This is the third and final installment of our blog series for health marketers that focuses on the world of data in an era of rapid digital transformation. Before diving in, you may want a refresher on parts 1 and 2 of this series.
Consumer data is big business, with some estimates valuing it at $200 billion. If you’re a health marketer, you’re probably buying or working with some kind of consumer data. But NAI and DAA restrict the use of personally identifiable health information (PII), except in cases where you have consumer consent. Given how difficult it is to get informed consent from large volumes of people, most consumer data that we buy is modeled data.
Because of how actively we’ve been using modeled datasets, they’ve gotten pretty sophisticated. But caution: not all audience models are the same. The next time you’re looking to buy condition audiences or other health segments, consider the following:
- Source of the seed data
- Modeled data attributes
- Targeted consumer audience
- Seed output size
- The balance between scale and accuracy
- Model maturity
Source of the seed data
Models are developed based on seed data. They study a group of ‘seed’ people, then project out their observed attributes and behaviors to identify similar people in the larger population. It follows that the bigger and more robust the seed data, the more reliable the model. So make sure your seed data source is coming from national organizations or initiatives.
Data suppliers use multiple seed sources to get to opt-in health data, including surveys, coupons, offers, etc. All sources have their positives and negatives. Take surveys, for example. With surveys, the supplier gets opt-in consent, as well as behavioral and attitudinal insights into the survey respondent. On the flip side, these are self-reported so probably not completely reliable.
When purchasing data, make sure you understand how robust the seed data is and where it’s coming from. Think about how accurate the seed data is, and how much it reveals about your audience.
Modeled consumer data attributes
Some data suppliers use medical claims data to seed their databases. These seed data are extremely robust and include real historical diagnosis, treatment and comorbidity information.
But since we cannot target using healthcare data, they model out this data to find other proxy targeting attributes like gender, age and zip code. So, if, for example, the model finds that high income women have a higher propensity to be diagnosed with Alzheimer’s Disease (AD), then an AD drug using this model would end up targeting high income women with their ads. Clearly this approach yields innumerable false positives and false negatives.
It’s important to understand how the model you’re buying works. Ask what the data is modeled against, and what it’s modeled towards. Ask about false positives and negatives, and get a sense of model efficiency.
Targeted consumer audience
Does the data you’re buying match the audience you want to reach?
Take contextual data for example. Let’s say the segment you’re interested in includes people who’ve consumed diabetes content. They are most likely to be motivated content seekers: people who are symptomatic but not yet (or newly) diagnosed, people who’re having an issue or complication, or caregivers. Conversely, people who are living with diabetes long-term are probably not actively consuming content.
If you’re looking for the broader diabetes population, contextual data is unlikely to get you there. But if you’re a drug or a device and are looking for the newly diagnosed, you can be confident that contextual data will get you to the right people. When evaluating an audience segment, try to get to the bottom of who is in the segment so you can evaluate if they’re your audience.
Consumer data seed output size
The more you model out the seed data, the less accurate the output is going to be.
Here’s how models work. Let’s say you start with migraine sufferers in your seed data. You then use machine learning to understand the various attributes shared by these people. You create the model based on these attributes, then use the model to go find people who look like your seed people.
The first group of people identified are going to have extreme fidelity to the model. But this decreases as you bring more and even more people into the segment. At really high numbers, the model can get so diluted that it’s worthless.
When working with data vendors, find out their seed to output ratio. If they’re using, say, a seed group of 1000 people to create a segment of 10 million people, you know you have a problem.
The balance between scale and accuracy
There’s a constant tradeoff in the world of data between scale and accuracy. As discussed in the seed output size consideration, the bigger the scale, the lower the accuracy. When buying data, you need to decide where you want to land on that spectrum, based on your objective.
If you’re launching, say, a disease awareness campaign, you want your message broadcast far and wide. In that case, opt for scale. However, if you’re implementing a bottom-of-the-funnel conversion campaign, you want accuracy to ensure an efficient time-to- and cost-per- conversion.
Consumer data model maturity
Maturity is either a good or bad thing, depending on how the model works. Some models are static and get stale. Others, that rely on machine learning, get smarter with time. When making your decision about an audience segment, find out how old the model it’s using is. Then find out if the model is blooming or atrophying with age.
In addition to these considerations, make sure you’re familiar with the NAI guidance, so that you’re keeping your organization in compliance.
Asking the right questions of your consumer data supplier is half the battle
Before you sign the contract for your data, you should make sure to ask the right questions about both the supplier and how they are procuring their data. Not only will this help ensure you remain in compliance (by using data that was collected according to NAI guidance), but that your marketing efforts will have a higher chance of success. Consider asking the following:
- What’s the source of your seed data?
- What targeting attributes does your model use?
- What target does the data cover?
- What’s the seed to output size ratio?
- Where do you fall on the scale vs. accuracy spectrum?
- How old is your model and is it getting smarter with time?
At PulsePoint, we use real-world data in real-time to optimize campaign performance and drive ROI. To learn more about our data and technology solutions, and how they can support your business and its goals, contact us to request a demo.