Views, clicks, visits, purchases: deterministic data is the record of what people do, and its endless supply is exactly what makes performance media optimizable in real time. But what is easy to know is not the same as what is worth knowing, and the most valuable questions in business are exactly the ones this data cannot answer.
Do people know us? Trust us? Would they choose us? What do they feel, believe and intend? What did our message change? The answers to these questions live only in people’s minds, and asking enough people, often enough, in enough places, has never been possible.
That is the problem AMP solves.
At scale means
Global: 117 countries
In real time, while the campaign is live
On every channel where media runs
10x the sample of legacy survey research offerings
Building a brand strategy takes knowing what people think. Executing one takes knowing how every investment is changing minds while the campaign is still live. Most of it starts today, with a pixel and a media plan. The rest opens up when your identity platform and ours are connected.
Composite examples drawn from live studies. Categories are real, the situations are real, the names are not shown.
A business computing launch aimed at a narrow professional audience. Off-site lead generation could not be attributed, so the agency was defending the buy with platform metrics that said nothing about the brand.
A brand lift study fielded continuously through the flight, with the pixel broken out by tactic and by audience strategy so every line carried its own read. Three KPIs chosen against the objective: familiarity, favourability and message association.
The average hid almost everything. One audience strategy delivered several times the favourability lift of another. One message out of four carried the message association result. And the role with the lowest starting favourability produced the highest lift of any group.
At the halfway point the survey answers themselves became targeting. Respondents who answered the right way were turned into a live audience and pushed back into the running campaign, and budget moved onto the tactics that were working.
The in-flight move
Respondents who viewed the brand positively, respondents already familiar with the product, and respondents in the three target roles. All of them identified by the study that was already running.
Those seeds became an addressable segment inside the same campaign, extending reach past the original targeting rather than waiting for the next flight to apply the learning.
Spend moved onto the tactics and audience strategies carrying the lift and away from the ones that were not moving, while there was still budget left to move.
What it changed for the business
The outcome was not a slide at the end of the flight. It was incremental brand lift bought with the same budget, a target audience reached at a scale the original plan could not deliver, and a measurement framework that replaced platform metrics as the standard for the following year.
Retail media networks reported units moved and stopped there. The brand team could not tell whether any of it was building the brand or simply harvesting demand that already existed.
Brand outcome measured alongside the retail buy, across five retailer networks, on the same method used for the rest of the plan.
Two of the five were moving volume with no measurable brand effect at all. That changed how the next year's retail media budget was argued for internally.
Purchase happens once every several years, considered slowly and rarely online. Optimizing to clicks or completed views was actively selecting for the wrong people.
Optimization pointed at stated consideration rather than at any behavioural proxy, with the audience score trained on the study's own responses.
The score stopped chasing habitual clickers and started finding people who said they were in market, which is a population you cannot identify from behaviour alone.
Bookings were soft mid-season. The team had theories about destination preference, price and competitor activity, and no way to test any of them before the season ended.
A five question study in field within days across four markets, asking the category question directly rather than inferring it.
The blocker was price perception, not destination preference. Creative and messaging changed inside the flight rather than in next year's planning cycle.
Every brand on the roster measured brand outcomes with a different vendor, a different question set and a different control methodology. Nothing compared, so nothing could be optimized across the portfolio.
One instrument and one control construction across the roster, in every market where those brands buy media.
A single league table across brands, markets and channels. For the first time the agency could argue about where money should sit rather than about whose numbers were right.
Media ran continuously all year. The brand tracker landed once a quarter, months after the spend it was supposed to explain, and never lined up with a specific flight.
Continuous measurement on every flight, plus a brandless category question fielded alongside it to find people actively choosing where to eat.
A weekly read instead of a quarterly one, and an audience of people who had just said they were switching, which then became the retargeting pool.
Measurement on its own is the on-ramp. This is where the integration takes it: no case study can be public yet, so here is the whole machine running once, a top ten global advertiser, every channel at once, one continuous loop.
It starts before any campaign: a live crosswalk between your identity platform and ours. Matched hold-in audiences, full-fidelity measurement, no match-rate noise.
The business problem arrives:
To solve any of them you need to know what people think. So you bring your first party data, find your customers in the crosswalk, and field a consumer insights survey on them.
A follow-up study turns learnings into structure:
Real declared answers, not personas.
Segments become addressable audiences, onboarded to every partner on the plan. They travel as your first party data; we hold the twin.
Before launch, lock the primary KPI: the business goal, quarter over quarter, plus the secondaries. In flight, one survey program covers the whole plan: validated opportunity-to-see qualifies exposure inside the walled gardens, while pixels and data files attribute everything else deterministically.
The click is not the business goal.
The result: a cross-channel single source of truth.
In flight, respondents who answer the right way become the seed of optimized audiences, rebuilt and refreshed daily, extending beyond your CRM into new markets. Granular reads cover every deterministic channel: creative, format, placement, strategy. A partner-level relative view covers everything, walled gardens included.
