Kyle Fohlin, VP of New Business Development for Wiland; Emma Nicoletti, VP of Predictive Client Solutions Department for Wiland; and Stephanie Taylor, SVP of Acquisitions for Belardi Wong, discusses the science behind data models and some bonus aspects to working with co-ops

As discussed in part one of this two-part article, not all data and the way it’s acquired is the same. First- and third-party data are only part of the picture; data collaborations and co-ops are crucial to acquiring new customers and keeping them.

Through acquisition modeling — predictive models, which are used to acquire new customers or consumers to the brand — businesses can reach more customers than ever before. But what does the science behind modeling look like? And how can customers handle issues like privacy compliance, customer reactivation and new product expansion when working collaboratively?

The process of acquisition modeling

In the collaborative data paper that Winterberry published, they surveyed close to 8,000 businesses and found 59.5% of them were already using co-ops and 26.2% plan to in the future. This means 85.7% of businesses surveyed are planning to get into a collaborative data situation in the future.

“Depending on what your business is and what the metrics are, co-ops can tell you — based on what else the [customer] is doing — that these are not good buyers for you to model or try to reactivate or continue to send your offer,” says Taylor.

Using a data co-op is just one way subscription businesses can expand their customer base and their bottom line.

In the same study, Winterberry found that for licensing partner to partner, where two partners share data on their own, 64.6% are currently doing that and 14.3% plan to in the future.

“It’s a good 80% of marketers out there that are capitalizing and using this to rapidly grow [their] businesses,” says Fohlin.

The precedent for using data and collaborating with partners exists. Actually, building models is the next step in the journey. Companies like Wiland, where  Kyle Fohlin, VP of New Business Development, and Emma Nicoletti, VP of Predictive Client Solutions Department, work are one of the simplest ways for businesses to get involved.

“For me, being in sales, that comes down to getting an agreement in and working with the people to discuss the kind of data they have. But the next thing that really makes it exciting for me is setting up the team to have a kick-off call,” says Fohlin.

A kickoff call includes the business development team, the client, and predictive sciences. And it’s during the kick-off call when a lot of interesting information is discovered. A client may have 100,000 people in their membership program and share that they’d like to grow that aspect of the business, for example.

“We talk to clients first about what they know about their buyers. The brand knows a lot about their buyers from their perspective, but they don’t know what else is happening across the collaborative database. We hear from them who their buyers are, who they believe their comps are and, once the data is onboarded, we get reports back,” says Stephanie Taylor, SVP of Acquisitions for Belardi Wong.

All of this is discussed in the kick-off calls after the co-op has onboarded the data. The panel agreed that clients tend to be pleasantly surprised because they find out their universe is much larger than they thought. This data allows clients to do different segmenting around different offers they’ve considered in the past. But now they have data on their buyers to know whether or not they would purchase based on the offer.

The science of acquisition modeling

Once the kick-off call takes place, all of the data and desired outcomes are handed over to the predictive model team.

“We have eight primary categories of models, which either sounds like a lot or a little, depending on which direction you’re coming from. The big thing we have to decide is what does [the customer] actually want? For subscriptions, it’s almost always a higher response rate. So we have models that are designed to specifically say, ‘This person responded five times, they’re very slightly better than this person who responded four times, but they’re both better than this person who responded once,’” says Nicoletti

Wiland uses a variety of models, including the extreme gradient boost model and the extreme elastic net regression model. These models all fall into the bucket of machine learning. Once the data from a large pool of businesses is input into the model, it immediately offers thousands of variables. The machine learning models are very good at sorting through the variables and looking for interactions between them.

Sometimes, because of human behavior, some of the strongest variables make intuitive sense — like pet magazine subscribers also being interested in a pet food box subscription. However, sometimes the variables and their connections make less sense.

“If you’re a beauty box subscriber, there might be things that the machine learning finds that wouldn’t really occur to a person trying to predict buying behavior on their own. It could be something like a specific brand of shoes. It makes sense when you see it, but you never would have thought of it,” says Nicoletti.

The algorithms pull all of the variables together and identify the traits all of the best customers have in common. That algorithm is then applied to the entire U.S. The algorithm looks for other people in the U.S. that look like the best customers, from people familiar with the brand to people who aren’t, and then those customers get sent the name of the company.

