Boat sailing on a sea with horizon in sight

Scaling customer marketing

Customer marketing is stuck between unsustainable mass campaigns and unscalable automated triggers. Discover Tinyclues' vision.

David Bessis

CEO

Time to read10 min read
October 29, 2020

Customer marketing is stuck between unsustainable mass campaigns and unscalable automated triggers. But there is a path forward.

This article inaugurates Tinyclues Vision Blogs, a series by the Tinyclues team on the vision, science, and technology behind our clients’ successes.

1. The magic is broken

Any sufficiently advanced technology is indistinguishable from magic. Judging by Arthur C. Clarke’s famous adage, customer marketing is insufficiently advanced.

If it was indistinguishable from magic, then consumers wouldn’t be so annoyed. Marketers wouldn’t be so frustrated. And CMOs wouldn’t have the shortest tenure in the C-suite.

Despite all the hype around AI, the sad truth is that 99% of what people call intelligent marketing is glorified retargeting. A consumer saw a pair of socks on your website. Now he’s categorized as a sock-lover and you are going to chase him with items he “may also like” (but actually hates.)

There goes the promise of intelligent marketing.

Marketers were sold magic.

All they got was workflow automation.

Right after Joe visited your website, it was relevant to retarget him with related content. But this relevance has a very short shelf-life. Joe hasn’t returned in months. What’s your plan? Send him random discounts?

Marketing was never supposed to piggyback on explicit demand. Its core mission is to detect unexpressed demand and to generate new demand.

So what happens if intelligent marketing breaks down when there is no explicit demand? Spam. Spam everywhere. Spam on a massive scale. Personalized triggers and journeys account for less than 10% of customer marketing messaging volume. More than 90% of messages sent by brands to their own customers are mass campaigns that are neither customer-centric nor data-driven.

Joe’s experience sucks because personalized campaigns (in green) triggered by his actions (in blue) do not scale. He mostly gets mass campaigns (in red.)

The most shocking part: this illustration isn’t even an exaggeration.

We all know it and we experience it every day. Four months ago, I bought a pair of shoes from a brand I love. Right after this in-store purchase in NYC, they sent me two personalized messages: the transaction receipt and a nice welcome email. Then a carpet-bombing of batch-and-blast messages, including discounts for products I don’t care about. In four months, they’ve sent me 117 newsletters. None felt personalized.

CRM teams are deeply frustrated by the disconnect between what they experience as consumers and what they practice as marketers. They know it is wrong, yet they keep doing it. Why? Because their business cannot afford to lose the revenue driven by mass campaigns and automation sequences cannot scale and compensate for it.

This frustration is major driver for switching to a new marketing automation solution, in the hope that better automation capabilities will solve the problem.

But automation alone cannot solve the problem and marketers may end up being equally frustrated with the new solution.

How can we fix this? A first step is to recognize the nature of the problem:

  • It isn’t a lack of automation. False beliefs about how much automation can achieve have contributed to the current situation.
  • It isn’t a lack of data. In fact, marketers sit on tons of customer data.
  • The issue is the disconnect between data and actions.

Mass campaigns dominate customer marketing because marketers need reach. So the only way forward is to build and execute a marketing strategy that is scalable yet truly informed by customer insights. As Forrester puts it“a successful solution must bridge the gap between data aggregation and marketing execution.”

A fishing analogy helps make this tangible. Customer data is an ocean: big, deep, ever-changing and mysterious. The fish are the opportunity to engage customers with something relevant. A marketing cloud is a fishing boat with automation capabilities that recognize the fish that are swimming right under the boat. But there aren’t enough fish swimming right under the boat. So marketers end up deploying gigantic nets and blindly trawling the entire ocean. This is brutal, ineffective and unsustainable. Marketers don’t need new fishing boats. They need a sustainable approach to fishing.

The Tinyclues approach starts with a predictive technology that scans first-party customer datasets and maps out all the opportunities in real time. By predicting where their marketing will be efficient, marketers can launch the right campaigns to the right customers. They no longer need to kill baby tunas when they’re fishing for mackerel. They no longer miss out on an undetected concentration of monkfish on the way back.

In short, they can finally build and execute a customer marketing plan that scales.

