DarkMirror Pilot
Find more people like your best customers.
DarkMirror learns what your best customers have in common, including patterns no single field captures, and finds more households like them, even where the data your model needs is missing.
The missing-field problem
A model that needs an income field can’t see 4 in 10 households. DarkMirror reaches them through attributes that correlate with income.
DarkMirror is DarkMath’s semantic audience product. It transforms enhanced, embedded data, from deterministic fields such as age to CTV viewing and other behavioral data, into semantic attributes, creates unique audiences from them and carries the activation data to reach them. DarkAgent curates and deploys those audiences.
How DarkMirror works
Every audience starts from resolved people. A seed list full of duplicates teaches a model the wrong lesson, so DarkMatch cleans it first.
In earlier testing, DarkMirror delivered 22% higher audience reach with precision maintained.
Add semantic attributes
DarkData turns sparse fields into multidimensional profiles: life stage, digital behavior, media habits and more.
Profile your best customers
The seed becomes a centroid in vector space, the shared signature your filters miss.
Exclude first
Current customers, suppressions, opt-outs and patterns you want to avoid come out before expansion.
Bridge the gaps
Households missing a required field qualify on the attributes that predict it, with a confidence level you choose.
Curate and activate
DarkAgent curates the audience and deploys it through DarkMirror activation data.
More reach, with the precision on the table
Every expansion trades some precision for reach. DarkMirror shows you the trade before you buy media.
Curated audiences that learn
DarkMirror creates the audience. DarkAgent curates it, connecting your customer and response data, plus other demand signals and supply when available, to customize the audience on the fly.
Every campaign feeds a machine-learning loop, so a curated audience improves over time. Because the whole cycle can run through DarkAgent automatically, this is a new kind of audience for adtech.
What makes it work
Activation data, not a borrowed ID
DarkMirror includes the activation data audiences actually run on.
That means mobile ad IDs (MAIDs), hashed emails (HEMs), phone numbers and personally identifiable contact data such as names and postal addresses, used only where permitted. These are the identifiers IDs are built from. Working at this level makes DarkMirror agnostic to any one ID and never beholden to it.
Most audience platforms activate through a proprietary ID, such as LiveRamp’s RampID, so every audience depends on how that ID is built and where it is accepted. DarkMirror goes to the connective tissue itself: the identifiers behavioral audiences activate on across DSPs and SSPs.
What activation data makes possible
- ID-less connections: no single ID decides who you can reach
- Activation on both DSPs and SSPs
- Feedback at the level of each identifier, where permitted
- Granular machine learning, so curated audiences keep improving
One field becomes a profile
A field that reads “Age: 22” says little about how to reach someone. DarkData semantic attributes expand deterministic fields like age, along with CTV viewing and other behavioral data, into signals a planner can use, then combine them with behavioral personas such as value seekers or premium shoppers to build composite audiences.
| Dimension | Signal from one field | How marketers use it |
|---|---|---|
| Life stage | College age | Student services, first-time purchases |
| Generation | Gen Z | Short-form video, mobile-first experiences |
| Digital behavior | Digital native | Social platforms and apps over email |
| Economic behavior | Rental market | Renter’s insurance, flexible subscriptions |
| Media habits | Streaming primary | Connected TV and audio over linear TV |
| Career stage | Student or early career | Entry-level offers, starter products |
Guardrails built in
Similar is not the same
DarkMirror finds lookalikes. It never claims two records are the same person. That is DarkMatch’s job.
Reach, never eligibility
Semantic attributes support marketing and measurement only, never decisions about credit, insurance, employment or housing.
Exclusions first
Suppressions, opt-outs and your own exclusion rules are applied before any household is added.
DarkMirror questions
What is a semantic lookalike?
A person or household whose behavior and context resemble your best customers in vector space, even when the specific fields your model uses are missing. Standard lookalikes usually match on a short list of demographic fields.
How is DarkMirror different from platform lookalike tools?
Platform tools model inside one platform’s data and return an audience you cannot inspect. DarkMirror builds the seed profile from your resolved customers, shows the counts, exclusions and contact coverage, and delivers to the channels you choose.
Does DarkMirror rely on cookies?
No. It works from resolved identities and their attributes, and delivers through standard identifiers such as postal addresses, hashed emails and mobile ad IDs where available and permitted.
What do we need to start?
A seed list of the customers you want more of, ideally a few thousand, and the objective you are optimizing for. DarkMatch resolves the seed first so the profile is built on real people, not duplicates.
What is a curated audience?
A DarkMirror audience that DarkAgent customizes on the fly with your customer and response data, plus demand signals and supply when available, and then keeps improving through a machine-learning loop as campaign results come back.
What activation data does DarkMirror include?
Mobile ad IDs (MAIDs), hashed emails (HEMs), phone numbers and personally identifiable contact data such as names and postal addresses, used only where permitted. These are the identifiers IDs are built from, so audiences activate on DSPs and SSPs without depending on any one ID.
Does DarkMirror depend on a proprietary ID?
No. Proprietary IDs are built from the same underlying identifiers DarkMirror carries. Working at that level keeps audiences portable across DSPs and SSPs and lets feedback return at the level of each identifier, which is what makes granular machine learning possible.
Can DarkMirror audiences be used for credit, insurance or employment decisions?
No. Semantic and inferred attributes support marketing reach and measurement only. They are never used to decide eligibility for credit, insurance, employment or housing.
Show us the customers you want more of.
Bring a seed list and an objective. We will size the audience and show the precision trade before you spend.