Stop recruiting in the dark: how deep signal targeting finds exact-fit B2B and healthcare respondents.

Most sample suppliers start every project the same way. They take your screener, sweep their entire network and wait to see who qualifies. That is why recruiting drags, and it's why respondents don't always match your criteria. Deep signal targeting changes that. Emporia Research recruits from more than 3,000 signals already attached to respondent profiles, so the right people get invited and your screener only has to confirm fit.

This live session, recorded on September 23, 2026, walks through how the signal catalog is built, shows the live targeting interface and walks through three recruiting scenarios. The full recording is below, followed by key takeaways, common questions and a complete transcript with timestamps.

Key takeaways

Chapters

Each timestamp opens the recording at that moment.

• 0:00 Welcome and introduction
• 1:00 Housekeeping
• 1:54 Attributes versus signals
• 2:31 The problem: your screener is doing two jobs
• 4:02 What a signal profile looks like
• 5:28 Over 3,000 unique data signals
• 6:53 What signals are not
• 8:22 Why AI is the right tool for tagging
• 9:26 Live demo: building a signal audience
• 10:18 Buyer journey signals: user, evaluator, decision maker
• 11:44 Example 1: cybersecurity buyers
• 13:07 Example 2: high-net-worth rental property owners
• 14:53 Example 3: physicians
• 16:29 What respondents get out of it
• 18:22 The results: more accurate, more efficient recruiting
• 19:43 Roadmap: lookalike audiences and confidence settings
• 21:29 Audience Q&A
• 27:44 How to get started

Frequently asked questions

What is deep signal targeting?

Deep signal targeting is recruiting survey respondents based on behavioral signals attached to their profiles, such as the software they use, their budget authority, what products they are evaluating and what they have qualified for in past studies. Targeting happens before the screener, so the screener only has to confirm fit.

How many signals does Emporia track?

Targeting is available on 3,000 signals and new signals are added regularly. At launch, there are 1,200 different software usage signals, and each can be combined with respondents' role in the buying journey: user, evaluator and decision maker.

Are signals synthetic or AI-generated data?

No. Every signal is generated by a real person's responses and behaviors, plus enrichment data. AI then files real answers against the fixed catalog of tags.

Does deep signal targeting replace my screener?

No. Signals decide who gets invited. Your screener still confirms every respondent exactly as it does today. The difference is that far fewer invited people terminate, so studies fill faster and respondents have a better experience because they qualify more often.

How long do signals stay valid?

The default expiry is 18 months, and it varies by signal. A durable fact like being a physician lasts longer, while time-sensitive signals can be shorter. A signal is refreshed whenever a later interaction reconfirms it.

Welcome and introduction

0:07 Ashley Lapin: We're so thrilled to have you here. Thank you for taking the time out of your day to learn how you can stop recruiting in the dark and start using deep signal targeting to find the respondents you need for your studies. For those of you who are unfamiliar with Emporia Research, we are a B2B and healthcare research platform that helps companies field studies with verified, fraud-screened professionals. I'm Ashley Lapin, head of marketing here at Emporia, and I'm getting us started today. We also have our CEO, Michael Hess, with us, and you will primarily be hearing from Andrew Blum, a senior engineering manager who led the development of the deep signal targeting we're going to dig into today.

Housekeeping

1:00 Ashley Lapin: Before we get started, a few housekeeping items. You are in listen-only mode, which keeps you muted throughout the session. You can talk with one another in the chat and submit any questions you have there throughout the session. Drop them in when they come up, and at the end we'll go through them and answer as many as we have time for. You will automatically get a recording of this session by email as soon as it's ready, so if you're multitasking, you can go back and revisit anything we cover.

1:47 Ashley Lapin: With that, I'm going to turn it over to Andrew to take you through the good stuff.

Attributes versus signals

1:54 Andrew Blum: Thanks for joining us today. Here is what we're going to cover: the way recruiting works today, which is the status quo of recruiting based on what we call attributes; what signals are, how Emporia collects them and what signals are not; several examples of signal-based audiences and how we use them to target and recruit; and the benefits for you and for the people who take your studies. At the end we'll have time for Q&A with the questions you've dropped in the chat.

