The perimeter problem: how to draw a geo-fence that pays, and why Performance Max is the only campaign type that can hold it

Somebody walks into the marketing meeting with a circle on a map. Twenty-five miles around the store. That is the perimeter. The budget gets built on it, the geo-fence gets drawn to it, and nobody in the room asks where twenty-five came from.

It came from nothing. It is the same number the last agency proposed, and the one before that. A radius drawn from a store address is a guess with a media budget attached, and it usually spends forty percent of the money in places that will never buy a car from you.

Here is how to actually determine the perimeter, and why Performance Max is the only campaign type that can operate inside one this precise.

The circle is the wrong shape and the wrong size

A radius ignores three things that decide whether a customer exists.

Drive time is not distance. A twenty-five mile radius across a river with two bridges is a ninety-minute fence. The same twenty-five miles down a highway is thirty minutes. Your customers experience the drive, not the geometry.

Demand is not evenly distributed. It clusters by suburb, by school district, by which side of the interstate the service bay sits on. A circle spends equally across all of it.

The buying decision is not one decision. A service customer chooses on proximity and today. A used-vehicle buyer will drive further for the right truck. A parts buyer may not be in the metro at all. One fence cannot serve three behaviors, and a single circular fence serves none of them well.

The three layers that define a real perimeter

Every defensible perimeter is built from three layers, in this order. Skip the first and you are modeling someone else's market.

Layer one: your own closed-loop demand

This is the only layer that is true rather than inferred. Pull it, geocode it, plot it.

Sort the delivered-unit addresses by drive time, not distance, and find the band that contains eighty percent of them. In most dealer datasets that eighty percent sits inside twenty to thirty minutes for sales and inside twelve to eighteen minutes for service. Do not accept those figures. Compute yours, because a store on the edge of a metro, a store with a dominant service brand, and a store inside a dense urban grid will all produce different shapes.

Layer two: market demand you do not own yet

Your closed-loop data tells you where you have won. It does not tell you where you could win.

• Search demand by geography. Search terms report and geographic reporting in Google Ads, plus Search Console for organic, aggregated to the postcode or metro level. Look for demand clusters with low impression share. That gap is the acquisition opportunity.
• Registration and garaging data. Third-party registration data by postcode shows where the vehicle parc sits, filtered to your makes and to the age and mileage bands that generate service and trade cycles.
• Google Business Profile interactions. Direction requests and calls by distance band tell you how far people already travel. If your profile draws from thirty minutes out, your fence is too tight.

Layer three: friction and eligibility

The layer everyone forgets, and the one that decides whether a wider fence can be serviced.

• Drive-time reality at the hours that matter. A forty-minute drive at 2am is a twenty-minute drive at 5pm.
• Title, tax and registration boundaries. State lines and county lines create paperwork friction. A fence that crosses one needs a reason.
• Competitive density. A cluster of same-brand stores fifteen minutes away is not an acquisition market, it is a price war with your name on it.
• Dealer group store-of-record. If you are part of a group, the fence must not buy leads the sister store already owns. That is not perimeter targeting, it is internal cannibalisation.
• Lift cost. The incremental cost to serve or deliver beyond a band: shuttle time, delivery fees, trade appraisals that require an extra visit.

Draw it in drive time, in bands

Once the layers are plotted, express the perimeter as concentric drive-time bands rather than one hard edge. Bands let you bid differently at the center than at the fringe, which is the entire point.

The expand band is not a rounding error. It is where you buy growth, and it is where the marginal arithmetic has to work.

The marginal-mile cut

The perimeter ends where the next mile stops paying. Run the arithmetic band by band and the fence draws itself.

The table is illustrative. The method is not. You are looking for the band where cost per unit crosses the gross you actually keep, and where close rate falls off a cliff. In this shape, the 20 to 30 minute band is still worth funding at a lower bid and the 30 to 45 minute band is a rounding error with a media invoice attached.

Two rules that fall out of the arithmetic.

• Never set the fence at your average customer distance. Averages get dragged by the outliers you spent the most to acquire.
• Set the fence where close rate breaks, not where lead volume breaks. Cheap leads at the fringe are cheap for a reason.

Where the fence actually gets enforced

A perimeter drawn in a spreadsheet is a hypothesis. It only becomes an operating constraint when it is set in the platform and policed monthly.

• Location targeting set to presence only. "Presence or interest" serves ads to people who show interest in your area from anywhere. It is the single most expensive default in Google Ads for a dealer, because it spends your local budget on people who are not local. Set presence.
• Exclusions drawn deliberately. Sister store territories, neighbouring metros where a different brand store is the correct answer, and any postcode where your own closed-loop data says you lose more than you win.
• Radius targeting where the interface allows it, wrapped around the store and around the service drive. Radius plus presence-only is the tightest fence the platform supports.
• Metro-level reporting joined back to the polygon. Matched location, not user location, when you audit spend.

