How We Picked 11 Counties Over Statewide Using a 268 Community Dataset

Most geographic targeting decisions get made by feel. The client says "we cover the whole state," somebody selects the state, and the campaign launches. Nobody checks whether the state is actually where the customers are.

Here is a case where we did check, using a public dataset, and the answer was different enough to change the buy.

The situation

A manufactured housing campaign covering New Jersey. Under Meta's Housing special ad category we could not use ZIP codes, could not apply geographic exclusions, and faced a 15 mile minimum radius on any dropped pin. In a state as narrow and as densely bordered as New Jersey, that combination makes radius pins close to unusable, which we worked through in why a 15 mile minimum radius breaks targeting in dense corridors.

So the real choice was between county selections and statewide. The question was which counties, and whether the trimming was worth the complexity.

Finding ground truth

The useful thing about manufactured housing is that the customers are not evenly distributed. They are concentrated in land lease communities, and communities are physical objects that somebody has counted.

The Federal Reserve Bank of Philadelphia published exactly that count. In June 2024, Eileen Divringi's report Manufactured Housing Communities in New Jersey introduced a custom dataset of 268 manufactured housing communities across the state, broken out by county and by size category.

That is a proper inventory, assembled from public records and property data rather than estimated from survey extrapolation, and it is published with its methodology. It is the kind of source that turns a targeting argument into arithmetic.

The distribution is heavily skewed. Per the report's county appendix, the largest concentrations sit in Atlantic (38 communities) and Ocean (38), then Monmouth (31), Cape May (28), Cumberland (26) and Burlington (21). Gloucester holds 17, Middlesex 15, Camden 10 and Salem 9. At the other end, Essex County has none at all, and several northern counties have one or two.

That shape is the entire argument. When roughly three quarters of the target inventory sits in a handful of counties, treating all 21 counties as equally worth buying is a choice you should have to justify. It is the geographic version of the point we make in how to find motivated mobile home sellers: go where the inventory actually is, not where the map is convenient.

The three options we priced

We built three candidate geographies and scored each on two numbers: how much of the 268 community inventory it covered, and how many people we would have to reach to cover it.

For the second number we used Meta's own estimated audience size readout at build time, and we combined the Fed's community counts with lot counts to get an approximate homesite total per geography. That gives a single efficiency figure we could compare across options: estimated lots per million people reached.

Option A: eight counties. The core southern and coastal block: Atlantic, Ocean, Monmouth, Cape May, Cumberland, Burlington, Middlesex and Mercer. Covered 199 of 268 communities, about 74 percent. Efficiency came out at roughly 8,216 lots per million people reached.

Option B: eleven counties. The same block plus Gloucester, Camden and Salem. Covered 235 of 268 communities, about 88 percent. Efficiency came out at roughly 8,342 lots per million people reached. Total estimated reach about 2.7 million.

The arithmetic reconciles against the Fed's county appendix. Atlantic 38, Ocean 38, Monmouth 31, Cape May 28, Cumberland 26, Burlington 21 and Middlesex 15 give 197, and Mercer's small count takes the eight-county block to 199. Adding Gloucester 17, Camden 10 and Salem 9 gives 235.

Option C: statewide. Covered all 268. Reach about 7.4 million people. Efficiency dropped to roughly 4,089 lots per million people reached.

The result that decided it

Look at what happens between A and B.

Adding three counties took coverage from 74 percent to 88 percent and improved efficiency at the same time, from about 8,216 to about 8,342 lots per million.

That is the counter-intuitive one, and it is the reason this exercise was worth doing rather than guessing. The usual assumption is that expanding a geography always dilutes it, so you trade coverage against efficiency. Here it did not. Gloucester, Camden and Salem are dense in manufactured housing relative to their population, so adding them raised the numerator faster than the denominator. Intuition would have told us to stop at eight. The data told us to keep going.

Now look at what happens between B and C.

Going statewide picked up the last 33 communities, about 12 percent of the inventory. It did so by adding roughly 4.1 million people, taking reach from 2.7 million to 7.4 million. Efficiency more than halved, from about 8,342 to about 4,089 lots per million.

Those 33 communities are the ones scattered across the northern counties, mostly in ones and twos, in exactly the corridor where Meta's audience estimates balloon because of proximity to New York. We were being asked to roughly triple our reach to add an eighth of our target inventory.

What "efficiency" actually buys you

It is worth being precise about the mechanism, because "efficiency" on its own is a word rather than an argument.

Under the Housing category, Advantage+ Audience is effectively the targeting model. It explores inside whatever geographic bounds you gave it, looking for conversion patterns. The bounds are the single biggest instruction you get to give it.

If two thirds of the bounds you hand it contain almost no manufactured housing, that is where a large share of the exploration budget goes. You pay for the learning, you get impressions in expensive metro auctions, and you get form fills from people who live in apartments and rowhouses. The campaign is not broken. It is doing precisely what you told it to.

Tightening the geography does not make the algorithm smarter. It makes the search space smaller, which matters most on the accounts with the least signal, which is most accounts.

The honest tradeoffs

Three things we gave up, stated openly.

We gave up 33 communities. Real ones, with real owners in them. If the client's business had any northern New Jersey concentration, this would have been the wrong call. We checked that it did not before we made it, which is the sort of service-area question worth settling before the media plan rather than after, as we argue in lead generation for mobile home dealers.

County selections are more work to maintain. A statewide buy is one line. Eleven counties is eleven lines, and somebody has to notice when the service area changes.

The analysis has a shelf life. The Fed dataset is a 2024 snapshot. Communities close, get redeveloped, and change hands. It is the best available inventory, not a live feed.

And one thing that does not follow from this analysis: none of it says anything about what the campaign will cost or produce. Targeting efficiency is an input. Results depend on the offer, the creative, the season, the competition and the follow-up, and they vary widely by account. We get into that in what a manufactured home dealer lead actually costs.

Tightening the geography does not make the algorithm smarter. It makes the search space smaller, which matters most on the accounts with the least signal, which is most accounts.

How to run this for your own market

The method transfers to anything where the customers are countable objects rather than an abstraction.

1. Find a real inventory of the thing you sell to. Federal Reserve community development research, state licensing registries, county assessor records, permit databases, association directories. For manufactured housing specifically, the Philadelphia Fed has published equivalent reports for Pennsylvania and Delaware.
2. Get it to county level. Counties are the finest grain most public data publishes at, and conveniently they are also the finest grain Meta's Housing category lets you target.
3. Price each candidate geography in Ads Manager and record the estimated audience size rather than guessing at population.
4. Compute a single efficiency ratio, targets per million reached, so the options are comparable on one number.
5. Expand until efficiency stops improving, then stop. The stopping point is usually not where your intuition said it would be, in either direction.

The whole exercise took a few hours. It is the cheapest thing you can do to a media plan.

Caveats

The 268 community count and the county breakdown come from the Philadelphia Fed report and are theirs, not ours. The coverage percentages, lot estimates, reach figures and efficiency ratios are our own calculations, made during a 2026 campaign build, combining that report with Meta's estimated audience size readout. Meta's estimate is an estimate and it moves. Our lot counts are approximations derived from the report's size categories rather than a homesite census, so treat the efficiency figures as a comparison between options rather than as absolute values.

Sources: Eileen Divringi, Manufactured Housing Communities in New Jersey, Federal Reserve Bank of Philadelphia, June 2024; full report PDF

Keep reading

• Why a 15 mile minimum radius breaks targeting in dense corridors
• Meta's Housing special ad category: the constraints nobody writes down
• Lead generation for mobile home dealers