“I’ve been on some of those model calls where we’re going through variables and one of those quirky things comes up and everybody is like, ‘This makes no sense.’ And then you talk through it and you’re like, ‘Well yeah, that does seem to fit our buyer,” says Fohlin.

Going through the data provides the ah-ha moments most companies need to understand to reach a wider audience.

“We can also do the opposite. We can say, ‘This is what all of your worst customers have in common,’…so the models work both positively and negatively,” says Nicoletti.

Sometimes, in running the algorithms, there are several models that yield strong results.

“If you’ve identified three that you think are really strong, of those three, how would you rank them and how would you encourage the client to test?” asked Taylor.

The answer to this question depends a lot on depth for most clients. But most collaborative databases, including Wiland, have models where they combine the models into a multi-model solution. A “multi” is where two collaborative databases say the same thing. A “multi-model” is where two or more models say the same thing. When combined, these models offer trickle-down sorting which can create interesting results.

“For any [company] who is big enough — which is a house file of 5,000 or more — we build at least three and combine them,” says Nicoletti.

This combination allows companies to home in on the absolute ideal customer, based on their goals.

Privacy can’t take a back seat

Though working with models and seeing what they can do to increase a customer base is exciting, when utilizing collaborative data, it’s crucial to understand privacy. Working with a legal team ensures the actions a company takes are not only legal but transparent. Gaining consent is only one aspect of privacy compliance; notifying customers and other collaborators of the use of third-party data and working in the collaborative data space is also important. Lastly, offering clear ways for buyers to opt-out that data usage allows companies to remain compliant with federal and international regulations.

“We’ve been through GDPR, we’ve been through CCPA, and we have other states coming online [like] Virginia and Colorado. Most of the co-ops have put out privacy statements or governance, and we’ve been through the compliance for these states as they come on board. We think we have it figured out, but we’ll see as we go,” says Fohlin.

Until there is a federal statute that oversees everything, it’s an active process that must be monitored and updated to fit the current climate.

“If you go back and read your privacy policy and think, ‘Uh-oh, maybe this isn’t privacy-compliant,’ don’t panic. Just fix it. Then make a date of when you fixed it because you want to make sure the data — if you’re contributing to one of these collaborative databases — is compliant,” points out Taylor.

If a privacy policy becomes non-compliant on November 2, for example, it’s important to share that update with partners because co-ops don’t want non-compliant data. This doesn’t have to be a cataclysmic event. Instead, staying on top of compliance by updating it with the help of a legal team and noting the date of the change is a simple way to remain compliant.

Reactivation: a secret bonus of data co-ops

There are times when a customer will simply up and leave. If the company has done its due diligence around helping customers increase their own lifetime value and minimizing involuntary churn, then turning to co-ops is a great next step.

“Co-ops can help with the reactivation process,” says Fohlin.

It’s a similar process to first acquiring customers using data modeling.

“It takes into account both what they did with you before they [left] and what they’re doing elsewhere so if they died, we’ll be able to say, ‘This is probably not who you should be reaching out to again,’” says Nicoletti.

However, in less extreme cases, co-ops like Wiland can identify if a customer is spending everywhere else and look at the buying patterns before the customer left to see if there’s a similarity between other customers that were reactivated. Not only is the modeling process similar to acquiring new customers, but so is the deliverable: an audience of a company’s customers who are likely to return.

Expanding offerings

In many cases, companies don’t want to continue to offer the same thing over and over again forever. Instead, they want to expand their offerings in order to reach an even wider audience and increase their bottom lines. When it comes to bringing on a new product or offering, there are a few different approaches companies can take with data co-ops.

“If you understand who your competitors or even soon-to-be competitors are, we can help design a model that lives in that space. Otherwise, it’s a matter of looking at who you currently have who you think is going to be the best audience for that product and focusing on them and looking to grow that,” says Nicoletti.

If a company has a group of people who express interest in a specific product, co-ops will try to grow that group instead of trying to grow from the entire audience.

“With data partners, there’s always a great opportunity to be collaborative,” says Fohlin. “To set up a call and say, ‘We’re thinking about doing this,’ or ‘Does our data support doing this?’ and we can really dive into that and work with anybody and doing that.”

The possibilities are vast, and it only takes getting started. By tapping into a variety of co-ops and getting on a kick-off call, subscription businesses can take the first step toward reaching a wider audience of people who will not only buy from them but become loyal subscribers. In the end, if companies want to remain competitive and experience growth, they must utilize all of the data available on their customers and the customers of others.