Move budgets, pause lines, steer to the KPI.
The post-campaign analysis: which combinations and sequences of channels moved the number, which creatives carried it, and what changed in your segments. A study of this scope targets a respondent base of 20,000 or more.
Every study feeds a training base tied to your outcomes. Around ten campaigns in, it unlocks two scores, in any market where the loop has run. Pre bid or post bid, at the speed of attention or viewability metrics, at a fraction of the cost of surveying everything.
Run the loop where it matters most. The scores carry it everywhere else.
Measurement is fully managed. The short version: you bring the campaign, we run everything else.
Campaign details and media plan at intake
Pixel trafficking on digital and CTV placements
Connection to the supply source for out of home play logs
Survey approval before anything fields
Sharing results with the end client as needed
Survey design and programming
Pixel generation, exposure ingestion and identity matching
Control group construction and maintenance
Survey fielding and pacing
Analysis, dashboard provisioning and post-campaign reporting
A dedicated account manager on the study throughout
A data services agreement covers pixel-based collection. Fielding is compliant with GDPR and other applicable privacy frameworks.
The next step is a working session: walk the methodology, scope the first campaign.
One integration
Three capabilities
One continuous loop
Bring us a campaign.
Audiences built from what devices reveal, and audiences built from what people say. Both at global scale, from one platform.
Most data businesses have one. The machine behind ours produces both.
All of it resolving to 2.6B surveyable profiles across 117 countries.
Scale is not a vanity metric. It is what makes everything that follows work.
Five families, twenty seven elements, joined to 2.6B device profiles in 117 countries.
Demographics start from a census foundation, then blend declared answers, publisher first party data and media inference.
Interest, from the media itself, is our wheelhouse. And every block is a starting point for a question.
Research companies have built survey based audiences for decades, because declared data works. The difference is the rails: surveys that travel like ads reach ten times the people a panel can.
Ten times the sample does not make a model slightly better. It changes what a model can be.
And the one segment that was never on the shelf is the question your campaign actually depends on: would they consider buying yours?
Write the question your business is really asking. We field it as media, in any country, in any language. The people who answer the way you need are not a proxy for your audience. They are your audience.
The seed is not who they resemble. It is what they said.
A supervised model, built for categorical data, trains on the answers to your question, with the platform's observed signals as its features.
It does not copy an audience's description. It learns what answering yes looks like.
Then it scores every profile in the country, 1 to 100.
Live across US, CA, EMEA, LATAM and APAC. Fielded in any market, in local languages. One survey per country, always.
Ask any question. Define any audience. Scale it to a tenth of a country, about two weeks from idea to activation.
When one travel category started moving on our shelf, we designed the segments the category was missing, from cruise intent windows to first time cruisers, and put them on the shelf before anyone asked.
Segments made by research, not by scraping.
Deterministic blocks when you know who you need
Predictive audiences when the question is the target
Custom builds when the shelf falls short
Delivered where you already buy data
Across 25 campaigns, audiences built this way drove up to 2x the click through of third party segments.
Built from answers.
One instrument: a short question, served as media, answered by a real person in the moment.
The survey platform serves a short question set as a full screen unit inside a mobile app, and returns the answers as they arrive. What changes between studies is the question, not the machinery.
There is no panel and no recruitment. The survey goes out through our own demand on our own supply, so the sample is whoever we choose to reach rather than whoever once signed up to be surveyed. Nothing is paid for the answer and nothing is asked that cannot be answered in a few taps, which is why people finish it. Responses land in the platform as they come in, so a study can be read while it is still in field.
How it runs
A survey in the wild: full screen, in a normal ad break, answered in a few taps and gone.
Why it gets answered
It arrives where attention already is, in an app somebody chose to open, in the slot a creative would have taken. It is not an email, a pop-up on a website or a portal somebody has to log into.
Nothing is offered in return, so there is no reward to farm and no professional respondent to filter out. Between five and fifteen percent of the people who see one answer it, which at our delivery is 800K answers a day.
What you can ask
The instrument is the same every time. These are the shapes it most often takes.
Awareness, consideration and favourability read continuously in market rather than once a quarter.
The same short ladder fielded to an exposed group and a matched control group while the flight is live.
What people buy, own, drive, watch, plan and switch, asked of the people actually doing it.
Two packs, two claims, two edits, put in front of real people in market instead of a lab.
Where somebody sits in a decision right now, and what would move them along it.
Any question a brand or a planner needs answered in market this week, drafted with you and fielded in days.
The facts
Runs in the ad slot, never in an inbox, a portal or an emailed questionnaire.
Non incentivized, so nobody is answering in order to collect something.
Brandless mode asks the category question without ever naming a brand.
Answers land live, so a study reads while it is still in field.
