Valcour Ad-Targeting · loading
Loading the targeting assistant. This can take a few seconds while the data layer warms up.The targeting assistant ranks geographies by where an incremental advertising dollar has the most leverage, for any advertiser, whether you’re placing a product campaign or a political one. It is not a black box: the formula is published below, every input traces to a named public data source, and the assistant tells you when a signal is a proxy rather than a measurement. The same score also reports a confidence range, because each input is itself an estimate.
Favorable markets with a well-matched, under-targeted audience score highest; markets where ad money already floods in are discounted. For a political campaign, “favorability” reads as race competitiveness and “fit” as persuadable-voter density; for a product, they read as market demand and audience-demographic fit. Each component is a 0–1 signal traceable to a named source, and the final score carries a 5th–95th percentile band.
Every opportunity score scales reach by two more signals and discounts for ad-saturation. The two scaling signals are each normalized to a 0–1 range so they combine cleanly:
How favorable the market is for the advertiser, on a 0–1 scale. For a political campaign this is race competitiveness, a toss-up scores near 1.0, a safe seat near 0.0; for a product it is market demand. Weighted toward markets where the outcome is actually in play.
A 0–1 proxy for how well a market’s demographics match the target audience, for a political campaign, the share of reachable voters who are plausibly persuadable; for a product, the density of the demographic you’re selling to. Built from Census ACS county fields; a heuristic, not a measured count, documented in full in Step 3.
How much ad money is already flowing into that geography, normalized 0–1. The ÷ (1 + ad_saturation) term discounts already-crowded markets, so the score favors places the airwaves aren’t saturated yet.
The signals are assembled from four public datasets. The assistant analyzes what these sources report; it does not invent figures on its own.
Reported outside spending by geography, one input to how saturated a market already is.
Ad activity disclosed through Meta’s transparency data, a second view of where ad money is concentrated.
Reported category ad spend, a third saturation signal, combined with the two above.
County demographics, race, education, and age fields that feed the audience-fit proxy.
Live API keys for the spend sources are still being provisioned. Where a geography has no per-geography spend data yet, ad-saturation defaults to zero and the ranking reflects reach × favorability × fit alone, the assistant records that this happened so a reader knows which signal was unavailable. We describe what the score analyzes, not a data scale we can’t yet stand behind.
The default fit proxy rewards counties that are demographically mixed rather than lopsided, on the premise that a broadly winnable audience clusters where no single group dominates. In the political case this maps to persuadable voters: a county that is overwhelmingly one group, uniformly high- or low-education, and very old or very young tends to be a partisan stronghold with few cross-pressured voters. So the proxy averages three “moderation” sub-scores, each peaking at a neutral midpoint and falling off toward the extremes:
One minus the largest single-group share among white, Black, and Hispanic populations. A 50/30/20 county scores higher than a 90/5/5 one.
Peaks where the college-degree share is near a national-ish midpoint and falls off for very low- or very high-education counties, which both tend to skew toward a single audience (and, in the political case, toward one party).
Peaks near a working-age median and falls off for very young or very old counties.
The three sub-scores are averaged over whatever fields are present; a county with no usable ACS fields scores zero rather than guessing. This is deliberately a transparent heuristic, every term maps to a real ACS field, and we call it a proxy because that is what it is.
A single score would imply a precision the inputs don’t have. So each geography is run through a seeded Monte-Carlo pass: the three inputs are perturbed by modest, documented Gaussian noise, reflecting that favorability estimates are noisy near a toss-up, audience fit is a proxy, and spend totals are partial, and the formula is re-evaluated thousands of times. We report the point score plus its 5th and 95th percentiles. When two geographies’ ranges overlap, the assistant says so instead of implying a difference the data can’t support. The pass is seeded, so the same inputs reproduce the same range.
Being explicit about the boundaries is part of the method, not a disclaimer bolted on after:
The score never claims that spending a given amount in a geography produces a number of additional votes or sales. It does not model causal outcome lift at all.
A high score means conditions, a favorable market, a well-matched audience, low existing saturation, favor advertising having leverage. It is a prioritization signal, not a forecast of what advertising will achieve.
The assistant tells you where to look. It does not place buys, negotiate rates, or manage flights. It points; you decide and execute elsewhere.
Ad-saturation combines Meta, Google, and FEC only, not every channel an advertiser can run on. Where spend data is missing for a geography, that component is treated as zero and flagged.
This is a tailored, early product rather than a mature platform with a long track record. We are honest about that. What we will not do is dress a heuristic up as a measurement or quote a data scale we can’t back. Every component of the score is published, every input is a named public source, and the assistant flags any signal that fell back to a proxy or a default. As the live spend sources come online, the score gets sharper, the formula and its honesty stay the same.