Lookalike: like what? Plus everything else we shipped this month

From the Product Team

Every month our product team ships a mix of work: new capabilities, upgrades to tools you already use, and smaller changes that make day-to-day work smoother. This is our roundup of what's new and why it matters.

This month has a common thread. Much of what we shipped is about the inputs: how precisely you can define an audience, and how quickly you can see whether it's worth building. Better inputs make better audiences, and that's especially true for the capability we're featuring this month.

Featured: Multi-signal lookalike modeling across diverse behavioral inputs

Intuizi's lookalike modeling takes an audience you already have, the seed, and finds more devices that resemble it. What sets it apart is what the model can learn from. Instead of judging similarity from a single type of signal, it can draw on location, app activity, demographics, transactions and profile attributes.

While lookalike modeling expands reach, incorporating broader inputs can help provide a more accurate picture of attributes and behaviors, defining which signals truly belong in a well-formed lookalike audience.

Start by deciding what "like" means

Someone reading running-shoe reviews has shown interest. Someone who regularly visits running stores has shown a different behavior. Both are useful to a running-shoe brand, but they give a model very different starting points. To reach people researching their first pair, web interest may be the right seed. To reach regular shoppers, start with store visits.

You can make a seed more deliberate. Audience modifiers such as visit frequency let you seed on frequent visitors rather than one-time drop-ins. Purchase criteria, such as repeat buyers or category spend, define a different starting point altogether.

The goal isn't to narrow until a handful of devices remain. It's a group whose behavior fits the campaign and is large enough to model. The same rule applies to familiar descriptions. "Adults 18–54 who like running" sounds specific, but without concrete behavioral signals, both the wide age bracket and the interest remain guesswork.

Let the model see more than the seed

A seed defined by store visits doesn't limit the model to comparing store visits. Because it can consider several signal types, patterns in app activity, for example, may help separate your seed from everyone else. Those patterns are worth testing, not assuming.

Intuizi works with permitted, de-identified device-level signals, not named customer profiles. Coverage varies by signal type, so if purchase data isn't available for a device, that doesn't mean no purchase happened.

Choose your reach deliberately

A smaller modeled audience stays closer to the patterns in your seed. A larger one gives a campaign more room to scale, but the fit loosens as reach grows. Neither end is automatically right; pick the size the campaign can actually use.

Before you build a lookalike, ask

  • What behavior defines the seed?
  • Have refinements made it more relevant without making it too small?
  • Which signals can the model use, and where is coverage limited?
  • How does the result change at the reach we need?
  • Can we deliver and measure it in the intended channel?

Also new this month

Audience estimates in seconds

Sizing an audience used to mean running most of a full audience build just to get a count, whether or not you were planning on activating that audience, historically this took minutes. Supported estimates now return in seconds, and uses less compute for processing, so it also lowers the cost for users. You can test ideas the way you'd have a conversation: try a definition, see the size, adjust, and send it to a full build only when you're ready to activate.

Fast estimates cover most single-signal audiences today; more complex combinations still run the full calculation.

Event-based location audiences

Not every visit to a stadium is a game-day visit. You can now tag a location with a specific event and its time window, such as an open house, a home game or a convention. Then build an audience of devices that were there during the event, not just at the location.

This is stricter than our existing time-of-day and day-of-week filters, which narrow the hours rather than tie an audience to a specific event. It suits real estate open houses, recurring sports and venue events, conventions and festivals.

Plus a run of smaller improvements

Alongside these, the team shipped a set of smaller updates across data exports, campaign analysis, delivery controls and account security. None needs its own announcement, but together they make everyday work in the platform faster and smoother.

Interested in learning more about these new product releases? Talk to your Account Manager or Book a demo today!

Your greatest asset is the data you already have.See what it’s been holding back.

Get a Demo