As Rory Southerland cleverly points out, there are two ways to find out which of your kitchen implements are dishwasher safe:

  1. Spend hours painstakingly Googling care instructions or looking for little etched symbols, or
  2. Put them all in the dishwasher and see which survive

Marketing segmentation is much the same (albeit a little more gentle 😅).

First, let’s cover the facts:

‍

The more specific and focused marketing segments you can create, the more effective they will be

I was just reading a really great case study from the owners of the Providers app.

For context, the Providers app is a FinTech app focused on serving folks who receive EBT recipients in the US. Because this particular segment of the population tends to be lower income, their user population also largely qualifies for medical assistance in the form of Medicaid.

The US government had implemented some sweeping changes to Medicaid, which caused around 5 million individuals to lose their health insurance, so it was vitally important that the Providers App was able to notify their users and get them to take action as soon as possible to save their health benefits.

‍

TL:DR: They really needed their messaging to connect to their audience. 

‍

No surprise, the Providers team found that push notifications were an incredibly effective channel, and they also discovered some more broad-reaching best practices (for example, using emojis and social proofing in their push notifications increased engagement by 17 and 20%, respectively).

But one finding, in particular, stood out:

‍“With the campaign well underway, Propel analyzed campaign performance metrics and survey data to understand who we were failing to reach. We found that parents with children were viewing our in-app content at lower rates. Developing additional messaging to target and engage this demographic resulted in a 26 percent lift in impressions among this group.”

‍

Here’s what the new messaging looked like:

‍

The second message obviously appeals more strongly to people who have children than the generic message, and the data backs this up. A 26% uplift from a previously non-responsive group is nothing to shrug off.

Tighter segmentation works.

‍

The challenge is, how do we find, define, track, and serve all of these different segments?

That’s the million-dollar question (...literally).

The closer we get to tapping into an individual’s specific situation, motivations, and goals, the more we see a lift in that user’s interaction, but the more specific we get, the smaller the size of our segment becomes.

The smaller our segment becomes, the more segments we need to manage and maintain, and the more segments we need to manage and maintain, the more resources we need to manage them.

‍

Think of it this way: 20 segments sounds like a lot

It’s certainly more segments than most organizations split their populations into, but maintaining 20 segments means 20 different onboarding user journeys, 20 different versions of each email campaign, 20 different messaging/retention strategies, etc. 

Everything you’re doing, you now need to do 19 more times.

But, if we have a population of five million users, having 20 segments leaves us with groups of 250,000 users each. 

  • How specific can we really get with a group of ÂŒ million people?
  • How closely do you think they share interests, dreams, and motivations?
  • Don’t we think we could get a lift if we narrowed this down even further?

Of course, we could! 
but what would that effort cost us in resources?

‍

This is the problem with manual segmentation: The effort/reward payoff

More — and more refined — segments result in better results for each individual segment, but they also require more work to manage.

At some point, the workload becomes unbearable, so we have to give up and settle for the results as “the best that we can do.”

‍

Back to the Providers app — 

They were able to find the “family segment” (perhaps through combing through survey data, demographic data, inference, or by some other means) and develop custom messaging for them, but there were likely more segments they could have built optimizations for, such as:

  • Elderly individuals
  • People with disabilities
  • Recently unemployed individuals
  • Gig workers and freelancers
  • Immigrant families
  • Young adults aging out of foster care
  • Individuals experiencing homelessness
  • People with chronic illnesses
  • Students and recent graduates
  • 
and more.

‍

Writing more focused messaging for each of these segments would likewise have produced higher results for that segment, but again,

  1. Finding which users fit into each of these segments is incredibly difficult.
  2. Building custom user journeys for each of these segments and subgroups is a massive undertaking.

So, what’s the solution?

‍

Let your users segment themselves

You don’t know which segments are the best fit for each of your users — They do.

So, we need to build a system that essentially lets your customers segment themselves in a way that’s intuitive and intelligent. 

‍

Beware of the survey trap

To be clear, I’m not talking about surveys.

Surveys have notoriously low response rates, and they lack one critical feature: adaptability over time.