The problem: your screener is doing two jobs

2:31 Andrew Blum: Right now your screener is doing two jobs. Every study starts about the same way. You write the screener, blast whatever network you're using and wait to see who qualifies. Screeners were built to confirm fit, but we use them to do targeting too. If you're doing a study about Quicken, you want to talk to accountants who use Quicken, so the screener asks, are you an accountant, and do you use Quicken? If they're not an accountant to begin with, the Quicken question is irrelevant. You invite a lot of people, most of them terminate and the study drags.

3:27 Andrew Blum: Depending on your industry, your panels and what you're doing, your numbers will look different, but the shape is the same. It's big at the top and small at the bottom. It's a funnel, and that is why fielding drags. Incidence is a guess, the terminate pile is huge, and we're using the screener to do a lot of that lifting for us. Quick question for you: what kinds of questions do you often put in your B2B screener because there's no way to target for them? Drop your answers in the chat.

3:58 Ashley Lapin: Ad spend. That's a good one.

What a signal profile looks like

4:02 Andrew Blum: Management level, fair enough. I'll let that cook while I move on. Here is an example profile of a person we have named J.R. He is a panel member and a director of IT. He has done three studies with us, and some of the signals we have about him are that he is evaluating Snowflake, he has an IT budget of one to ten million dollars, he is a final decision maker, he buys in the cybersecurity category and he loves using AI at work. This is the kind of profile we build about people. Every time we run a screener, it tells us structured facts about who they are: budget authority, current tech stack, what they're evaluating, whether they have the final say, and on and on.

4:48 Andrew Blum: These come from the screener and from behavior, so what they've told us and what they do next, including how they interact with our apps, and we're capturing it all the time. If a screener asked an unrelated or loosely related question, we capture that too, and it might be useful for targeting something later. We have thousands of these. We just launched, but we already have about 3,000 defined signals today, and they're all defined by us. These are not things AI is making up. We came up with all of them based on what we think is interesting and useful for the kinds of studies we run, and we're adding to the catalog all the time.

Over 3,000 unique data signals

5:28 Andrew Blum: To give you an idea of the breadth: software people use, what they buy, what they spend, what they're licensed to do, what they own and how they're adopting AI. A lot of our business is B2B focused, so we might have high net worth tags, an Epic user tag, whether someone is an RN, budget sizes, you name it. Here is how a signal gets attached to a person. A real person completes a real study, which is your survey on our panel, same as today, and their qualification answers become signals. The primary source is those screener questions, and the software, AI in this case, matches their answers against our fixed taxonomy, the 3,000-plus tags we have about people. The next study then invites on those signals. Everything they've told us before helps target them for future studies. The screener is still confirming fit. It's still the final gate. Signals just get better people into that screener from the start. The slide says signals expire after 18 months. That's the default, and it varies. Whether someone is a doctor lasts much longer, because we don't need to expire that in 18 months, and some signals could be shorter.

What signals are not

6:53 Andrew Blum: The question is how long we think a piece of information stays useful and relevant. After that it falls away unless it's reconfirmed by some other interaction we've had with the person. One other thing to call out: respondents are still flagged for quality problems. We have a lot of fraud technology, it's a big part of our business, and none of it was skipped for signals. That would be a whole other talk, but all of the Emporia fraud work is still running, so this is not a new avenue for attackers. Now, what signals are not. They're not synthetic respondents. These are real people who completed real studies and gave real answers. Nothing is generated. We're not guessing what their answer would be.

7:38 Andrew Blum: The attributes we use come only from the fixed list. If the AI tries to say, I think this person might be something we haven't defined, we ignore it. Signals are not a replacement for your screener. Signals are just for targeting. They decide who gets invited, and your screener still confirms every respondent exactly as it does today. We're just inviting better people. And signals are not your research data. They come from our qualification questions, participation history and other behavior, and they are not sold. We use them to target people into studies.

Why AI is the right tool for tagging

8:22 Andrew Blum: AI does the filing. It's not making up the person or the answer. People have different feelings about AI, so I'd like to make the case for why this is such a good use of it. We're taking a defined answer from a person and a fixed set of tags, and we're asking, do these match? That is something AI is very good at. The other reason it works is that AI can fail in situations where it has to be perfect all the time, and it does not need to be perfect here.

