The Ads Get Blamed First

When leads fall on a property marketplace, there are three suspects: demand, supply and plumbing. The ad accounts, the listings and the enquiry records each see one of them. Behavior on the site sees all three, and two investigations on sample data show what happens when it sits next to the rest.

On a property marketplace, the leads number is watched more closely than almost any other, because it is the number clients are paying for. When it falls, the first meeting tends to go the same way. Someone opens the ad dashboards, someone asks whether the agency changed anything, and the budget conversation starts before anyone has checked whether the ads moved at all. The reflex is understandable. Ad spend is the lever closest to hand, and the ad consoles are the dashboards that open fastest.

The trouble is that the reflex looks at one suspect out of three.

Three suspects, three systems

A lead is the last link in a chain, and the chain can break in three places.

The first is demand: fewer people arriving, or the people arriving caring less. Its evidence lives in the ad accounts and in the clickstream, the record of who visited, from where, and what they looked at. The second is supply: less on the shelf. A client takes listings down, a category thins out, stock goes stale. Its evidence lives in the listings and clients data, who owns each listing and when it went up or came down. The third is plumbing: interest that is still there but no longer turns into an enquiry, because a form, a call button or a booking step stopped working somewhere. Its evidence lives in the app’s own records of enquiries, often a production database that no analytics tool ever sees.

One record runs through all three. The clickstream sees people arrive and where they came from, which is demand. It sees which listings they open, and so which stock is still carrying interest, which is supply. And it sees them reach for the contact button, which is the top of the plumbing. It is the only record that watches the whole chain from the visitor’s side. On its own, though, it can’t say what changed: it doesn’t know a client took listings down, what the ads cost, or whether an enquiry actually arrived.

The usual setup keeps these apart. Each system has its own dashboard, so each suspect gets examined on its own, by whichever tool happens to hold its evidence, and the suspect with the most convenient dashboard gets examined first. But a drop in leads is not a question about any one of those systems. It is a question about the join: what people did, next to what they sent, next to what was there to look at and what it cost to bring them. The other systems say what changed. Behavior, sitting next to them, says which suspect it was.

Two investigations on the real estate page show what that looks like. Both run on sample data: a fictional property marketplace built for demos, with a clickstream, enquiries exported from the apps, listings and clients, and Google and Meta ad exports. The company is not real. The patterns are ones marketplace teams will recognise.

The enquiries that vanished

The first investigation starts from rental enquiries and narrows one small step at a time. Monthly totals for January to March. Then split by city. Then one city, Pune, split by app: Android, iOS and web. In the sample data, rental enquiries from Android in Pune went from 42 to 28 to 0 across the three months, and the last one arrived on February 16.

At that point there are two stories that fit the numbers, and the reflex reaches for the first one. Android users in Pune lost interest, which sends the team back to the ads and the listings. Or the Android enquiry step stopped working. The enquiry records cannot tell them apart; both look like silence. The clickstream can, once it is narrowed to exactly the same slice: views and gallery views on Android, of the Pune rental listings, before and after the last enquiry. In the sample data, those listings were viewed 84 times a day before February 16 and 85.5 times a day after. People kept browsing the same listings, on the same app, in the same city. They stopped enquiring.

The investigation’s own conclusion was careful: the pattern fits a broken enquiry step better than lost interest, but the data does not establish the cause. That is the right place to stop. The data can show the shape of a break; it cannot see the bug. The person who knows what shipped to the Android app around that date closes the case. The decision is an engineering ticket, not a marketing plan.

The drop the ads didn’t cause

The second investigation starts where the first meeting usually does, with a plain question: how many leads a day in February, and how many in March? In the sample data, leads fell from 124.6 a day to 99.6, about a fifth. The “a day” matters. February had 28 days and March had 31, and comparing monthly totals would have overstated the fall before the investigation had begun.

