Real talk about real-time CDPs

People are talking about real-time CDPs.  People are lying about real-time CDPs.  Others are claiming that real-time CDPs are as mythical as unicorns.

Let’s dispel  the myth, because Lytics is a real-time CDP and I’ll walk you through what you need to be looking for.

But first let’s make sure we’re on the same page about what real-time is and what a CDP does.

What does real-time mean?

In our industry, we’ll call a platform real-time if things are happening in under a second, end-to-end.

That is, in a real-time CDP, a customer does something and you can immediately do something about it.

That is, in a real-time CDP, a customer does something, which sends an event to your CDP, which then does a “bunch of processing”, and then your CDP triggers other actions in complementary systems – such as: present offers, send emails or push notifications, sync to ad tools, update your CRM.  

Most critical is that all of these steps happen in under a second.

That’s a difficult thing to do.  But it’s not impossible.

What does a CDP do?

The part of this that might be impossible is navigating the CDP landscape and deciphering what every CDP does.  It’s even tricky for the analysts who try to corral and classify the different types of CDP.

When we say the CDP does a “bunch of processing”, I like using David Chan’s list of his required capabilities in a CDP pipeline:

1. Collection
2. Transformation
3. Identity Resolution
4. Business Rules
5. AI/ML Modeling
6. Segmentation
7. Activation

If any of those take longer than a second, you’re no longer doing things in real-time.  That’s quite a task, so it’s worth going a bit deeper on each one.

How to go fast

(As a side note for the geeks, Lytics’ pipeline is written 100% in pure Go and is blazing fast.  You can hit me up if you want to geek out about it.)

1. Collection

It’s true that not all data collection mechanisms are real-time.  Your CRM likely isn’t.  Your data warehouse definitely isn’t.  But most of your other (modern) tools are: web, mobile, email, etc.  When plumbed to Lytics, your data’s real-time journey begins.

2. Transformation

Transformation is a key component of making systems talk to each other, and has been the “T” part of ETL for 50 years.  If you’re acting on real-time data, it’s because transformations have enabled systems to be interoperable, and like Lytics’ transformations, those transformations had better be quick.

3. Identity Resolution

This is the second-most likely place where a real-time pipeline slows down.  But it’s also the most important part of a real-time pipeline.

Stitching on-the-fly requires your data model to be sophisticated enough to handle fast connections to be made, like Lytics’ identity graph model.  Having a graph model (over a traditional relational model), yields both better performance and better accuracy.

Better performance because connections in a graph can be built on-the-fly, versus relying on relational database joins to populate connections. Real-time relational joins are never going to happen.

Better accuracy because native and flexible mechanics for building a graph, called graph traversal, allow for adding or removing connections, limiting the number or type of connections between identities, or augmenting identity with AI, all on-the-fly.

Having real-time identity resolution that is both performant and accurate opens up a world of high-value use cases, like fraud detection, that are otherwise infeasible.

4. Business Rules

Your business is unlike any other business, so your rules are not going to be exactly like any others’ rules.  Applying your business rules to your data means that your customers always see your fingerprint in their experiences.  Flexible, powerful business rule building, like Lytics’ powerful language for calculations and aggregations and conditional logic, are a must when facilitating sub second responses. 

5. AI/ML Modeling

This is the most likely place where real-time pipelines slow down.  This happens for so many reasons, but principally among those are that data scientists aren’t thinking about real-time.  As a data scientist  myself, I can attest that they’re usually (and thankfully) focused on accuracy, using the tools they know how to use.  And let’s face it, you will not get real-time performance out of any models built in R or Python (data scientist favorites).

In a world with real-time ML, you can personalize the web experiences of even anonymous visitors during their first (and likely only) session from the results of custom predictive models, like those built in Lytics.  We do that billions of times every day.

It’s possible because we’ve optimized our AI/ML pipeline for speed while remaining uncompromising on accuracy.  We even author open-source libraries when we find that things don’t meet our performance standards.

6. Segmentation

For many platforms, segments are either completely static or are batch based, either of which are a death knell for real-time.  Having real-time segmentation would mean two important things:

1. Membership for segments (both entries and exits) would be updated in real-time.  Ideally, you could even subscribe to these events and could build applications with them.
2. For a given profile, you would always know all of the segments that they were a member of at exactly the time you ask for it.  The segments would be available everywhere the profile is.

In Lytics’ real-time CDP, both of these are true.

7. Activation

Real-time identity resolution is an absolute requirement for real-time activation.  If you don’t have real-time identity resolution, your profile activation is either stale or inaccurate, negatively impacting your customer experiences and your match rates.

And if you can’t activate an accurate profile in real-time, there’s really no point in having a real-time pipeline.  Real-time technologies, like webhooks and triggers, are a must for a real-time pipeline.

As you’ve probably guessed by now, Lytics has all of that too.

Take it for a spin

Want to experience what a real-time CDP feels like?  Let’s talk.  Get a demo of real-time in action, and then try out one of our 30-day POC’s.  It’s free.

In all honesty, it does seem a little too good to be true.  I can see why the skeptics are so skeptical.