Roughly ten times the sample a conventional panel study would field.
Question sets are drafted by us and nothing goes live without client approval.
Custom experiences
Tap the one you would actually pick
A white labelled, gamified survey experience: the brand's own identity, rich media and a mechanic worth playing.
Built by our creative studio
Most studies run in the standard unit, and the standard unit is the right answer most of the time. When a client wants something more, our in house creative studio builds it: a white labelled experience carrying the brand's own identity, rich media, video, interactive product choices, or a gamified mechanic that earns the answer instead of asking for it.
These are scoped as special projects, designed with the client, and they field through exactly the same pipeline as a standard survey. The experience changes. What comes back does not.
The instrument does not care what the answer is for. Ask a brand question and it measures a campaign, ask a category question and it labels an audience, ask a market question and it researches a category. Same unit, same pipeline, same people answering.
Survey sample bought the way media is bought, through the ad tech ecosystem.
Every survey we field is a media buy. The research budget does not go into recruiting and rewarding a panel, it goes into the auction, and our own bidder decides which impressions are worth paying for.
That is a different discipline from research operations. Reaching a defined person through the ad tech ecosystem means winning them in a real time auction, at a price that holds, in an app where a full screen unit will actually be seen and answered. Get it right and sample scales with media rather than with recruitment. Get it wrong and a research budget disappears into impressions that never produce a response.
Research budget, spent in the auction
A conventional research provider spends the budget recruiting people, rewarding them, and keeping them on a list. The sample is whoever is on the list that week.
We spend the same budget in the open auction, buying the impressions where the right person is about to be. The survey runs in the ad slot that a creative would have taken.
Sample follows the audience definition rather than the panel roster, so the people answering a study are the people the study is actually about.
Why this is hard to copy
A bidder is not a feature a research company licences. Ours exists because LoopMe is a full stack ad platform: the exchange, the data, the identity graph and the models all feed the bid decision, 400B ad requests a day. A survey provider without that stack has to buy its sample through somebody else's demand side platform, which means paying a margin on every impression, losing profile level control over who is reached, and giving up the ability to field an exposed group and a matched control group in the same moment.
We also do not have to reconcile two systems after the fact. The bidder that buys the media is the bidder that buys the survey, so exposure and response are the same event seen twice rather than two files that have to be matched.
Performance per research dollar
Scale is not only a function of how much inventory we can see. It is a function of how efficiently each dollar converts into a completed, non-incentivized response, and that is an in-house discipline we have been refining for years.
Real time scoring of every bid request for how likely it is to produce a completed response, not just an impression.
Direct supply through our own SDK, so there is no reseller margin between the budget and the person.
Pacing, frequency and daypart control tuned to when people actually answer.
App and country level yield management, so spend moves to where responses are cheapest.
Unit and placement selection matched to the question being asked.
The same optimization loop that lowers a campaign's cost per outcome lowers a study's cost per response.
Owning the bidder is why a research dollar buys sample at media efficiency, and why the scale is ours rather than rented.
Three pieces of intellectual property that make measurement, audiences and optimization possible.
Thirteen granted AI patents sit behind the platform. Most of them make the ad business run more efficiently: bid pricing, pacing, supply selection, creative decisioning. Three of them are the reason AMP exists at all.
Each one solves a different problem. One decides who a campaign should be measured against. One turns a few thousand survey answers into a few million addressable profiles. One decides, minute by minute, which model gets to spend the money. They are separable, and each is difficult on its own. Together they are the difference between a research product and an operating system for brand outcomes.
RISA, the Residual Index Selection Algorithm
RISA answers the hardest question in brand measurement: compared to what? A lift number is only as good as the group the exposed audience is being compared against, and that group has to look like the exposed audience in every way except the exposure itself. RISA selects that control group from the same live population the campaign is buying into, matching across more than fifty variables covering demographics, device, media behaviour, location and category context.
The part that is genuinely hard is that the match does not hold still. As a campaign delivers, the exposed pool changes shape: new supply comes in, frequency builds, the optimizer shifts spend. A control group that was matched on day three is quietly wrong by day twelve. RISA rebalances continuously against the exposed pool as it evolves, so the comparison stays honest for the life of the flight rather than only at the moment it was struck.
The control is drawn from the same universe the campaign buys into, not from a separate panel.
No media has to be withheld to create a holdout, so measurement costs nothing in delivery.
Matching is continuous, so the comparison does not decay as delivery shifts.
Lift reads from 10 exposed and 10 control, with significance from 30 and 30, at 95 percent.
The reach extension algorithm
A survey reaches thousands of people. A campaign needs to reach millions. The reach extension algorithm is what closes that gap: a gradient boosted model, built specifically for categorical data, which is what human attributes actually are. Survey answers are the labels. The deterministic signal set held against every profile is the feature set. The model learns what distinguishes the people who answered a particular way, then scores the wider graph for everyone else who fits.