8:58 Andrew Blum: If we're tagging whether someone likes coffee, and we've tagged 100 people that we think like coffee, and two of them are completely wrong, that's fine. If we invite those 100 people to a study and two of them turn out to like tea instead, that's still a huge win compared with blasting a huge list, asking everyone whether they like coffee and terminating the people who don't. I also want to show off the interface. This is a quick demo of the UI, which is live in production right now. I'll play it and talk over it so you get a sense of what it looks like.

Live demo: building a signal audience

9:36 Andrew Blum: This is the signal tag UI. You can sort by popular or alphabetical. I'm going to search for the tag Cybersecurity Category Buyer and open it. It gives me a description, and at the bottom it shows how long the tag lasts, which is its decay rate. It also shows synonyms and related tags. I'll select it. There is and, or and not syntax at the bottom, so I can say any of, all of or none of. We'll go with all of and add another tag, Marketing Technology Category Buyer, so they have to be both. Now I'm going to add a new condition, and this one is a not.

Buyer journey signals: user, evaluator, decision maker

10:18 Andrew Blum: So they have to carry those two tags, but they must not be an Ad Platforms Category Buyer. Now we have some very specific targeting built out of signals. This next slide is, I'm told, the most interesting one in the deck. The tags get very granular. Some tags might sound like attribute-based sourcing. If a tag just says someone is a nurse, that's not new. We can target nurses today. But we can get far more granular. Take someone who uses Slack. That's not remarkable on its own, but there are different kinds of Slack users you'd want to talk to.

11:01 Andrew Blum: There's someone who just uses Slack at work, like many of us do. There's a Slack evaluator, likely more senior, who evaluated Slack against Discord, Teams and other options and gave their opinion. And there's a Slack decision maker, the person who actually decided to purchase Slack for the team. Those are three different kinds of Slack users, and we split them out for all 1,200 of our software signals. Slack, Jira, Epic, whatever it is, there's a user, an evaluator and a decision maker, and we differentiate between them.

Example 1: cybersecurity buyers

11:44 Andrew Blum: Uses Salesforce is a screener question you would ask. Has the authority to choose the CRM might be what your study is actually about, and it's exactly the kind of thing a signal tag can filter for. Let's look at a few examples, with traditional sourcing on the left and signal-based sourcing on the right. A cybersecurity vendor wants IT decision makers at mid-sized companies who have not standardized on CrowdStrike yet. That's quite the ask. If you wrote a screener for it, you'd ask a lot of questions and terminate a lot of people. On the left, everyone in the network with IT in their title and a few attributes like company size gets invited.

12:29 Andrew Blum: We invite a lot of people, and in the screener we ask about role, company size, security budget, the tools they use and whether they have buying authority. Many terminate on budget or authority, and asking whether they use CrowdStrike dumps a lot more. Ten days in field is not uncommon for us on a study like this. With signals, we go straight to the tags: they are a cybersecurity category buyer, which you saw me search for earlier, they have an IT budget of one to ten million, they are a final decision maker and they are not a CrowdStrike user. Any tag we have can be negated.

Example 2: high-net-worth rental property owners

13:07 Andrew Blum: If someone is a CrowdStrike user, we can also look for people who are not. The only people invited to this study carry all four signals. It's a much smaller set of people, and most of them pass the screener. That means a much lower termination rate and a shorter time in field. Same screener, different people coming through the door. Use case two: high net worth households, a slippery group to find. We want them for a tax software study, and specifically people who already use Intuit products and own rental property.

13:47 Andrew Blum: Today we might go to a consumer panel and filter on age and income bands, since older people generally have more money, and then in the screener ask about investments, current tax software and the property they own. People don't love answering those questions in a screener, so termination is high. With signals we go straight to high net worth, Intuit user and rental property owner. The slide also calls out a fringe benefit of signals: they're collected all the time, including in unrelated studies.

