Traditional research can be slow, expensive and hard to scale but agencies are increasingly testing AI-generated personas to supplement things like media planning and concept testing.
In theory, the pitch makes sense: Synthetic audiences can model real consumer behavior, pressure-test ideas and uncover insights without the long wait times that come with human focus groups. In reality, it’s a harder sell.
Synthetic audiences are only as good as the data they’re built on. Even if a project needed to launch right away, a fully synthetic data set would only take the project 80% over the line, leaving the remaining 20% to be vetted by real humans, said Andrea Berki-Nnuji, senior vp of data analytics at Crowley Webb, an advertising, PR, and digital marketing agency that has worked with clients like Evergreen Health, Niagara University and M&T Bank.
The agency has used synthetic audiences on behalf of some of its clients over the last 18 months layering synthetic data on top of human insights and social listening data.
Digiday recently caught up with Berki-Nnuji about how the agency is using synthetic audiences, how it tests and vets synthetic data tools, and how AI changes the agency’s pricing policies.
This interview has been lightly edited for clarity.
What marketing decisions have clients made after using these synthetic personas? #
Concept testing, we did naming, we did outreach, media message testing. Pricing, even, we are working with — it’s still a confidential project, but [one client] that is going to open and they had no idea the interest or how much would people [would be] willing to pay or would they want to have a membership versus what the appetite was. We created these personas based on family or single, or based on interest and we were able to get the price right.
How did you validate that your synthetic audience approach actually works? #
When it comes to syndicated data, that’s the large-scale data. We use MRI [MRI-Simmons, a syndicated consumer intelligence database used to power and train synthetic models], and I already trust that the data getting out of that tool is reliable. So mixing these three types of data points and using less of primary, we can create these synthetic personas.
When you do traditional research, it’s something that you have that data at a given point of a time. So we have done lots of segmentation, research, and perception and brand [tracking] over the years. So, of course, then we started using synthetic. We are, I would say, traditional researchers here at heart and we were able to paint those personas that we created using SPSS [Statistical Package for the Social Sciences, an IBM software program used for data management and statistical analysis], and using statistical modeling and week’s worth of statistical analysis. We took a portion of those real human feedback, mixed it with social listening, and syndicated data, and the results were almost identical.
How do you decide when to use synthetic audiences versus traditional consumer research? #
We had one of our clients, they already had personas built. It was in a travel segment, and they have been using those [personas]. They didn’t want to invest any new dollars into it, but they wanted to do a marketing campaign targeting one of the personas among the four [audience personas]. They came to us, and I looked at those personas, and I was like, “This is great, but research was done prior to COVID.” We know that — especially when it comes to travel — everything changed. People travel differently. In their case, I recommend refreshing and making sure that we validated. We recruited just a small number of consumers, and we used the [synthetic data AI] tool, and actually super similar personas came out of the tool, but there was an additional persona added that the client would have never thought of. That made perfect sense when we looked at it. This was the persona that changed behaviors since COVID.
What can marketers actually do with synthetic personas once they’re built? #
You have these personas, they are living in this tool, and you can always go back and talk to them. You can do concept testing. That’s the most powerful thing about synthetic personas, that they’re going to live in your tool, and you are able to communicate with them.
**Are there decisions that clients insist be validated by a human or be relegated to a human as opposed to a synthetic audience? **
We’ve worked with lots of pharmas [pharmaceutical companies] and sometimes there’s just no AI [usage allowed]. For certain ailments, there are such rare diseases, you really need to find those people that are suffering from that ailment. There’s always going to be some exceptions.
What factors mattered most when choosing a synthetic audience platform? #
I want to make sure that any analyst can pick up a project and it makes sense. Also, saving your things in the tool, having multiple users without extra cost… I don’t have to pay extra once the personas are created or the synthetic users. Other tools were limited, and some of them were outrageously priced.
When it’s a new business type of thing, everybody wants to create their own AI playbook and everybody wants to come up with these tools. I have to do investigating about the company itself. When did they start doing it, what is their experience? Where do they get their data from? We have a compliance department here at the agency, and they did help with reviewing them. I ended up with three choices, the three final choices. That’s when I went to the compliance department and they made sure that these companies [are viable].
How are you packaging synthetic audience research for clients? Is it now a standard offering or an add-on service? #
At this point, I let the client decide and say, “Yes, we can do the traditional research and this is how many hours.” We usually say, “[Hours] calls the budget,” because each client has [a] different hourly rate — or, “We can do this for you, and then this is the benefit. If you use this synthetic and at the end of the day, you are going to save time and money, and then you will have this living persona.”
Everybody chooses that [hybrid] option because they see that they still recruit humans. Right now, I feel like I would have [a] hard time selling option three [fully synthetic audiences].
More in Marketing
[
How Bloom Nutrition is expanding internationally ](https://digiday.com/marketing/how-bloom-nutrition-is-expanding-internationally/)
Bloom Nutrition is launching in multiple foreign markets that include Australia, France and the U.K.
[
Fears of Big Tech bias underpin debates around W3C’s Attribution API ](https://digiday.com/marketing/fears-of-big-tech-bias-underpin-debates-around-w3cs-attribution-api/)
Critics may label the initiative as ‘Privacy sandbox 2.0,’ but advocates counter with privacy arguments.
[
Nike versus Adidas: Which brand won the World Cup? ](https://digiday.com/marketing/nike-versus-adidas-which-brand-won-the-world-cup/)
We sift through data from AdClarity, Meltwater, Sprout Social and MediaRadar to work out which brand had the best World Cup performance.