This is not modelling off a pixel pool or a list of site visitors. The seed is a stated answer from a real person, given in the moment, which means the audience has a verifiable basis rather than an inferred one. Each model is trained for the study that produced it rather than pulled off a generic taxonomy shelf, and it is rebuilt as new answers arrive, so an audience gets sharper the longer a brand keeps asking.
Labels are stated answers, not clicks, visits or inferred intent.
Features are our own deterministic signals rather than licensed third party segments.
Built for categorical data, which is how most human attributes behave.
Retrained as responses accumulate, so the audience improves with every wave.
The PurchaseLoop economy of models
Most optimization runs on one model per campaign, retrained on a schedule. PurchaseLoop runs an economy instead. Over 2,400 models are retrained every day and they compete against each other on live performance. Every five minutes the best performing model is selected and put in charge of bidding, and the ones that are losing are simply not used until they earn their way back.
What makes this valuable is what the models are competing on. They are not optimizing toward clicks, completion rates or any other proxy that happens to be easy to count. They optimize toward the survey verified outcome the study is measuring, which means the campaign is being steered by whether it is actually changing minds. A single scheduled model can decay quietly for weeks. An economy cannot, because something better takes over the moment it exists.
Over 2,400 models retrained daily, competing on live performance.
The best performing model is selected every five minutes, not every quarter.
Optimization targets the surveyed outcome rather than a proxy metric.
Decay is self correcting, because a weaker model stops getting spend.
The rest of the portfolio
The other patents are ad platform intellectual property: they make bidding cheaper, delivery more efficient and supply selection smarter, and they are a large part of why the economics work. They are not what a client is buying. These three are.
The patents matter less than what they enable: a defensible read on whether a campaign moved anyone, an audience built from what people said rather than what they clicked, and a bidder that keeps choosing the model most likely to move the next person.
The spine that proves we are allowed to survey somebody, and that we can actually reach them.
The graph does more than resolve an impression to a person. It is the mechanism that decides whether a person can be surveyed at all, and it is entirely our own.
Everything else in the stack depends on it. It is how consent is verified and enforced, how reachability is confirmed before a study is scoped, how a phone is connected to the household screen next to it, and how an exposure, an answer and an outcome end up attached to the same person. It holds 2.6B profiles across 117 countries, built from deterministic joins and refreshed continuously against live volume rather than sitting as a static file.
Consent, verified twice
We operate a double opt in. Consent is not assumed from the presence of an identifier, and it is not inherited from a data supplier.
Publishers in our exchange pass a consent signal when a user agrees, inside the app they are using, to have their data shared with LoopMe. No signal, no profile.
When the survey is served, the person accepts our privacy policy before answering. Consent is given again, in context, by the person actually responding.
Both signals are stored against the profile. We only survey people for whom both are present, and the graph is what makes that enforceable at scale.
Reachability, checked before anything is promised
Consent alone does not make somebody surveyable. They also have to be findable. Every audience we field against is matched to the spine first, whether it is the deterministic group of users exposed to a client's campaign or an off the shelf third party audience a planner has selected.
That match answers two questions at once. Do we hold consent for these people, and have we seen their identifiers recently enough, in apps they are actively using, to be confident we can put a survey in front of them. It is the difference between a feasibility estimate and a feasibility fact, and it is why a study scoped on our numbers tends to deliver on them.
Any audience, client first party or third party, is resolved against the spine before it is fielded.
Recency of identifier activity is part of the test, not just presence in a file.
Reach is assessed against apps the person is actually using, in our own exchange.
Feasibility, market minimums and expected sample all come out of the same check.
Devices, and the household around them
The graph resolves mobile devices to people and people to households. That is what lets us attribute an individual smartphone user to the connected TV in their home, and it is the reason we can measure brand lift on CTV at all: the screen that carried the exposure is not the screen that answers the survey, so the two have to be connected before the question is worth asking.
MAIDs for mobile and out of home, IP addresses for CTV, desktop and audio.
Household resolution links a connected TV exposure to the phone that answers.
Deterministic joins only, so an exposure is an exposure rather than an estimate.
A client identity platform can connect spine to spine. That crosswalk is the integration.
Why it is a moat
The graph is fully in house. We do not licence it, rent it or reconstruct it from somebody else's file, which means nobody can withdraw it and nobody else can sell the same thing twice. It exists because we are a full stack ad platform: it was built out of years of real ad transactions and it is maintained by billions more every day. That is not a dataset that can be bought and switched on. It takes years to create and continuous global delivery to keep alive.
This is the clearest line between us and other providers who also run surveys inside ad units. They can serve a question. What they cannot do is verify consent at the profile level, confirm reachability before quoting a study, connect a household screen to a personal one, or field globally against a base that behaves like a panel roughly ten times the size of a conventional one. The graph is why our footprint is global and why the sample is there when a study needs it.