14:26 Andrew Blum: If you run a study looking for cattle farmers, you're paying a thousand dollars and the first question is, are you a cattle farmer, there is a strong incentive to say you are something of a cattle farmer. Whereas if someone mentions in a cybersecurity study that in their free time they look after the cows on their ranch, there's no incentive to lie. Most signals are collected outside those high-pressure questions, and we've found they are nice quality. Use case three: board certified physicians whose practices choose Epic.

Example 3: physicians

15:09 Andrew Blum: For anyone who doesn't know, Epic is a massive healthcare software company that a lot of hospitals run on. It's very expensive software, and you want to know who makes the decision to buy it. Specifically, you want MDs who are making that decision, which is unusual. The traditional way is a healthcare panel filtered on people who self-declare a physician title, and a lot of non-clinicians and office staff slip through.

15:47 Andrew Blum: Depending on how you word it, that's not even wrong, because a lot of Epic decision makers are administrators or in similar roles, and a lot of MDs have nothing to do with Epic and don't want to. If you're a doctor at Kaiser, you're not making software purchasing decisions and you have no interest in being asked about them. With signals you tag for both, which is rare, so you reach a small group, but the termination rate is close to zero. If they clicked on the study and they were targeted by those signals, they're exactly who you want to talk to. Terminates drop to near zero, and it works fantastically for us.

What respondents get out of it

16:29 Andrew Blum: We've been talking about the benefits for us. We love faster studies. But what about the person taking the study? Research only works if good people keep saying yes to it. Every time we waste an invitation, pull someone into a long screener and terminate them without a clear reason, we make the next study harder, because many of those people decide they're never clicking one of those again, even though they showed up with good intentions and were ready to share their expertise. If it happens too many times, they stop participating.

17:07 Andrew Blum: A good respondent today gets sent to studies they'll never qualify for, gets terminated, gets no money and often gets poor communication about it. They stop opening our emails. That was a person with good things to say who is now not going to participate. If you make sure they're only invited when they're very likely to qualify, they have a better experience, they finish more of what they start, they get more feedback, they get paid and it doesn't feel like a scam. We also have a smaller, tighter audience.

17:41 Andrew Blum: That means we can spend more time and care on each person we're trying to get into a study. In the Epic example, if you knew the 50 MDs who are Epic decision makers that you want to talk to, you could spend real resources making sure they had a good experience. If you had to message large health organizations all at once, you couldn't afford that. They have a better time, we have a better time, everybody has a better time. What this gets you in total: one, faster fielding, with fewer terminates and shorter field windows.

The results: more accurate, more efficient recruiting

18:22 Andrew Blum: The screener confirms that the right people are getting in instead of doing the targeting and discovery work. Two, the right people. Targeting is based on how they've qualified and participated over time and what we already know about them, not only what they claim in a single moment. Three, it's sharper every study. Every project adds signals to hundreds of profiles, including the people who terminate, so we know more about them and can target them more accurately next time. It's not a waste for them or for us. We're also adding tags all the time and we can backfill.

19:03 Andrew Blum: Say you want to do a study about socks, and we've never studied socks, so we don't have a sock tag. We can add socks, sock color, sock preference, sock texture and sock thickness to the taxonomy, go back through every answer, interaction and behavior we've ever collected, and tag people. Then we have it going forward in addition to every study we run in the future. As we work with new partners on new studies, the catalog gets better and better. On the roadmap, the most exciting item is lookalike audiences for hard-to-fill briefs.

Roadmap: lookalike audiences and confidence settings

19:43 Andrew Blum: I didn't get to talk about the engineering under the hood, but this is a huge knowledge graph about people. If you're having a hard time finding the right audience, you can say, I want people like this person, but I've already talked to them and there's no one left with the exact tags I want. You can find people next to them in the graph. They are similar in ways we don't necessarily understand, they're a very good place to start targeting, and they may well be that person. We just didn't have the chance to tag them.

20:23 Andrew Blum: That lets us target more broadly. Number two is custom confidence and time settings. We talked about time settings: tags expire over time. We also have a confidence score on every tag with a minimum threshold. For something like, are you a doctor, the threshold is high. We have to be quite sure before we tag someone as a doctor. For a coffee lover tag, the threshold is lower. If they mentioned coffee and might be into it, tagging them is fine.