Then the ads, platform by platform, and here the clickstream did the work the ad exports couldn’t. Spend per day on Google and on Meta held about level, but spend only says what was paid. Paid visits per day from each platform, counted from the sessions on the site, held about level too: the ads were still bringing the same people. That matters because of a trap in the sample: the Meta export stops on March 21. Averaging March spend over all 31 days would have produced a fall in spend that never happened, and a ready-made culprit with it. The visits are what close it. Meta’s visitors kept arriving after the export ended, so the gap was in the file, not in the traffic.

With demand ruled out, the next question went to supply: which client’s leads fell the most, and did that client take anything down? In the sample data, one client’s leads went from 38.3 a day to 12.5, after it took 40 of its 120 listings down on March 10. Set that client aside and the rest of the marketplace’s leads held steady.

The decision that follows is different from the one the first meeting was heading towards. It is a conversation with one client about its stock, not a new brief for the agency or a cut to a budget that was doing its job.

Behavior is the witness

Put the two cases next to each other and the pattern is the point.

The enquiry records could show that Android enquiries in Pune stopped, but not whether anyone was still looking, and only behavior on those exact listings could separate a broken step from a bored audience. The ad exports could show what was spent. Only the visits could show the ads were still working, and only the visits could show that the Meta gap was in the file rather than in the traffic. The listings data could show a client’s stock coming down, and the enquiry records could show its leads falling with it.

In both cases, the other systems said what changed and behavior said which suspect it was. Neither half was enough alone. The clickstream without the enquiry records can’t see a lead; the enquiry records without the clickstream can’t see interest. Put together, at the level of a single platform, city and set of listings, they turn a puzzle into a short list. That is the case for asking one question across all of it, rather than reading four dashboards and reconciling them in a meeting.

The useful checks along the way were small: per day rather than per month, export coverage rather than calendar days, the rest of the marketplace with one client set aside. None of them is clever. Each one is a choice about what counts, and each one can quietly decide the answer if nobody sees it being made. That is why BumbleB Crunch™ shows how every number was counted, under every answer, in the language of the business.

“We’d have found it eventually”

Probably. A team that knows its marketplace will usually get to the client who pulled its listings, or the Android form that broke, sooner or later.

The cost is not that the answer can’t be found. It is which suspect gets acted on first. Every week spent re-cutting a budget, briefing an agency or redesigning a listing page for a drop that came from one client’s stock, or from one broken button, is a week spent on the wrong suspect. Meanwhile the client conversation that would actually help doesn’t happen, and the fix that would actually help doesn’t ship.

The faster path is simply asking the question across all of it at once.

Pointing it at a marketplace

The setup follows the three systems. A Google Tag Manager tag on the site sends listing views, searches and contact clicks. Enquiries come from the marketplace’s own database through bb-connector, or as uploaded exports, alongside clients, listings and ad-spend files. Then the questions are the ones teams already have: leads fell this month, was it the ads or the listings? Which city, app or listing type lost enquiries, and from when? Which clients get the most leads for what they pay? Which listings get viewed but never contacted?

Crunch investigates across all of it and shows its working. The team still makes every call that matters: whether to talk to the client, which fix to ship, where the budget goes. What changes is that the call starts from the right suspect.

When the leads move, the ads will still be the first thing anyone looks at. They just shouldn’t be the only thing.

Try the same investigation on your own listings: bumbleb.co/real-estate.

Key Takeaways

• A lead drop has three suspects: demand, supply and plumbing. The ads are only the first of them, and often not the guilty one.
• The ad accounts, the listings and the enquiry records each see one suspect. The clickstream is the only record that watches the whole chain from the visitor’s side.
• The other systems say what changed; behavior, sitting next to them, says which suspect it was. The enquiry records can’t tell a broken step from a bored audience. Views of the same listings can.
• Small choices about what counts (per day, export coverage, one client set aside) can decide the answer, which is why they belong on the page under every number.
• The cost of the reflex isn’t that the answer goes unfound. It’s the weeks spent acting on the wrong suspect.

Both investigations described here run on sample data: a fictional property marketplace built for demos. The full case studies are at bumbleb.co/real-estate. This piece continues “Crunch Is Open. Hold It to Both.”