The metrics in the chain
- CPM — cost per thousand impressions.
- CTR — percentage of impressions that become clicks.
- CPC — cost per click.
- Conversion rate — percentage of clicks (or visits) that turn into a result.
- CPA or CPL — cost per action or per lead. CPA = CPC ÷ conversion rate: if CPA rises, the cause is the cost of the click, conversion, or both.
- Cost per customer — spend divided by actual customers, from the CRM.
The stretch the platforms can’t see
Between click and lead sits the website. Two opposite problems produce the same symptom — few leads —: people who click and never arrive (often tracking parameters lost in a redirect, or slow pages), and people who arrive and don’t convert (the page doesn’t deliver on the ad’s promise). Telling them apart takes the site’s analytics data, not just the platform’s.
Between lead and customer sits the CRM. One campaign can bring in lots of cheap leads that never buy, and another a few expensive leads that become customers. Optimizing on cost per lead would cut the second.
Attribution, without fooling yourself
Every platform tends to claim the conversion for itself: adding up the results reported by several platforms often gives you more than the real total. Attribution models — last click, first click, linear, data-driven — distribute credit in different ways, and none of them is “the truth.” Two practical rules: compare campaigns counted the same way, and treat associations as associations, not causes.
How to do it
1. Bring the sources together
Ad platform exports, site analytics data, outcomes from the CRM. Recalculate CTR, CPC and CPM from the base data: every platform rounds them its own way.
2. When a cost changes, decompose it
How much comes from the cost of the click, how much from conversion, how much from budget shifting between campaigns. And define a threshold below which a change is noise.
3. Follow the post-click
For each campaign: how many clicks reach the site, how many of those convert, and on which pages people drop off.
4. Go all the way to the customer, with mature cohorts
Calculate cost per customer only on leads that have had time to close: last week’s leads haven’t had a chance to become customers yet, and they drag the rate down misleadingly.
5. A few moves, with verification
What to change first, why, and how to tell whether it worked.
The most common mistakes
- Optimizing on cost per lead. You reward the campaigns that bring contacts, not customers.
- Adding up the conversions the platforms report. The same customer gets counted several times.
- Comparing platforms that count differently. A click on one platform isn’t always a click on another.
- Cohorts that are too recent. New leads haven’t closed yet.
- Reacting to noise. Small week-to-week changes aren’t a signal.