20:59 Andrew Blum: In the future those two settings will be custom per project. You could say, I want to go wider, so lower the confidence, or I want to tighten the time window to people who evaluated Slack in the last three months. Every project you run with us makes the next one faster. With that, let's take a look at the questions.

Audience Q&A

21:29 Ashley Lapin: No questions coming through yet, so if you have them, please drop them in while we give folks a moment. Andrew, I think you mentioned it, but I missed it. With the software tags, about how many types of software can we zero in on today?

21:51 Andrew Blum: 1,200. It's 1,200 right off the bat.

21:55 Ashley Lapin: That's such a game changer. I'm also curious about the user, evaluator and decision maker tags. When you talk about the time settings and how signals get updated, if someone moves from evaluator to decision maker as they move up in their career, is that something we can see and start to target on?

22:30 Andrew Blum: Yes. One, things naturally fall off over time. Two, some tags have exclusivity with each other. You can't be an intern and a CEO, so the system has to pick one when it tags. That takes care of some of those cases. Some people are simply both, like a software engineer who is also a real estate agent, and I know several people who have done that, in which case they carry both tags.

22:59 Ashley Lapin: We have a question from Kristen. It seems like the more studies a respondent completes, the more signals we have about them. That's true. How do we prevent a respondent from becoming a professional respondent, and do we restrict the number of studies someone can participate in per quarter?

23:23 Michael Hess: I can take the first part of that, and Andrew can add anything. To Andrew's example that you can't be tagged as an intern and a CEO within the same company, we're capturing a lot of information that could be contradictory. If you take a nurse practitioner study and then somehow navigate into a study for accountants, that raises a flag in our system, and we review those cases. We want to make sure that is not happening and that people are not becoming professional respondents. On the limit of studies per quarter, that is something we can set if it's a client requirement. The short answer is yes, and we don't enforce it on all studies.

24:13 Michael Hess: Keep in mind the nature of B2B work. The total addressable universe for a lot of these roles is much more limited, so if someone is the perfect fit, they may be open to participating in multiple studies in a quarter. If a client requires that they can't, we can make sure of that. Great question, Kristen.

24:41 Ashley Lapin: What are some of the most surprising signals our tagging system has captured?

24:56 Andrew Blum: I'll be arrogant and say I wasn't surprised by any of them, because we came up with all of them. What has surprised me is a system we have internally that suggests tags we don't have, which we review and decide whether to add. That is all human review, and it produces a lot I hadn't thought of, mostly jobs I didn't imagine existing. Military ordnance maintainer came up yesterday. It's a real job. I didn't think of that tag, and I'm probably not going to add it.

25:41 Ashley Lapin: As we've talked about this over the last few weeks, a couple have stood out to me. Andrew mentioned rental property ownership in one of his examples, and the level of investable assets people have, which has made it easy for us to target high net worth individuals. We're squarely focused on the B2B and healthcare world, but that turns out to be an audience we can pull into a study relatively easily. And someone mentioned ad spend early on. Being able to target on different levels of ad spend is exciting if you're doing research with marketers.

26:37 Michael Hess: Andrew, a question that has come up in past conversations. Are signals limited to what is collected from screeners, or can we take a file mapped to participants, or pull in other data sources from data partners, to enrich these profiles and use them as signals? How might that work?

27:10 Andrew Blum: It's definitely not limited to screeners, and we are pulling from other sources. We re-enrich our contacts from a number of partners and use that information. They return varying kinds of data, so it isn't standard between them. Some return lists of hobbies, for instance, and others don't. All of that is read in, along with people's interactions with our applications. We use everything we can to find the right people.

How to get started

27:44 Ashley Lapin: It doesn't look like we have any more questions. If you have a study coming up where you could benefit from deep signal targeting, please let us know. It's easy to reach us on our contact page, and we'd love to help you get studies started with fewer terminations during screening, better and more accurate respondents, and a much faster path into field. Thanks for taking the time to join us today and learn more about deep signal targeting. If you have questions after the fact, please don't hesitate to reach out. We'll be sure to get you the recording very soon.

28:43 Ashley Lapin: Thank you all so much. Hope to see you at the next one.

28:47 Michael Hess: Thanks, everyone.