Here’s what I learned this week

Predictive LTV - The Holy Grail of Marketing

I read a post about someone trying to crack it, I know smart people working on it, we’ve been working on theories around how to approach it.

If someone actually figured it out, you’d know about it.

The truth is, I don’t think it exists, not in the way that people are claiming.

3rd Party Data, 2nd Party Data, 1st Party Data all being combined with Pixels to try to figure out what you’re likely to purchase next and when.

Amazon has the biggest advantage of anyone in this game, they have your purchase data, your address or multiple addresses, your browsing history on the website, your saved cart items, and they have an advertising network that can help companies get products in front of you that you’re likely to be in the market to purchase.

It’s why they own so much of the ecommerce market.

There are companies trying to apply the same logic to their websites, trying to predict when someone will come back and shop again or what they might like.

Everyone thinks they can figure it out via observed data.

The problem is no one store will ever have enough data to be able to make accurate claims around this.

I had this debate with Tobias over at Ocurate.

We talked over lunch one day about his theory and their idea that it could be predicted, don’t get me wrong, I love talking to people with PHDs in big data from Stanford that headed up Facebook’s political influence research.

I think he’s correct in saying that for a lot of larger purchases and large life decisions like politics this data would be incredibly valuable to gauge timing, for smaller purchases though I find it really hard to believe.

My counter argument is always the same: swimsuits.

My need for a swimsuit can change from lifestyle changes that are not captured directly by any 1st party data. I might be going on a vacation, I might have left a swimsuit at someone’s house far away, I might have got a gym membership at a place with a pool and decided to take up swimming, I might be traveling during winter to the southern hemisphere, I might have been drunk and fell down ripping my existing suit, or the elastic might have given out because it’s so old, or I might have gained or lost weight, or just didn’t like the way any of them looked anymore.

Because we don’t live our lives in platforms, not every waking moment, not every experience, not every interaction is fed into a system for these platforms to analyze.

There’s no way of really accurately predicting this stuff. It’s what makes us as humans so unpredictable. It’s what makes people turn on brands in a heartbeat, it’s wonderful.

Everyone has been trying to figure out the God Tier power of knowing not only that someone is likely to purchase and how likely but also their likelihood of purchasing again.

Predictive LTV before the first sale is God Tier power.

The problem is, really smart people, even smarter than me, with more specialized degrees have been working on this for a while and well turns out no one has really cracked this bit.

There was a thing with Target back in the day that did some damage.

You can read the article here: How Target Used Data Analytics to Predict Pregnancies

Now this works when you sell lots of products, it’s not the same for your normal DTC store.

We’ve been working on this at Formtoro for years now because I like a challenge too, but I cannot confidently say that we are anywhere near cracking it.

Here’s what we can tell you:

Based on data collected I can tell you the following about someone based on historical data (not entirely predictive of future behavior but all we have to go on).

When someone signs up we get this information:

[email protected]
Shopping for themself.
Interested in Men's Activewear.
Primarily for exercising.
What matters most is comfort.
Looking to purchase in a few days.

Came from Facebook Campaign tcamp101
Came from Facebook Ad Set aud54
Clicked on Ad ac01c85lp806

People that answered similar:
Average 3 items in their cart.
Typically have t-shirt in their cart.
Have a first purchase AOV of $67.
Convert 28% of the time, typically within 7 days.
They repurchase 15% of the time within 90 days.

This is pretty darn good, but it’s not perfect, but it’s representative of where we are currently with all the technology we have available to us. Everyone is guessing. We’re just using structured data relevant to the intent to guess a little more conclusively.

My personal goal has always been to get a print out of the number of subscribers and their data and source and be able to create a statistical modeling of the revenue they will create over a time period.

How do you do anything close to this without zero party data? I have no idea if you’re a smaller store.

I keep hearing about these magical algorithms that can do all sorts of fairy dust magic and I keep shaking my head because we simply don’t think any of it is really possible. At least not without tons of products and at super scale.

For the average store, we’re yet to see how anyone’s magic pixel or 1st party data alone can lead anyone to make any concrete causal relationships between behavior and purchases happen that didn’t already have some underlying intent that could have been surfaced by zero party data.

I’m also yet to see correlative analysis relevant to revenue, order count, conversion rate correlated with 1st party data in a meaningful way that is structured.

It’s just a black box of an algorithm.

Trust me, I would love for someone to solve this problem, it’s a multi-trillion dollar one, but those that are going to be the best at it are going to be the social networks and they are going to combine with the largest marketplaces to make it happen.

But remember we had Cambridge Analytica, so those day of freewheeling data to the highest bidder are likely behind us.

Notice that news about Amazon partnering with Facebook last week, yeah they aren’t dumb.

This is the only downside of my being part of a team of two building software and tackling a problem that is sizable. We simply don’t have the access necessary to take on this problem for real, the data set and inputs are just not enough.

Our quality of inputs is often higher than a social network can have our intent signals definitely higher than other types of data including 1st party (we use a combination of 1st and zero party), but it’s still not enough to build a financial forecast on.

Do I think Accurate Predictive LTV is currently possible? No.

Am I still going to work on predictive models based on data? Yes.

Do I hope that one day it will be possible? Eh, I don’t think it will ever be strong enough unless we had all our behaviors attached to a registry, I think that statistically we’ll probably get to about a 30% confidence interval which means you’re better off just flipping a coin.

The Takeaway

Don’t believe the hype. I’m all for being proven wrong, but in all the long term tests that I’ve run of more than a year holding variables consistent, I can say that we’re still a long ways off from knowing if a model can work for the average store.

Have a great week!

-Jon