It’s like clockwork: When times get tough, companies send more messages.

Promotions. Sales. Discounts. Offers. Whatever you want to call them, they all start hitting users' email boxes, phone screens, and even mailboxes in a desperate attempt to drive more sales.

The latest data from Omnisend showed that the number of promotional SMS messages sent by eCommerce retailers increased by 376% from 2020 to 2021 and then by another 75% in 2021. Similarly, the number of push notification campaign sends increased by 260% in the 2020-2021 time period.

Here’s the thing, in the short term, this increased volume does drive more revenue. 

…and that’s what makes this so dangerous.

Higher message volume doesn’t just drive conversions — it also drives your customers away.

Per research from VWO Engage, 62% of users already felt they were receiving too many push notifications (and this was from before the big lift of 2020 and beyond), and research from Helplama found that sending five or ten notifications per week boosts unsubscribe rates to 30%+.

Again, a “Turn ‘em and burn ‘em” approach can produce revenue in the short term, but it has effects that go well beyond the walls of your app:

Is this how you want your customers to talk about your brand to their friends and networks?

Even if users don’t uninstall your app, turning off your notifications greatly reduces the value of your existing customers

Research from Custora found that order frequency is the KPI that has the largest impact on the top-line growth and revenue of eCommerce companies—it has over 2x the impact of Average Order Value (AOV)—so it’s clear that to grow effectively, you need to keep your customers engaged and buying over time.

Graph source:

As it stands, a shocking 21% of users abandon an app after one use, but users who have push notifications enabled stay for at least nine sessions (with 46% of them staying beyond the unofficial retention point of 11 sessions).

Conversely, almost half of the users who don’t have push notifications enabled leave after only two sessions.

Your most valuable users have their notifications enabled, and driving users to disable them has a huge negative and lasting impact on your most impactful growth metric.

So, if sending more messages isn’t the right answer, what is? 

Per research from Sajith Pai, VC at Blume Ventures, ~5% of Zomato users account for at least 45% of its total orders and likely over 50% of its total value.

Why does he feel this way?

“Because their [power users’] AOV (avg order values) are higher than the infrequent orderers (as confirmed in the Q1FY23 transcript).”

And this story isn’t uncommon: The majority of the revenue for most businesses and apps comes from a relatively small percentage of “power users.”

The question is, how do you identify these power users…and how do you make more of them?

The first question is more straightforward (You can look at order volume, overall value, etc., and track those metrics back to particular users), but creating more “power users” is much more difficult.

At Aampe, the way we increase the number of “power users” is through dynamic personalization: 

The better we understand what a customer wants (in terms of the way we speak to them, when we speak to them, how often we speak to them, and what we speak to them about), the more likely we can turn them into a power user.

For example, here’s a dashboard from one of our largest eCommerce customers (Note that the “Very High,” “High,” “Still Exploring,” “Low,” and “Very Low” categories denote how well our model understands the timing, frequency, copy, and offering preferences of each user):

(If you want to learn more about how we construct our control groups, you can find more info here.)

From the top row (titled “Messages Distribution”), you can see that, on average, only 20% of our messages are in the “Very High” category. This means we’re sending messages that we’re really confident that people want, and these messages are averaging around a 1.7% revenue rate.

(Note: This isn’t just a measure of clicks—these are actual conversions.)

On the other hand, about 40% of our messages are in the “Still Exploring” category, which means our model is more or less sending messages randomly in an attempt to learn more user preferences. These messages are only yielding a 0.6% revenue rate.

So, we’re sending twice as many messages by volume but accomplishing less than half as much with them with regard to revenue.

To put it into easier numbers, if we sent 1,000 “Very High Confidence” messages and 2,000 “Still Exploring” messages, we’d get 17 conversions from the “Very High” messages and only 12 from the “Still Exploring” messages.

We get 42% more conversions with 50% fewer message sends.

**Note: Despite the complexity, it doesn’t take long to start seeing results with this approach. Most apps start to see results of copy personalization within the first few days and timing personalization within the first couple weeks of using Aampe.

…and, since we’re not burning users out with increased message volume, these results only improve as our model continues to learn more and more users better and better.

One more thing: You can’t A/B test your way to these results

The app we discussed above has over 1 million users and sends out almost 500,000 messages each day. 

A large percentage of each of these messages is unique and features different offerings and incentives. They’re also written in different tones with different copy, and they’re all sent at all different times, with each of these variables determined by factors such as a user’s app activity, historical actions, and demonstrated interests, as well as the actions of “similar” users (systematically clustered by relevant shared traits).

There’s simply no way you can accomplish a feat like this by “comparing the performance of Message A to Message B.”

To quickly and efficiently achieve this level of personalization at scale, we employ a mix of data science tools, including reinforcement learning and bandit algorithms (More on that here), to dynamically learn each individual user’s preferences and then serve them the appropriate message from our customers’ messaging catalog at the appropriate timing and frequency.

In addition to radically increasing key KPIs with these methods, our customers have also reported significant increases in their teams’ efficiencies (up to an 11x productivity increase) by not having to manually build segments, create user journeys, etc.

If you recognize the diminishing returns that ever-increasing message volume brings, and you’d like to learn how we can help you use these tools to create lasting results by increasing your ratio of “power users,” please reach out to

Cover image credit msgrowth on Freepik