--- title: "Demographics did not predict cost per lead. Comparable locations do." description: "Before building a forecast for new sites, we tested the industry's favorite inputs against 865 ZIP codes and about $102,000 of real ad spend. Here is what Audience uses instead." section: Inside Sneeze It author: David Steel published: 2026-10-10T20:15:23.522Z url: https://one.sneeze.it/blog/demographics-did-not-predict-cost-per-lead tags: ["inside-sneeze-it", "audience", "site-selection", "forecasting", "franchise-development"] --- # Demographics did not predict cost per lead. Comparable locations do. _Before building a forecast for new sites, we tested the industry's favorite inputs against 865 ZIP codes and about $102,000 of real ad spend. Here is what Audience uses instead._ Every site-selection pitch looks alike. A map, a circle, and a table of who lives there: median income, median age, share with a degree, how long they commute. The table implies a promise. Find the neighborhoods that look like your best members, and your ads there will bring cheaper leads. Before we built Audience, our market tool for new and existing locations, we tested that promise against our own results. We had the ad accounts. We had the leads by postal code. So we asked a plain question: when the demographics of a ZIP code change, does the cost of a lead change with them? :::stats 865 | ZIP codes tested $102K | of real ad spend, approximately 37 | client locations in the comparable set source: [Audience, About](https://audience.sneeze.it/about) ::: ## The test We lined up 865 ZIP codes and about $102,000 of real ad spend, and compared each ZIP's demographics with what a lead there actually cost. Median income, the input every broker leads with, showed effectively zero correlation with cost per lead on one account. On another it pointed the wrong way. Commute time predicted nothing anywhere we looked. Income did show a real effect in fitness on its own, but it did not repeat across accounts, and a pattern that appears in one account and vanishes in the next is not something you can sign a lease on. The test also caught something we were not looking for. An early version of the tool ranked ZIP codes, and it put federal mail-only ZIPs (the kind assigned to a government agency's mailroom, where nobody lives) near the top. They looked perfect because nothing bad was measured there. Nothing was measured there at all. We added a rule that an empty ZIP cannot score, and the lesson stuck: the absence of a bad number is not a good number. :::pullquote A forecast you cannot trace back to real locations is a guess with a chart on it. ::: ## What Audience does instead The forecast in Audience is not a curve fitted through demographics. It is a match. Given a new address, it describes the market around it, then finds the client locations Sneeze It runs whose markets most resemble it, and reports what those locations actually produced: enquiries a month and cost per enquiry. The match still uses market facts, but it weighs them by what our data showed matters: :::chart column "How much each market fact counts in the match" Density | 1.3 Competition per 10,000 people | 1.2 Households | 1.0 Median income | 0.6 Median age | 0.5 Share with a degree | 0.4 Commute time | 0 source: Audience matching weights, from the code ::: Density and competition lead because they describe the auction a location is buying into, and contested markets cost more. Income is in the match at a low weight: it informs the comparison without driving it. Commute time is out entirely, because including something that predicted nothing would only add noise. :::compare "A demographic forecast" "A comparable-location forecast" Takes the neighborhood's income, age and education, runs them through a model, and returns a number. Nobody can say which real business the number came from. --- Finds the locations we run whose markets look most like this one, and reports what they did. Every forecast names how many matches it came from, and says "low confidence" when nothing is close. ::: There is a second advantage that turned out to matter more than accuracy. A comparable-location forecast can show its working. A franchise development team can see that the range came from five locations with similar density and competition, and judge for themselves whether those are fair comparisons. Same brand beats same industry, because locations that share an offer and a sales process are the fairest comparison of all. ## The rules that keep the forecast honest A forecast built on real locations is only as good as the locations it uses, so Audience is strict about which ones count. :::flow "How Audience forecasts a new site" Address | The site you are weighing Market | Drive-time trade area, census, competitors Match | The most similar locations we run, weighted Check | Drop stale locations, mark down old ones Forecast | Enquiries and cost per enquiry, with a match count ::: **A location that stops reporting stops counting.** If a live location's leads have not arrived in our records for 21 days, its numbers describe a period, not the location, so the forecast leaves it out and says so. **Old accounts fade.** A location we no longer run can still count, but its cost per lead was set by an auction, an offer and creative from that year. It keeps full weight for a year, then counts for less and less, and after five years it cannot move a forecast on its own. **No match, no number.** On Audience's public homepage, the sample comparison shows three sites side by side. Two get a range of enquiries a month. The third gets "No forecast. Low, no close match." That is the correct answer when we have nothing like it, and we would rather show it than invent one. **Presales wait for evidence.** For a club that is not open yet, Audience will not forecast presale joins until it has at least four measured presales to learn from. :::figure wide ![An illustrated franchise development office where a person lines up small cards, each showing a simple street grid, beside a paper map with one magenta pin](/blog/media/a7cb863d7040b77b58db38ea.jpg) The forecast for a new site is built from the real locations it most resembles, and it says how many it found. ::: ## What this changes for a franchise team As of October 10, 2026, the live strip on Audience's homepage shows 3,860 enquiries measured in the last 90 days and 31 open locations feeding the forecast. Those numbers will move, and that is the point: the forecast is fed by locations that are open and reporting now, not by a demographic file bought years ago. It also changes the conversation with a franchisee. A demographic report invites an argument about whether the neighborhood is "right". A comparable-location report invites a better one: are these really the right comparisons, and what would make this site different? That is a question a franchisor and a franchisee can actually answer together. There are limits, and the tool states them. It works at the ZIP code level and nothing finer. It reports what comparable markets did, not whether a market is good. And it cannot report cost per member for most locations, because few of them report signed memberships back to us. Audience is live and used by our team. Clients see its output as a printed market report, not the tool itself. The drive-time trade area at the start of every forecast has its own story, which we tell in [A circle is not a trade area](/blog/a-circle-is-not-a-trade-area). For the same habit applied to ads, see [We judge ads by cost per lead](/blog/we-judge-ads-by-cost-per-lead). :::takeaways "What this means if you are opening your next 10 locations" - Ask any site report where its forecast came from. If the answer is a model run on demographics, ask which real locations it has been checked against. - Density and local competition told us more about cost per lead than income did. Look at those first. - A forecast should say how many comparable locations it used, and should be willing to say "no forecast". - Treat an empty or quiet data point as unknown, never as good news. ::: :::cta button="See Audience" href="https://audience.sneeze.it" Weighing a new site? We will walk you through a market study on a sample address, from drive time to forecast. :::