When I arrived at eHarmony, the media operation looked like most consumer businesses of the era: a capable agency, a healthy budget, and a monthly deck full of green arrows. On paper it worked. Underneath, we were flying blind — paying a margin to sit one step removed from the decisions that actually moved acquisition cost.

So we brought it in-house. Over the following stretch we cut digital customer acquisition cost by 81%. That number gets attention, but the discount was never the point. The point was what made it possible — and why it kept compounding after the initial drop.

The agency black box has a hidden tax

An external media team is optimizing toward the KPI you hand them, on a reporting cadence they control, with a model you can't see. The fee is the visible cost. The invisible cost is latency: every insight has to travel out to the agency and back before it changes a bid. In a subscription business where signups, activations, and retention all feed each other, that lag is where money leaks.

Bringing the function in-house collapsed that loop. The people setting bids were the same people reading the cohort data — sitting next to the analysts who knew what a good subscriber actually looked like ninety days out.

Owning the data changed the math

This was the early era of large-scale data modeling in marketing, before it was fashionable. The advantage wasn't a tool. It was that we owned the full signal — every impression, every signup, every downstream conversion — and could model against lifetime value instead of a proxy.

Three things changed once the data was ours:

  • We optimized to value, not volume. Cheap signups that never converted stopped looking like wins.
  • We iterated in hours, not weeks. A creative or audience test could be read and acted on the same day.
  • We could prove incrementality. Spend that wasn't actually driving new subscribers got cut without a debate, because the data settled it.

Why 81% compounded instead of plateauing

A one-time efficiency gain is a discount. A compounding one is a system. Because we controlled the loop, every week's learning became next week's floor. Better targeting produced better subscribers, better subscribers sharpened the model, and a sharper model made the next dollar of spend more efficient than the last. The 81% wasn't the finish line — it was the point where the flywheel became obvious.

In-housing isn't about saving the agency fee. It's about owning the feedback loop that turns spend into a learning system.

When in-housing is the right call — and when it isn't

I don't think every company should bring media in-house. It's the right move when three things are true: acquisition is core to the business (not a seasonal push), you have enough volume for the data to teach you something, and you're willing to hire and retain the talent to run it. If acquisition is episodic, or you can't own the data, an agency's breadth usually wins.

But if growth is the business — as it is for most subscription and streaming companies — control is worth more than convenience. That principle carried straight through Pantaya, Paramount+, and the AI-driven marketing systems I build today. The tools keep changing. The advantage of owning your loop doesn't.