For example, many of the segments we discussed above (e.g., “Student,” “Unemployed,” etc.) are transitory, meaning that people who are students today may not be tomorrow. Likewise, individuals who are unemployed may become employed the following week.

Building segments with surveys is inherently risky for these reasons.

‍

Our approach to user-led segmentation


Ok, back to the dishwasher tweet.

Instead of digging through every single cabinet and drawer in your kitchen to find out whether each individual spoon, fork, egg slicer, spatula, and dish is dishwasher safe, you could just put them all in the dishwasher and see which survives.

‍

What does this practically look like for marketing professionals?

It means you send someone a message (SMS, WhatsApp message, or Push Notification) that appeals to a specific niche and see if they interact with it.

If someone interacts with it, it’s likely that they belong to this segment. If they don’t, it’s possible that they’re not in this segment.

Then, you lather, rinse, and repeat with the next niche message, and then the next one, and the next one, and before long, you have a decent idea of which users belong to which “segments.”

We call this concept “Message-Led Personalization” or “Message-Led Segmentation.”

‍

Disclaimer: There are a few things this method of segmentation doesn’t account for

To be clear, this approach can get you 60% of the way there, but there are several legitimate concerns and optimizations that can make a huge impact on the results.

For example,

‍

There are almost unlimited niches and combinations of niches

Not only can people be split into demographics, but these demographics can be split even further by psychographics and even things like the preferred tone that messages use (Some people, like me, appreciate puns and dad jokes. Others, like my wife, do not. 😅).

We need a system that can handle this level of complexity to understand which impact each individual element of a message has on any particular user.

‍

Because someone clicks or doesn’t click one time doesn’t demonstrate a strong preference

Sometimes, people don’t click a message because the timing is wrong or they accidentally dismiss it. Other times, they click completely by accident.

We need a system that learns from patterns and sees how those clicks align with propensity-driven expectations, so the question isn’t “Did they click?” but “Did they click when we expected them to?”

‍

Also, we don’t want to over-message users

You don’t really want to have to message each user a message about each niche topic.

Using AI to cluster users based on demonstrated commonalities in behavior before messaging allows you to learn a small subset of a cluster from messaging. You don’t have to test every single user if you can learn from messaging a few representative users.

‍

We need to stay adaptable

As we said above, this can’t just be a one-time thing. We need to periodically test other motivations — even ones that haven’t worked before — to understand if our user’s situation, wants, or needs have changed.

Because they do. Often.

‍

“What if we send a user a message they don’t like?!”

By far, this is the most common objection we get to this approach, so I feel it deserves its own section.

My answer: Surprise! That’s exactly what you’re doing right now. 

Most companies like to believe that every one of their messages scratches the itch of every one of their users (well, they’re typically more realistic than that, but in general, they think most messages more or less hit the mark), but between the segment analysis I did above, the fact that most messages have a 90%+ “not click thru rate,” and the data we’ve collected from over 28,000 messages we’ve stored in our message library, I can confidently say that this method for segmentation is among the lowest risk methods for segmentation you can possibly do.

‍

How to implement user-led segmentation yourself

User-led segmentation is something you can do on your own. It won’t account for all of the concerns in the disclaimer section above, but again, it’ll get you 60% of the way there or more. 

Here’s how you do it:

  1. Think about all of your different possible user motivations and niche segments.
  2. Write messages for each of those motivations identified in Step 1.
  3. Message all of your users with each of those messages in a controlled way.
  4. Use your analytics to see which users interacted with each message
  5. Repeat this process periodically

Compare these results to your current segments, and you might just be surprised to see what you find!

‍

The easy button

If you want advanced user-led segmentation out of the box, add Aampe to your MarTech stack.

User-led segmentation is exactly how our platform operates, and it includes controls and safeguards that account for each of the concerns in the Disclaimer section above. Most companies who switch to Aampe are up and running in a couple of weeks and seeing results within a week after implementation — without having to change their existing MarTech stack.

If you’d like more information or you want to get started with user-led segmentation for your org, smash that bright orange button below.

‍