads Data Transfer into BigQuery for a single manager account, read directly. No sampling, no modelling, no third-party estimate.
- Window
- 8 September 2025 to 8 September 2026, a rolling 365 days.
- Population
- 38 advertising accounts that spent more than $100 in the window. 37 of them recorded at least one conversion. Together they spent $1,695,957 and recorded 285,052 conversions across 126 distinct conversion actions.
- Revenue-linked
- A conversion whose category is purchase, store sale, qualified lead or converted lead. These are the categories Google reserves for an outcome representing money or a confirmed sale.
- Real value
- A conversion action whose value varies with the sale. An action whose total value divided by its conversions is exactly one, under five, or zero is treated as a placeholder rather than revenue.
- Not published
- Account names, conversion action names, per-account figures and per-category account counts. Conversion action names in this data frequently contain live client domains, so they never leave the warehouse.
- Reproducing it
- The query is a single aggregation over the account conversion stats table. We will share it with anyone who asks, and we will correct this page in public if someone shows us it is wrong.
- This is one portfolio, not an industry. 38 accounts under one manager account, weighted towards hospitality, travel and service businesses in the United States and India. It is what we can see, read directly, rather than a survey.
- It measures counting, not matching. We are comparing what accounts count against what could represent revenue. We are not comparing against their accounts, because we only hold both halves for clients who gave us both.
- A soft conversion is not automatically wrong. A lead generation business is right to count an enquiry. The gap is that almost nobody carries that enquiry forward to the sale and tells the platform which ones were real.
- Categories depend on setup. They are chosen by whoever built the conversion action, and they are sometimes chosen badly. This portfolio contains a phone call action categorised as an outbound click. Miscategorisation moves these numbers at the margin.
- A placeholder value can be deliberate. Some advertisers set a flat value on purpose to weight one action above another. It still cannot tell bidding what a customer was worth.
Nothing has been corrected yet. When a published figure changes, the change and its date appear here rather than being edited away quietly.
Next edition: December 2026. Found an error? Tell us and we will publish the correction.
Want this run on your own account?
The free Attribution Leak Audit is the same method applied to one account, with your own numbers instead of a portfolio average.