The Marketing x Analytics podcast recently invited Ahmed to talk about data, our activity schema, and why it's so hard to get answers using more traditional approaches.
Listen to the whole thing here or read on for some highlights.
It can be hard to get answers to simple questions
"Let's go with the simple question. How many people that came to our website called us. Should seem relatively easy, right? If you ask it to a data team will tell you, oh, give me like a month.
And the reason why it's really hard to answer that question is that your website data is often in your front end tracking -- Segment or Snowplow, Heap. And your call data is often in your Salesforce or any call center app. Now that data, both of them live in the warehouse, which your data team controls, but stitching the data together is extremely hard. They use different user identifiers – a phone number or a browser cookie or an email address
Another question that also causes a lot of trouble is if you're sending email campaigns and you're like which email campaign led to the order. That also seems like an easy question, but it's extremely hard. It's so hard that almost every single CDP like Amplitude, Heap, Mixpanel, and Segment, what they often do is they kind of hack at it. They say, give me everyone who is part of this campaign and also bought. The problem is that everybody is part of like 30 campaigns because you've tried different things. So it's very hard to know which campaign relates to that order. That question is extremely hard and most tools just kind of ignore it."
Why is user identification so hard?
"Whenever you want to combine data from one system (a web visit) to another (a phone call) you need to match some kind of user identifier. The problem is that a common identifier between two systems never works, never exists. If you try to combine data from a system like a call center to your website or your website to your order system or your order system to your internal database there's no lookup ID. You have to make one. And making one is a thousand lines SQL query, and you often do it wrong and it's really hard. "
How Narrator quickly matches data across systems
"So Narrator decided to change how we structure data. We transform all data from each system into what we call an activity - something a customer did 'visited a web site', the time they did it, and a customer identifier (an anonymous user id, or an email, or a phone number).
So instead of having all these tables and trying to do a lookup, which never exists, we use this customer activity in time to stitch things together. Say I opened an email and then right after that I called you. I can then easily associate these actions across my entire warehouse because they happened somewhat close in time.
And this works when you have a billion rows or a hundred billion rows or trillion rows. So doing that temporal way of relating data is really, really different in data. And getting that to actually work took us several years."
The importance of asking data questions frequently
Narrator's unique approach allows marketers to ask as many types of data questions as the want, without waiting on the data team. This means they can apply analytics to everything rather than rationing their questions to the ones the data team has the bandwidth to answer.
For example, let's say you were curious about the ROI of offering upgraded shipping to certain customers for free. You have a hunch that some people aren't price sensitive about shipping speed, but some people do care.
You could start using analytics to figure out which customers are likely to care about shipping costs. You could quickly find customers who are clicking onto the more expensive shipping then clicking back. What if you upgraded them for free? You send them an email saying: I noticed you tried to change a price. Here's a discount on shipping, or we've waived the fee this time and given you free shipping. And in a few days, if you have analytics, you can actually see that that the customers who were upgraded might actually return a higher LTV. That customer might come back and buy way more frequently because of that shipping upgrade instead of spending the $50 to retarget them using ads.
The point is that every possible user interaction with your business can be explored with data to make better decisions.
How we think about data
"Narrator doesn't ever give you unactionable data like 'women convert twice as much as men'. That's useless. We like to give you stories (which we call Narratives) that help you understand what's happening. Like here's how many orders you're getting. Here's how here's how people were coming, men and women. And here's how they're converting. And here's how it's been reliable over time. And now, if you actually were to shift your targeting to get more of this kind of group, here's what could have happened to your business and here's how much, what you would want to change. And here's how it makes sense. And it's already in English and we'll run this analysis for you every week automatically. So if the outcome starts changing or just becomes not actionable, we'll email you and let you know. And that's the thing that makes analytics still beautiful that it's like a deep connection you have with a billion, random people."
There's a ton more
These snippets only cover a small part of the conversation. Listen to the full podcast to learn a ton more about marketing analytics, how Ahmed got started with Narrator, and some of his philosophy around how to be properly data-driven