Objective-first marketing built on real purchase behavior.

Geminus turns a business objective into who to reach, using individual cross-retailer transaction data, and measures whether that objective actually moved.

Start a conversation

The problem

Campaigns start with audiences. Businesses start with objectives.

Brands care about outcomes: new households, category growth, trade-up, reactivation, product adoption. But campaigns typically begin with whatever audience is easiest to activate: a retailer's definition of "new," a CRM list, a platform interest segment, a third-party cohort.

Most of that spend hits households that have no real likelihood of making the desired change in behavior anytime soon: becoming customers for the first time, returning after lapsing, increasing purchase frequency, moving to a higher tier of products, trying a new product from the brand. And most of the households that are actually somewhere on a trajectory toward one of those changes are missed entirely.

The message makes it worse. A broad audience forces a generic campaign, and a generic campaign does not move the households that could be moved. Brands know what they want: find the households already moving toward their product and concentrate the message and the spend on them. So instead, clients ask for what is feasible: a proxy audience.

How it works

An automated loop, end to end: from objective to measured outcome.

  1. 01
    Define the objective
    State the business goal in plain language: new households, category growth, product launch, trade-up, reactivation.
  2. 02
    Strategies & audience definitions
    The objective is translated automatically into executable strategies and precise audience eligibility definitions.
  3. 03
    Campaign-specific language models
    A compact language model, fine-tuned on real cross-retailer purchase outcomes, scores the full eligible population per strategy. World knowledge of what products and categories mean is built in and applied automatically to every audience.
  4. 04
    Scored cohorts & holdout
    The eligible population is ranked and split into treatment and holdout by design, before activation.
  5. 05
    Activation
    Cohorts activate through existing platforms: LiveRamp, TTD, DSPs, retail media networks. Geminus does not buy media.
  6. 06
    Lift measurement
    Outcomes are read in the same cross-retailer transaction data. Not clicks. Not platform attribution. Real purchase behavior.

See the full architecture

Why Geminus

Objective-native, not seed-native
No starting segment required. The audience is derived from the objective itself: households that achieved that outcome historically, based on their pre-outcome behavior.
Multiple automated models per campaign
Each strategy gets its own supervised model, trained automatically per campaign: NTB propensity, repeat conversion, winback hazard, trade-up propensity.
Individual cross-retailer transactions
Training signal is individual observed purchase behavior across retailers, not modeled proxies, not single-retailer data, not platform behavioral signals.
Holdout by design
The control group is defined before activation. Incrementality is built into the workflow, not retrofitted after the fact.
Execution stays where it is
Geminus sits upstream. Campaigns run through DSPs, retail media networks, walled gardens, and CTV. Nothing in the execution layer changes.
Outcomes in the same data
Results are read in the same cross-retailer transaction layer the model trained on. The question and the answer live in the same data.

Where it fits

Geminus strengthens the starting point for every activation platform.

LiveRamp
Objective → cohorts → identity resolution → activation across all destinations → lift in transaction data
The Trade Desk
Geminus improves the starting audience. TTD execution unchanged. Lift in purchase data, not platform attribution.
Retail Media Networks
Market-level objective and cross-retailer measurement without retailers sharing raw transaction data.
Commerce DSPs
Geminus defines the audience upstream. DSP execution is strong. The gap is market-level objective → audience + cross-retailer lift.
Agencies
A repeatable bridge between strategy and activation, with market-level proof of impact beyond platform metrics.
CPG Brands
Start from the business objective grounded in cross-retailer purchase behavior. Measure whether it moved.

From the blog

Latest thinking

View all posts
July 2026 · Geminus
The audience you're buying is not the audience you need
When a brand buys an audience, what they're actually buying is usually a stand-in for the audience they need. Both sides know it, at some level. The transaction happens anyway, because there hasn't been an alternative.
Read the post
July 2026 · Geminus
A model that's only seen your own transaction data can't connect a gym membership to a sports drink
A person connects a gym membership, deodorant, and a sports drink instantly. A model trained only on your own transaction history can't, and that gap is the whole argument for using a language model.
Read the post
July 2026 · Geminus
Search ads systems don’t understand what they’re doing
Ask a person to judge whether a search result is good and they do it instantly. Ranking systems can't, because they don't know what things are. They know what users clicked before.
Read the post

If the objective matters, the starting point matters.

Built by operators who spent twenty years inside the execution layer and know exactly where it breaks upstream.

Start a conversation

Or write directly: