Northline Tower: the A/B test that changed how we produce renders

A creative A/B test for a premium residential tower in Chicago

For years the visualisation industry has trained everyone to produce shots in one style, fixed in the brief up front, and then drop them into a marketing campaign. We were curious about doing it differently — about how to make the material live up to its potential and push sales results as far as they go.

Northline Tower is a project we made to show the level we work at, and to have something to test that hypothesis on for real rather than in theory. The test wasn't a simulation: we ran an actual campaign and measured actual numbers. We set up a situation where a developer is selling apartments in a premium tower in Chicago — a market where a buyer compares a dozen developments at once and first meets the offer as an ad on their phone. There's a budget, there's social media and there's a sales team. What there isn't is an answer to one question: which material actually sells.

The question sounds trivial until you work out what getting it wrong costs. Across a campaign running for several months, the gap in cost per click between two creatives is thousands of dollars and hundreds of lost contacts. And the decision that “this render is the nicer one” is usually made in the fifteenth minute of a meeting — on taste, not on data.

We started as a visualisation studio that wanted to check its own work rather than assume it. We ended up with a method we now use on every development: instead of one final render we hand over variants built for testing.

Where the answer disappears

The standard process goes like this. The architect hands over the model, the studio produces one final render per shot, marketing puts it into the campaign, and a month later someone asks why the results are poor. Nobody knows, because there's nothing to compare it with.

In a campaign you measure budget, audience, format and placement. Creative is the one variable nobody tests, even though it's what stops the scroll and decides whether any of the other settings get a chance to work at all.

The reason is mundane and has nothing to do with marketing acting in bad faith. There's nothing to test, because the order was for one render. A creative A/B test is lost at the brief stage, not at the campaign stage.

How we designed the test

We proposed reversing the order. Instead of one “best” render we produced three variants of the same shot: day, night and winter. Same proportions, same framing, same composition, same building — the only difference being mood and light.

The whole difficulty of a test like this is keeping it clean. We stuck to three rules.

One variable. The variants differ only in time of day. If the framing or the presence of people had changed along the way, the result would still be interesting, but there would be no telling what had done the work.

Everything else identical. Same budget, same audience, same format. So everything we saw in the results came from the image, not from other settings.

Two markets at once. Chicago, where the development might be built, and Los Angeles, where part of the investor audience lives. One market gives you partial data; only the second shows whether the advantage repeats or is a local fluke.

It was a turning point in how we work today. A render stopped being a finished product and became a hypothesis to be tested, and a way to push sales results further.

What the results showed

The differences were bigger than we expected. The night variant won in both markets: in Chicago it delivered 20% more attention and traffic than the winter one, and in Los Angeles the gap reached 90%, almost double. It's worth saying that every test can come out differently and that a lot of factors feed into it. That is precisely why it pays to run them while you are selling.

Cost per click for the night variant was $0.31 in Chicago and $0.55 in Los Angeles — the gap between markets is natural, because media costs differ. The number that matters is a different one: the winter creative had a CPC 190% higher than the night one. Almost three times the price per click for the same building, the same budget and the same audience, with one difference — the time of day in the render.

That is what's at stake in a creative A/B test. The question isn't “which version is nicer”, it's “which version will you pay three times more for the same click”. The mood of an image isn't a matter of aesthetics, it's a line in your cost of sale — and it can be measured, provided you produce the material with a test in mind rather than a portfolio.

What this test did not prove

The night render “won” in this case, but that isn't a rule. On a development with a lot of greenery, with a water view, or with a different audience, the result can go the other way.

If we turned one project into the rule that “night sells better”, we'd be making exactly the same mistake as before — deciding up front, just with a nicer justification. The test didn't tell us the single formula for a render that sells best. What it gave us was a way of finding out in two weeks instead of guessing across a whole campaign cycle.

So we don't carry the Chicago result over to the next development. What we carry over is the method.

Creative A/B tests for your development

We plan the test variants during production, not after the campaign. It's a cost difference, not an ideological one: once the model, the framing and the composition exist, the second and third variants are a fraction of the work of the first — only lighting and post-production change. Adding a variant after the project has closed costs disproportionately more, and usually ends with nobody ordering it.

In practice: two or three variants of the key shot, differing in one thing at a time — time of day, season, or whether there are people in frame. Below two you have no comparison. Above three you spread the test budget so thin that the results settle nothing.

Then the test runs in the campaign, data comes in over two to four weeks, and you decide: what to scale, what to switch off, what to produce next. Creative stops being a matter of taste in a meeting and becomes a line you account for like the media budget.

So this isn't an idea we came up with to fill out an offer. We tested it on ourselves first, on our own project and our own ad budget, and only then wrote variants into our production standard and the test into how we run campaigns.

Beyond the renders we build the whole system

In the test we were checking the image alone, because that was the point. On a real development, though, the render is the way in, not the whole thing — around the winning variant we add the full set of materials that work at the later stages of the buyer's journey.

An interactive model for exploring the development — the buyer browses the building, the floor plans and the location from their phone, with geolocation and a map of the area. It's a tool for the agent in a meeting and for the buyer between meetings. Embedded in the development website it works around the clock and sells in the middle of the night.

Urban context map — in a premium development you aren't buying square metres, you're buying an address and a view. The map shows what's in the immediate area.

Digital brochure, website, sales materials and social content — one visual language across every medium, from the same sources, with no re-production for each new campaign.

The logic of putting it together this way is simple: the campaign brings the traffic, the model and the website hold it, the sales materials close it. If any one element speaks a different language, you lose attention at exactly the point where you paid for it. A creative test only makes sense if the winning variant has somewhere to lead.