Scale is not the point on its own. The point is that the same person can be recognized at exposure, at answer and at activation, with consent on file for every one of those moments.
A live signal stream on 2.6B people, and the reason the models mean anything.
Every bid request we see is a small, timestamped observation about a real person. We see 400B of them a day, in 117 countries, and we keep the ones that tell us something durable.
This is not a file we bought. It is a stream we are already inside, produced by our own delivery, arriving continuously and attached to profiles the graph has already resolved. Most of the industry reads a bid request once, prices an impression and discards it. We route every useful field to a place it belongs and then watch how it changes over months.
The lifecycle of a bid request
A request arrives with a set of fields. Some of them say who this is. Some of them say what this person is doing right now. Some of them only matter once you have seen the same person a thousand times.
One request tells you almost nothing. A person observed continuously across the apps they choose, at the hours they choose, in the places they go, becomes a picture that is difficult to assemble any other way.
What we are genuinely good at knowing
Which apps somebody actually opens, how often, for how long, and which formats they will sit through. This is the strongest thing we own.
When a person is awake, reachable and receptive. The shape of a week, and how that shape changes when their life does.
Home geography down to postal level, the places they repeatedly return to, and how far they move in a normal month.
What they carry, what they watch on, and which other devices sit behind the same connection.
Interest read from consumption rather than from a purchased segment. Somebody who watches three hours of motorsport a week is telling you something.
What people have told us in a survey: sentiment, usage, price sensitivity, preference. The only place belief enters the file.
What we do not know deterministically
Being straight about the edges is what makes the rest credible.
We do not observe offline purchases. What somebody actually bought comes from what they told us, or from a client's own data under an integration.
Income and household composition are modelled, not observed.
Demographics start from a census foundation and are corrected by declared answers rather than measured directly.
We see behaviour in the apps and channels we transact in, and not outside them.
Content detail inside a publisher's app is limited to what that publisher chooses to pass.
Anything a person has not consented to share is simply not in the file.
Every signal we carry, by family
Where the data ends up
It has two jobs. First, it is the training data. Every model in the platform learns from these signals, which is the difference between a prediction with substance behind it and a number with a confidence interval attached. Strip the deterministic layer out and the models have no features to learn from and no accuracy to offer.
Second, it builds audiences. The same signals assemble the segments in our syndicated taxonomy, available off the shelf in the platforms buyers already use, alongside the survey-built audiences a study produces. One signal set, two outputs, both resolving back to the same profiles.
Deterministic sources rather than modelled panels or inferred cohorts.
Refreshed continuously, so a profile reflects last week and not last year.
50+ variables are available to any single model build.
Every element resolves to the same 2.6B profiles the graph holds.
Feeds the syndicated taxonomy as well as bespoke, survey-built audiences.
Client first party data can be onboarded here under the integration.
The models get the credit. The data is what makes them true. Without this layer there is nothing for a model to learn from, and nothing underneath a score except an assumption.
Direct access to the environments where people are actually paying attention.
Attention has never been harder to buy. A person's media day is split across dozens of apps, feeds, screens and services, and almost all of it is either skippable, muted or half watched. Getting a real question in front of a real person, and getting it answered, is a media problem before it is a research problem.
Mobile apps are the exception worth building on. Somebody who has opened an app has chosen to be there, is looking directly at a single screen, and is in a frame of mind where a short interruption is tolerated rather than resented. That is why a survey with nothing offered in return still gets answered here, at between five and fifteen percent, when the same question sent to an inbox would be ignored.
Owning the whole lifecycle
Direct SDK inventory through Chartboost, acquired in 2025, puts us inside thousands of popular apps rather than buying our way in through a chain of intermediaries. That changes the economics of a study, not just the experience of it.
The budget goes into recruiting people and rewarding them for staying. Reach is capped by the roster, refreshed slowly, and the people on it know they are being studied.
A newer competitor can serve a question in an ad unit, but the placement is rented. Every hop in the chain takes a margin, and the attention on offer is whatever the intermediaries left behind.
We own the SDK, the publisher relationship and the placement. There is one hop between a research budget and a person, which is where the efficiency comes from.
Because we own the request, the placement, the unit and the response, there is nobody in between taking a cut and nobody deciding on our behalf which impressions we get to see. Every point of efficiency compounds: a research dollar buys more impressions, more impressions in lean-in environments produce more completed responses, and more responses produce a faster, deeper read for the same money.
Why an exchange is a moat
There is no shortage of ad exchanges. There is a shortage of intelligence platforms that own one. Building this took years and thousands of individual publisher relationships, each negotiated, integrated and maintained one at a time, and that network is what the whole thing runs on. It cannot be licensed, bought off a shelf or stood up in a quarter.
A platform trying to hold a two way conversation with consumers globally, at scale, needs somewhere to hold it. Owning the room is what makes the conversation possible, and it is why the survey layer, the bidder and the data layer all have something real underneath them rather than a rented pipe.
