Geographic targeting in paid media is almost always set up as a binary decision. A market is either in the plan or out of it. A zip code is either targeted or not. The question is "where do we advertise?" rather than "how much do we advertise in each place we've decided to be?"
That framing makes the buy easier to execute, but it leaves substantial efficiency on the table. Sales penetration almost never distributes evenly within a targeted market. Some zips inside a DMA produce ten times the sales per impression of others. Treating them all the same wastes spend on the weak ones and starves the productive ones.
What the Data Looked Like
For a consumer brand we worked with, the imbalance was stark. Sales penetration concentrated in a small subset of zips within each targeted DMA — pockets over-indexing by 5x or more relative to surrounding zips. The brand's media plan, set up at the DMA level, was buying impressions in those zips at the same rate as adjacent under-indexers. The result was the kind of mismatch that's invisible in standard reporting because the reporting is also at the DMA level.
Pre-optimization, 37% of media impressions were going to zip codes that produced just 6% of total sales. The plan was technically targeted — but inside the targeted geography, the math was nearly random.
A Continuous, Not Binary, Optimization
The fix wasn't to drop the underperforming zips. Some have growth potential the historical data understates; others are just noise from small populations or short measurement windows. Cutting them entirely creates the opposite problem — over-optimizing on the recent past and missing the next batch of buyers.
Instead, we built a continuous prioritization layer. Every zip in the targeted geography received a priority score based on absolute sales contribution and sales-per-target-household penetration. High-volume, high-density zips got the highest scores; low-volume, low-density got the lowest. The scores drove a custom delivery algorithm that allocated impressions in proportion to priority — more weight to productive zips, less to unproductive ones, none of them dropped entirely.
Crucially, the optimization didn't run through CPM bid adjustments. CPMs stayed flat. The delivery shift happened at the impression-allocation level, which kept the buy clean and avoided the auction-dynamics problems that come with aggressive bid manipulation.
What Changed
weakest-performing zip codes
manipulation in the optimization
binary on/off targeting
The team stopped framing geographic decisions as "should we target this zip?" and started framing them as "what weight should this zip receive?" That's a more useful question, and it scales naturally as the business expands into new markets — where the early data is always thin and a binary decision would be premature.
Most consumer businesses with measurable geographic sales data have the same opportunity sitting in their media plan. Restaurants, DTC brands, retailers, services — anywhere the customer base concentrates more sharply than the targeting plan does, there's efficiency to be recovered by replacing binary geo logic with weighted prioritization.
Building a smarter geo-prioritization layer for your paid media? Say hello.