Should we measure acquisition cost on paid spend only, or across all customers?

The two numbers can move in opposite directions

Paid acquisition cost improving while blended worsens usually means paid is capturing demand that would have arrived anyway. Branded search is the classic case: the ad platform records a cheap conversion, the customer was already yours, and total new-customer volume is unchanged while spend went up.

The reverse also happens. Paid cost rising while blended cost falls often means paid is doing upper-funnel work — creating demand that shows up later as branded search, direct traffic and referral. Judged in isolation, that channel looks like it is failing.

Which one to manage to

Manage paid at the campaign level, because that is where the levers are. Manage blended at the business level, because that is what actually determines whether growth is affordable.

The mistake is letting one team own only one of them. When the media manager is graded on paid cost and nobody owns blended, the rational move is to shift spend toward the cheapest-looking conversions, which are usually the ones you were going to get for free. Pairing the two numbers on the same marketing report removes the incentive.

What has to be true for blended to be trustworthy

• New customers are deduplicated. Otherwise the denominator inflates and the number improves for no reason.
• All marketing costs are captured. Fees, tools, creative and lead purchases, not just media.
• Referral and repeat are separated out. You want to see the share of new customers arriving without a per-lead price, because that share is the real health signal.
• The window is long enough. Blended metrics inherit every lag in the business, so short windows are noise.

The referral leak

Referrals are the most under-measured source in service businesses because nobody asks in a structured way and self-reported source data is unreliable. That leak biases blended acquisition cost in a specific direction: it makes paid look responsible for customers it did not create.

Reducing that error usually means inferring source from behavior and history rather than trusting a dropdown — matching inbound calls against existing customer households, prior addresses and referral patterns in the operational record. That inference is a normal part of customer intelligence work and it changes the blended number materially.

The correction usually moves the blended number in one direction only: upward, because referrals and repeat work stop being credited to paid campaigns.