Direct SDK integration rather than a resold supply chain.
Thousands of popular apps across gaming, entertainment and utility.
Full screen, high attention placements rather than banner real estate.
Owning the supply means no incentive is needed to get an answer.
The same supply serves the campaign and the survey, so exposure and response sit together.
A transparent path from request to impression, with no hidden hops.
Every other layer assumes we can reach somebody. This is the layer that makes that assumption true, and it is the hardest one to copy because it was built one publisher at a time.
A fully owned, vertically integrated technology stack
Click any layer for how it works, what it runs on and what it delivers
Every layer of the stack operates in every country on the map. The exchange serves, the bidder buys, the graph resolves, the data refreshes, the models train and the surveys field in all 117, so measurement, optimized audiences and consumer insights are available anywhere you do business. Depth of coverage shapes how fast and how granular the reads get, never whether the capability is there.
Unshaded countries sit outside the footprint today.
Positions we hold, argued in public. The ideas behind AMP, written for the industry that measures it.
Why the biggest questions in marketing cannot be answered with behavioral data, and what answering them at scale takes.
Every dataset eventually gets a second job on the shelf. Survey data is about to get its first.
Seed and spread is a blunt instrument. Scaling should look like a behavioral signature.
The data shelf stocks everything people did and nothing people think. That aisle is opening.
When asking is as fast as advertising, insight stops being a quarterly event.
Real-time brand lift on any campaign, any channel, any country. Surveys delivered as media, exposed versus matched control, and a number you can steer by while the flight is still live.
Surveys delivered as media to the people your campaign exposed and to a matched control group, on any channel in any country. The difference between the two is your lift, read while the flight is still running rather than after it.
One method, every channel, every country
In flight, not post mortem
Granular to channel, partner, audience and creative
No holdouts, no panels, no unmeasurable campaigns. One hundred percent of your impressions do the work of the campaign, and reach is never sacrificed to measure it.
There is no channel on a global media plan we cannot measure. What changes from one to the next is only how exposure reaches us, and that splits the landscape in two: the channels that hand over deterministic exposure, and the channels that keep it to themselves. Both are measurable. One of the two needs the integration.
Deterministic exposure, nothing to integrate
A pixel, an exposure file or a play log hands us the exposure directly. This is most of the plan, and it runs from day one.
Trafficked on the measured placements, delivered same day, carrying macros so the read breaks out by tactic, creative and placement.
Where a pixel is not an option, deterministic exposure arrives as a file or a feed on an agreed schedule. Same measurement, different plumbing.
Play logs joined to deterministic location data give real exposure for digital and classic placements. Three weeks of active flight where programmatic out of home is on the plan.
Automatic content recognition data identifies the households that actually saw the spot. It takes more work on our side, and it puts linear in the same chart as everything else.
The walled gardens, once your platform is connected to ours
The problem: the biggest platforms never tell you who saw your ad. No pixel, no exposure file, nothing. So the largest line on most plans is the one line nobody can put a brand number against, and it sits outside every comparison.
How we solve it: we match your customer list against our profiles, and you activate that matched group as the target audience on the platform. We know exactly who is in it, so we survey those people and ask whether they saw the campaign. Their answers give us the exposed group the platform would not hand over, and the same matched control does the rest. The result lands in the same chart as everything else.
A hold-in audience you activate on the platform, plus a short screener of the recognizable media on the plan, establish a validated opportunity to see. Every partner then sits in one relative view instead of sitting outside the read entirely.
Between the two, that is every channel where advertising money is spent anywhere in the world, read on one method against one baseline.
Illustrative dashboard views, refreshed throughout the flight.
Indexed against positive responders.
Every bet you make, ranked by brand outcome
Every read carries its own matched control, built the same way, so every bet lands on the same baseline while the campaign is live.
You bring the campaign and the media plan. We run the study: survey design, control construction, fielding, analysis and the dashboard. One market or twenty, fully managed, live in days.
Campaign delivery on any channel.
Pixel integration or partner exposure files. Identifiers: MAIDs for mobile and OOH, IP addresses for CTV, desktop and audio.
Raw exposure data.
Raw exposure data.
MAID exposures match 1:1 to profiles. IP exposures resolve the household IP through the graph to member devices. A seven-day active look-back keeps the segment dynamic.
The exposed segment.
The exposed segment plus the full graph.
RISA fingerprints the exposed segment’s composition and assembles a statistically identical, non-exposed control, mirrored in real time as the campaign evolves. No media holdouts.
A persistent matched control segment.
Exposed and control segments.
The bidder wins normal ad auctions and serves a survey instead of an ad: full-screen, in-app, five questions max, no incentive.
Survey responses tagged exposed or control.
Survey responses.
Exposed positive rate versus control positive rate. Absolute lift in points, relative lift in percent. A z-test at 95 percent confidence.