Marketing success is all about product-market fit. Tinyclues detects hotspots between offers and customers with whom these offers will resonate. It predicts latent customer demand before intent is expressed, at scale, across all offers and customers, based on all available data points.

The dilemma between mass marketing (reach without sustainability) and trigger marketing (sustainability without reach) is a false dilemma: predictive marketing offers both reach and sustainability. Our clients have discovered these fascinating laws of predictive marketing:

More marketing with less customer fatigue.

More marketing that creates more loyalty.

As long as it is well targeted, customer marketing can be much more active than previously thought.

This may sound like magic, but leading global brands have proved it at scale. They apply this magic to sustainably grow their customer marketing, driving revenue, customer loyalty and strategic business outcomes:

  • A top 10 global hospitality group is using Tinyclues to build and execute its global CRM plan: thousands of campaigns are intelligently allocated to millions of customers. The result: a scalable marketing plan that dynamically adjusts to business needs while improving customer experience metrics; Forrester estimates the 3 years revenue impact of Tinyclues at $82M.
  • Three of the top 10 global luxury brands are using Tinyclues to decide when and how to engage their customers on channels from email to clienteling. In an industry so focused on experience, smart decisioning is the only way to scale customer marketing without damaging the brand.
  • In a year, more than 100 companies in retail, travel, luxury, hospitality, banking, betting, media used Tinyclues to make the right decisions on more than 100,000 campaigns and more than 2 billion messages, based on insights learnt from more than 100 billion dollars of past transactions.
CC BY 3.0, SKA

2. Capturing tiny clues in deep customer data

From a technology standpoint, transitioning from lower-funnel intelligence (personalization based on recent intent) to full-funnel intelligence (personalization in the absence of intent) is a pretty big deal.

Personalization based on recent intent is easy

Off-the-shelf “collaborative filtering” algorithms can provide Netflix-style recommendations of new items based on recently seen items.

But most likely your business isn’t like Netflix and your customers aren’t returning every week.

In that case, such recommendations will have very limited value for you and your customers. If Joe doesn’t return to your website, automated recommendations will become repetitive and increasingly irrelevant.

Personalization in the absence of intent is much harder

First, it requires an element of novelty: new products, trends, seasonality, special offers, etc. Sending the right message isn’t enough, you have to maintain a consistent conversation over multiple months, even if Joe doesn’t respond.

For this you will need humans in the loop, as they beat machines at telling stories. Predictive technology will have to co-operate with your marketing team, not replace it. The question isn’t which socks to recommend to Joe based on the socks he saw last week, but:

Could your new beachwear category be a successful outreach topic? And would it be the right occasion to re-engage Joe who’s been silent for six months?

The problem is that predicting inactive Joe’s potential interest for new items is mathematically hard because of a phenomenon known as the curse of dimensionality, that basically asserts that you cannot learn too many things from too few data points. You can learn 3 parameters from 200 data points, but learning millions of parameters requires an unthinkable number of data points.

If you have tons of data about Joe (his first name, age, zip code, what he bought 2 years ago, the items he saw during his visits, which emails he opened and which links he clicked on, the dates of all these events, etc.) but few recent beachwear transactions and no way to directly connect them with Joe, then you are in curse of dimensionality territory, because each customer profile is incredibly complex and unique in its own way and you have no idea which aspects matter.

The secret behind Tinyclues is a deep learning technology specifically designed to predict customer unexpressed demand using 360 first-party customer data, without falling victim to the curse of dimensionality.

Deep learning is the unsupervised machine learning approach behind the recent breakthroughs in image recognition and natural language processing. It imitates how our brains work by extracting the latent information contained in raw data.

What is the latent information? In the context of customer data, latent information is the meaningful signal hidden in the correlations that exist between the different data assets.

Customer data is like a giant multi-dimensional graph (mathematicians call that a tensor) and there are correlations all over the place. Examples:

  • First names and email domains are correlated with zip codes and years of birth (and vice-versa) abut also with product_ids and category_ids (through the transaction table) and link_ids from newsletters (through the email click logs)—this contains latent social and demographic information.
  • Product_ids are correlated with payment methods, days of weeks, acquisition channels, with first, with other product_ids bought by the same people—this contains latent product information.
  • Campaign_ids, link_ids are themselves correlated with everything above—this contains latent information about your content.

