How to Read Ad Metrics Without Fooling Yourself

An ad dashboard is a sales instrument. It is built by a company that wants you to keep spending, using data it collects about itself. That does not make it dishonest. It makes it a source that needs reading carefully, the way you would read a supplier's own quality report.

Here is how to read one.

The four numbers, ranked by how much they matter

Impressions and reach. These tell you the platform delivered something. They are inputs, not results. Impressions are the thing you bought, not the thing you wanted.

Clicks and click-through rate. These tell you the creative got attention. A high CTR with no sales usually means the ad promised something the page did not deliver.

Results, meaning the conversion you optimized for. This is the first number connected to your business, and only if you chose the right event. Optimizing for "landing page views" produces a lot of cheap results that mean nothing.

Revenue and margin. The only number that settles anything, and the one the ad platform can see least well.

Most bad decisions come from managing the first two because they move fastest.

Attribution is a modelling choice, not a fact

Meta's own documentation refers to attribution settings including "1-day click, 7-day click or 1-day view." That last one matters enormously. A one day view attribution means a person who saw your ad, did not click, and later bought is credited to the ad.

Sometimes that is fair. Often it is not. The person may have been going to buy anyway. There is no way to tell from inside the dashboard, which is exactly the problem the research addresses.

Gordon, Zettelmeyer, Bhargava and Chapsky compared randomized experiments against normal advertiser measurement using "data from 15 US advertising experiments at Facebook comprising 500 million user-experiment observations and 1.6 billion ad impressions," and concluded that "commonly used observational approaches based on the data usually available in the industry often fail to accurately measure the true effect of advertising." Two of the four authors worked at Facebook at the time.

Blake, Nosko and Tadelis showed the sharpest version of the same problem in paid search at eBay: "brand-keyword ads have no measurable short-term benefits." People searching your business name were coming anyway. The ad simply collected the credit.

Ask of any impressive number: would this have happened without the ad? The dashboard is structurally incapable of answering that.

Your web analytics does not see social either

Google Analytics defines its Direct channel as "sessions in which the user typed the name of your website URL into the browser or came to your site via a bookmark." In practice that bucket also catches traffic that arrives without a referrer, which includes links opened from messaging apps, some in-app browsers, and pasted URLs.

That is the phenomenon Alexis Madrigal named "dark social" in The Atlantic in 2012, reporting Chartbeat data in which "almost 69 percent of social referrals were dark." That figure is from 2012, from one analytics vendor's set of media sites, with no published sample size, so treat it as an illustration of the mechanism rather than a current statistic. The mechanism has not gone away: private sharing is invisible to referrer-based analytics, and private sharing is how most recommendations between real people actually happen.

Also note what Google says about the default model: "Paid and organic last click: Ignores direct traffic and attributes 100% of the key event value to the last channel that the customer clicked through." A model that ignores direct traffic and gives all credit to the last click will systematically flatter whatever you were doing at the end of the journey, and erase everything that created demand earlier.

How to read a benchmark

Benchmarks are the most quoted and least examined numbers in marketing. Here is a real example, and it is instructive.

WordStream publishes widely cited Facebook advertising benchmarks. The page states its methodology plainly: "This report is based on a sample of 256 US-based WordStream client accounts in all verticals (representing $553,000 in aggregate Facebook spend) who were advertising on Facebook between November 2016 and January 2017." It also notes that "average figures are median figures to account for outliers."

Two hundred and fifty-six accounts. Half a million dollars. A three month window from 2016 and 2017. Meanwhile the page carries a recent "last updated" banner, which is how a decade-old sample ends up quoted as a 2026 benchmark in a pitch deck.

Their Google Ads benchmark report is larger, stating "a sample of 14,197 US-based WordStream client accounts in all verticals (representing over $200 million in aggregate Google Ads spend)" for the period "August 2017 and January 2018."

None of this makes the numbers useless. It makes them specific: they describe the clients of one agency, in the United States, in a particular period, reported as medians. That is a real thing to compare against, as long as you say all of that out loud.

So the checklist for any benchmark:

1. Who was sampled, and how many?
2. What period does the data cover, and is that the same as the page's update date?
3. Is the "average" a mean or a median?
4. Whose accounts are these, and are they like yours in size and country?
5. Does the number come with a range, or just a single figure?

If the answers are not published, the benchmark is decoration.

What to actually do instead

Set the comparison before you spend. Write down your current weekly bookings, quote requests, or orders. That baseline is the only control group a small business can afford.

Use a holdout when you can. Run the ad in one town or one week and not another, then compare. It is crude, and it is closer to an experiment than anything the dashboard offers.

Ask people. A "how did you hear about us" field on your booking form is unfashionable, self-reported, and biased, and it will still tell you more than a view-through conversion will.

Judge on cost per real outcome against customer value. If a customer is worth $300 to you and a booking costs $60 to buy, you have a business decision. If a click costs $0.40, you have nothing.

Know when you cannot know. Meta's own Conversion Lift test requires a campaign "with a spend of $5,000 USD or more" and "a minimum of 500 optimized conversions." Below that scale the honest answer to "did the ads cause this" is that you cannot measure it, and you should decide based on cash flow and judgment rather than pretending the dashboard settled it.

The one thing dashboards never show

The customer who found you because someone in a group chat vouched for you, then followed you quietly for four months, then walked in. That path shows up in your analytics as Direct, with no source, attributed to nothing.

It is also the path most small businesses actually grow on. Measure what you can, then stop mistaking the measurable part for the whole business.

When this was last confirmed

Google Analytics documentation was retrieved in August 2026. The Meta Conversion Lift requirements come from an archived snapshot of Meta's help page dated January 2025. The WordStream methodology statements were read from their live pages in August 2026, where the stated sample periods are 2016 to 2018. The two academic papers are peer reviewed and dated 2015 and 2019, and the dark social article is from 2012.