Brand lift results, live in the dashboard.
Advertiser, objectives, flight dates, markets and the KPI that matters.
Trafficked on the measured placements, provided the same day, or delivered as a file where a partner holds its own exposure data.
One agreement covering pixel-based collection. That is the whole dependency list.
Two weeks is the minimum active flight on pixel-measured channels. Three where programmatic out of home is on the plan.
A real set, in the moment an ad would have run.
A real set, from a business technology study.
One question sorts the audience, so every read can be cut by it afterwards. Three carry the KPIs chosen against the business objective rather than a standard ladder. The fifth answers whatever the team actually needed to know this quarter.
The KPIs you can choose from
Three of these go into any one study, and this is not the whole list. Which three you pick is the decision that matters, because a set dialled into the business objective tells you something a standard funnel never will. If the metric you need is not here, we will write it.
AwarenessAd recallBrand recallMessage associationMessage recallBrand attributesBrand perceptionFavourabilityTrustRelevanceConsiderationPreferencePurchase intentIntent to visitIntent to switchLikelihood to recommendCategory usagePurchase behaviourSearch intentSubscription intent
Three questions the advertiser answers first
We write the survey, not you. Answer these three and a recommended set comes back for review, usually the same day. Adopt it, change it, or replace it with your own. Nothing fields without approval.
The real people the advertiser wants to move, not just a demo. Small business owners considering a switch in the next twelve months, rather than adults 25 to 54.
Becomes the audience question.
What outcome is the organization pursuing, commercial or otherwise? If the campaign works, what will people think, feel or do that they do not today? Answered in plain language.
Drives which three KPIs are measured.
Their learning agenda about this audience.
Becomes the final question and shapes the insights in the report.
Non-incentivized, delivered on our own inventory, within seven days of exposure, and one person is surveyed once however many channels reached them.
Every study is feasibility-checked before it is scoped. Three levels, separated by the response base they can support and the granularity that base allows.
Headline lift for the campaign, with limited breakouts. At 200 responses, cutting the data six ways leaves cells too small to hold up.
Every market where we operate, all 117 countries.
Full campaign reporting with breakouts by partner, placement, creative, format and audience, and enough base to steer the plan in flight.
United States, United Kingdom, Canada, Korea, Australia, Chile, Singapore, United Arab Emirates, New Zealand, Mexico, Saudi Arabia, Japan, Argentina, Qatar, Kuwait, Netherlands, Israel, Hong Kong, Oman, Bahrain, Malaysia.
Cross-tab demographics, unlimited macro breakouts, frequency and multi-touch analysis, and an activatable campaign audience built from positive respondents.
United States, United Kingdom, Canada, Korea, Australia, Chile.
The one million impression minimum applies to a light study and holds in every market. Standard and advanced studies carry higher minimums that vary by market, confirmed on the feasibility check before anything is committed.
Setup runs in days, not weeks. The read starts arriving while the campaign is still in market, which is the whole point of measuring this way.
Brief received: campaign details, media plan, target audience and KPIs.
Survey drafted for your approval, and the data feeds agreed: pixels for digital and CTV placements, file formats for out of home.
Campaign live. Exposure data begins arriving and the exposed pool builds daily.
First results visible in the dashboard, then updated in near real time.
The bulk of surveying lands here, as the exposed pool reaches full size.
Surveying finalizes with the flight, and can run on for up to seven days where the response target calls for it.
Optional post-campaign analysis delivered, where it has been scoped.
A dedicated account manager runs the study throughout. Where data science flags one for additional statistical rebalancing, final dashboard results generally take around three extra days.
Measurement tells you what happened. Optimization is what you do with it, while the campaign is still running.
The surveys running on your campaign produce a brand outcome signal nobody else has: real people telling you what the media did to them, in flight. We use that signal to train models, and the models produce two things you can act on. Brand optimized audiences, so the next impression goes to somebody more likely to move. And an inventory score, so the next dollar goes to inventory more likely to move them.
The point is efficiency. The same budget, pointed at the people and the placements your own survey results say are working, so the brand dollar works harder against the goal that actually matters to the business.
Standard on every study and live on the first campaign, because the model behind it is already trained on our own data.
An optional module, switched on once enough measured campaigns exist to train it on the inventory you actually buy.
Both are aimed at the stated brand outcome rather than a proxy for it, both work while budgets can still move, and both get sharper with every study you run.
Alongside the brand questions we field a brandless one, about the behaviour or the attitude you actually care about. Everyone who answers it the right way becomes a seed, and the model scores every profile against that seed. It is ready immediately, because the model behind it is already trained on our own data.
The target is a stated outcome rather than a proxy. Not clicks, not completion rate, not a signal that happens to correlate on a good week.
Move across the dial to preview. Click to set.