Everything is correlated with everything. These correlations contain millions of “tiny clues”, meaningful social and behavioral patterns.

With the right algorithm, it is possible to automatically extract these tiny clues and consolidate them into a high-resolution picture of latent demand patterns across a customer database.

Instead of building impossibly large telescopes, astronomers combine many small telescopes to reconstruct high-resolution pictures of the deep universe. In a way, the Tinyclues approach is the same. It effectively escapes the curse of dimensionality and enables our clients to accurately locate, within a database of millions of customers, the likely next buyers of a product that was sold a few dozen times.

3. From tiny clues to scaling the marketing plan

Because off-the-shelf deep learning algorithms cannot handle the complex relational data structures of marketing, we had to come up with our own mathematical approach. But if building the tech is hard, successful AI is first and foremost about addressing the right user needs and creating a meaningful user experience.

Making marketing heroes starts with understanding marketers. What does their job look like? Not just the official job, but the real job, the day-to-day job. What are their goals? What are their constraints? How much of the marketing plan is actually sourced by marketing and how much is pushed by other stakeholders in the organization? How can we empower marketers to step up and gain influence?

B2C marketing automation leaves these questions unanswered, because automation workflows do not integrate with marketers’ day-to-day jobs. If it really was about automating marketing, then CRM teams would have lost their jobs long ago. But the dirty secret of Marketing Clouds—and the reassuring news for CRM teams—is that triggers and journeys only account for less than 10% of customer marketing messaging volume.

The remaining 90% keep CRM teams busy on a day-to-day basis. This activity is below the radar of automation and doesn’t even have a universally agreed name. It has barely changed over the past 15 years. Some marketers call it business-as-usual marketing, others talk about tactical campaigns, mass campaigns or editorial campaigns.

Despite all predictions that they would go away, these campaigns continue to dominate customer marketing. There is a simple explanation: they solve two mission-critical business problems:

  1. Generate repeat demand from existing customers.
  2. Support the dynamic needs of the business (product launches, inventory issues, strategic cross-sell, commercial priorities, partnerships, etc.)

Marketers do not work in isolation. They build and execute a campaign plan based on inputs from other stakeholders in the organization.

There is a tension between the conflicting necessities of being customer-centric AND business-centric.

This tension is the defining dilemma of modern marketers.

There is no silver bullet and this tension cannot be resolved overnight. We’ve designed our feature set and onboarding process to offer marketers a pragmatic path forward, one step at a time:

  1. The best starting point is the existing campaign plan. Predictive targeting powered by deep learning works much better than rule-based targeting or gut-feel (and of course it works much better than no targeting at all, a surprisingly commonplace practice.) By showing improved metrics for revenue and customer experience from the first weeks of using Tinyclues, customer marketing teams build confidence.
  2. Then they can start adding incremental campaigns. A common misconception is that brands already do too much customer marketing. What they do too much of is indiscriminate customer marketing. If you accurately match topics and audiences, there is virtually no limit.
  3. Here again, the right way to start is from the existing business needs. Is there any campaign request (from commerce or category managers) that was rejected or hidden at the bottom of a newsletter because there was “no more room” in next week’s plan? You’ll magically find room for an extra campaign targeting the right 5% of your customers (those who will account of 80% of revenue for this topic) because it will deliver a great customer experience to them and add substantial revenue.
  4. Action-oriented insights from customer data will provide the inspiration to go further. Tinyclues generates dynamic maps of demand patterns across customers and offers, and identifies where marketing will have high efficiency. This creates an opportunity for marketers to elevate their conversation with other departments. Instead of just executing a campaign plan decide elsewhere, they can finally influence the plan.

In its 2020 Predictions for CMOs, Forrester anticipates a radical transformation of marketing’s mission and predicts that CMOs will gain a broader responsibility for revenue and customer experience because “you can’t build, express, sell, communicate, connect, or service today’s brands without continuity.”

This continuity is what is at stake. By bridging the gap between customer data and marketing actions, marketers can not only deliver on their revenue goals: they also gain an opportunity to step up and drive customer-centricity across the organization.

You might also like