That last point is the one that separates a fence from a nice drawing. If you cannot see spend and return by band, you have not set a perimeter. You have set a preference.

Why Performance Max is the only campaign type that can hold this fence

The perimeter is a demand map with price tiers. Performance Max is the only Google campaign type that is built to trade a single goal across every surface while respecting a geographic constraint. Six reasons, in the order they matter to a dealer.

One goal, every surface, one budget. Search, Shopping, Display, YouTube, Discover, Gmail and Maps compete against each other inside a single campaign for the same conversion goal. Search-only campaigns cannot reach the customer who never types your category. Display-only campaigns spend on impressions with no purchase intent. Performance Max puts your fence in front of the entire Google footprint and lets the system decide where the next dollar earns the most, inside the geography you allowed.

Presence-only location targeting applies across all of it. This is the whole argument. A geo-fence that leaks is not a fence, and the surfaces where leak is most expensive are the ones dealers never watch: Display placements and YouTube inventory. Performance Max enforces your location setting across every surface in one place rather than in nine campaigns you have to audit separately.

Value-based bidding aligns the machine with gross, not leads. Feed it conversion values that reflect front-end gross by model line or by band and it optimises toward the money. Lead-count bidding inside a fence produces the cheapest possible lead inside the fence, which is almost never the profitable one.

Asset groups carry the perimeter message. A "ten minutes from the service drive" asset group with its own headlines, its own imagery and its own offer can sit alongside a metro-wide new-vehicle asset group in the same campaign. Geo-fencing stops being a bid setting and starts being a message.

Audience signals and search themes short-circuit the learning period. A new store, a new fence, or a new model line has no conversion history. Signals built from your closed-loop customer data, the postcodes you already win, and the makes and models in your own parc give the system a starting point instead of a cold guess. This is the layer that most dealers skip, then blame the campaign for.

Account-level negatives keep the fence from leaking. Excluded brands, excluded competitor terms, excluded service terms in a sales campaign: the fence has to hold against the query as well as the map.

The honest limitation, and the architecture that solves it

Performance Max sets location targeting at the campaign level. You cannot put a different fence on each asset group. So banded bidding requires banded campaigns, not a single campaign with clever asset groups.

This structure is the perimeter. Each campaign is a band, each band has its own cost per unit, and the monthly audit decides which band grows.

The perimeter audit, monthly

One meeting, one page, five checks.

1. Share of spend inside the polygon. If more than a small fraction sits outside, something is misconfigured, usually presence-or-interest on a campaign nobody checks.
2. Cost per unit by band against the marginal table. Any band crossing your break-even line gets its bid cut, not its budget raised.
3. Close rate by band. If the fringe close rate is improving, the expand band is earning a wider fence. That is growth.
4. New clusters. A postcode that never appeared is a signal, not noise. Find out whether it is a new housing development, a competitor closure, or a delivery partner.
5. Exclusions to add. Sister store overlap, price-war postcodes, and any geography where the fleet or commercial data says you cannot serve.

Where Performance Max is the wrong answer

Three cases, stated plainly, because a studio that claims a tool always wins is not being read carefully.

• Budget under roughly $4,000 a month. Performance Max needs volume to learn. Below that threshold, a tight search campaign on high-intent service and model terms inside the core fence will out-earn it.
• A catalog with no feed hygiene. Performance Max for retail runs on the feed. Wrong part numbers, missing GTINs and thin descriptions produce a campaign that optimises toward the wrong products with great efficiency.
• A brand-new store with zero conversion history and no customer data. Here the perimeter is a guess no matter what you run, so buy data first. A small search campaign inside the tightest defensible fence, sixty days of results, then hand the fence to Performance Max.

The fence errors that cost the most

• Setting the radius once and never re-deriving it. Housing growth and competitor closures move the eighty percent polygon every six months.
• Using one fence for service and sales. The service customer lives closer than the sales customer, always.
• Ignoring drive time on the wrong side of a river, a rail line or an interchange.
• Letting presence-or-interest run because someone read that it "increases reach." It increases reach. That is the problem.
• Reporting on user location instead of matched location, then concluding the fence works when it is the targeting that is doing the work.
• Buying leads a sister store already owns, then counting them as growth.

Start with your own delivered addresses sorted by drive time, find the band holding eighty percent of them, put a price on the next band, and let Performance Max spend inside the shape your customers already drew. The circle was never a strategy. It was a placeholder.

We keep a working version of that argument on the report page: reach against rate, paid reach separated from organic, because a fence that buys impressions and a fence that earns saves are not the same result.