Pushed to the platform you already buy on as a standard segment, refreshed daily for the length of the flight. Nothing about your stack has to change.
Delivered back to your environment as a file or a feed, on your identifiers, for you to activate, model against or keep however you want.
The same responses that measure the campaign build the audience for it. No integration, no separate media buy, no waiting for the post-campaign read.
Every impression arrives with a bid request full of context: the app, the placement, the format, the hour, the market, the connection. None of it identifies anybody. The model reads that context and scores how likely the impression is to move your brand outcome.
It needs a critical mass of measured campaigns first, because your inventory universe is not ours. The model has to see enough of what you actually buy, scored against what your surveys came back with, before it can grade it.
Scored 0.0 to 1.0, pre bid as a filter or post bid as measurement on one hundred percent of impressions.
Why this scores better than a generic quality metric
Trained on declared brand outcomes tied to real impressions, not clicks or proxies for exposure quality.
Not lookalike modeling: supervised prediction, trained on the answers to your question.
Works pre bid, like attention or viewability metrics, but pointed at your outcome instead of a generic one.
The brand optimized audience is standard and live on the first campaign. The inventory score is an optional module: around ten measured campaigns train it on your own inventory, and from there it covers all activity in trained markets at a fraction of survey cost.
The same read that grades the campaign teaches the model which impressions were worth buying, and you spend the next budget on those.
One live crosswalk between your identity platform and ours. Everything above keeps running underneath it, and four things become possible that were not before.
The crosswalk matches your identity spine against ours. What matters is not either spine on its own, it is the overlap: the people we can both see. Every integrated capability runs inside it, and these are the things it opens up.
The overlap becomes a hold-in audience you activate on any platform, including the ones that release no exposure data of their own. Paired with a short screener of the recognizable media on the plan, that is what puts every partner into one comparable read.
Our deterministic signal layers onto records you already own: what those people use, what they watch, where they go. The same profiles you have always had, with the half of the picture you have never had attached.
Ask your own first and third party segments anything, and read the answer in days rather than quarters. Not a panel you rent and not a proxy for your customers. Your customers, on demand.
Model new segments from what you just learned, inside the footprint of the overlap. Everything built there is addressable by definition, so there is no match rate to negotiate.
One integration, and the two files stop being separate. Yours says what people did. Ours says who they are and what they think.
All of it standing on one integration
One source of truth for your brand metric across every campaign, channel and country, plus two scores from your own measurement: an audience score for every profile and an inventory score for every impression.
Segments built from what people declare, scaled across 2.6B profiles, delivered where you already buy.
The big questions answered at the speed of media, on your own customers and beyond.
Your first party data tells you what people did. The integration lets us keep telling you what they think.
That is the whole of it. Everything on this page is machinery in service of those two moves, and the reason they can happen at all is that the asking and the buying run on the same rails.
A brand number on every channel where you spend, in every market you run, read while the flight is still live rather than after it.
No holdout, no panel, no channel left out of the comparison, and one method so the numbers mean the same thing everywhere.
The same answers train the audiences you activate and the score that grades your inventory, so the read changes the campaign instead of grading it.
Connect your platform to ours and the same instrument reaches the walled gardens, your own customers, and the questions you have not asked yet.
Brand measurement that pays for itself, because it makes the next impression better than the last one.
Consumer research fielded as media: custom surveys, brand tracking, segmentation, category studies, graph enrichment. Real people, real answers, at the speed and scale of advertising.
The survey rides the same pipes as an ad: full-screen, in real media moments, answered by real people in 117 countries in any local language. Research embedded in the media and data you already use, not a project you send away.
Deterministic data cannot answer any of them. Asking can. These are the questions that used to take a committee, a vendor and a quarter of waiting: now they move at the speed of advertising, embedded in the data you already use.
?Who are we
?Who are our customers
?What do we mean to them
?What should we build next
Panels ask a small sample and model the rest. We ask ten times more, from an audience of thousands to national scale across 2.6B profiles.
Surveys that travel like ads reach ten times the people a panel can.
Answers from the places panels cannot go: in-app audiences, and markets where panels do not exist.
Non-incentivized answers, captured in real media moments rather than a paid panel session.
Results return as data you can activate: segments, enriched graphs, addressable audiences.
One loop runs every study: define, field, join, read, activate. Declared answers land next to observed behavior, so the results are ready to use the moment they arrive.
Define the question and the audience: general population, your first party data through the crosswalk, or any onboarded audience.
The survey fields as full-screen units in ad breaks: 800K surveys a day globally, one survey per country, in the local language.
Respondents resolve to the identity graph, so declared answers join the behavioral context the platform already observes.
Results read live as responses arrive. Cut them by any attribute the graph carries.
Outputs land three ways, all of them usable the day they arrive.
What you learn here does not sit in a deck. Insights seed segments, segments become audiences, audiences get measured and optimized: three capabilities sharing one integration and one continuous loop.