Predictive LTV Walkthrough
This guide will focus on how to access and understand your predictive LTV data in the Lifetime Value Report. For a full walkthrough of standard Lifetime Value Report features, please take a look at our LTV Report Walkthrough article.
- How to access your predictive LTV data
- How the predictive LTV model works
- What your projections are telling you
- How report settings work with predictive data
- The "LTV Predictions" section
How to access your predictive LTV data
All predictive modeling is integrated directly into the Lifetime Value Report, which you can always access by clicking “Lifetime Value” in the top banner of your screen.
Just below report settings (but above the report matrix) is a switch labeled “Predictive data”. When you click this switch, your LTV report will fill out the rest of your matrix with predictive LTV data.
Before getting deeper into the report, let’s take a quick look at how these projections are generated.
How the predictive LTV model works
When you toggle on the “Predictive data” switch, our Predictive LTV Model kicks into action. The goal of the model is to accurately project the future buying behavior of every individual customer. But how are these predictions made?
Essentially, the model takes whatever transaction data is available for your current customers, and then finds past customers that match on factors such as products purchased, reorder frequency, spending patterns, and more. Then it uses these matches as a baseline to predict not just how much each customer will spend in future sales, but when these sales are most likely to occur. What you end up seeing in your LTV report is all of these individual projections added up to the cohort level and broken out into a monthly forecast of the average accumulated sales each cohort will drive.
Let’s return to the report to see it in action.
What your projections are telling you
To get a better sense for what exactly the model's projection represent, let's zoom in on a few cohorts.
In our example report above, here is the cohort of customers who placed their first orders in December 2020:
The date range of the report shows the past six months by default, which in this case is October 1, 2020, to April 1, 2021. This means that we have 4 months of historical data from the December 2020 cohort (December through March, represented by the columns with headers 0 through 3). But when we toggled on Predictive data, two more cells were added to this row. What do these two cells represent?
- The first predictive cell ($89) shows the LTV model's projection of the average accumulated sales for this cohort by the end of April, which is four months after the cohort's first orders.
- The second cell ($92) show's the LTV model's projection of the average accumulated sales for this cohort by the end of May, which is four month after the cohort's first orders. Note that both of these cells projections are for months that are beyond the Oct 1 to Apr 1 date range of the report.
Now let's look closer at a newer cohort. Here are customers who made their first order in March 2021:
Because we only have data from the same month in which first orders were made (or "0" months since the first order), the majority of this row is projections of LTV.
- The first predictive cell ($76) shows the LTV model's projection of the average accumulated sales for this cohort by the end of April, which is the first month after the cohort's first orders. Note that even though this cell is a projection for April, it's in a different column than the Dec 2020 cohort's projection for April.
- The last (or fifth) predictive sell ($96) shows the LTV model's projection of the average accumulated sales for this cohort by the end of August, which is the fifth month after the cohort's first orders. Again, note how much further this project extends beyond the Oct 1 to Apr 1 date range of the report.
How report settings work with predictive data
Many report settings will work the same whether or not predictive data is enabled, but some settings have some important differences and/or considerations when looking at predictive data.
In our example above, we saw that a default report with a time window set to the last six months completes your LTV report matrix so that you see historical data and/or projections of LTV for each cohort up to six months. This means that for a default report, the projection that's farthest in the future is five months after your newest cohort's first orders.
But what if you want to project even further out? Your instinct might be to select a time window that extends into the future, but keep in mind that the LTV model needs at least first order data for a cohort in order to run its projections. If it's April 2021, the LTV model can't generate predictions for a May 2021 cohort.
However, the model can project out as far as 24 months! To see longer projections, you just need to choose a time window that reaches further back. Compare the projections here by a report from Oct 1, 2020 to April 1, 2021 (top) and April 1, 2020 to April 1, 2021 (bottom).
The end date of both reports is April 1, but the second report gives you projections out to 12 months after first orders for each cohort. For the March 2021 cohort, you can see LTV projections all the way out to February of 2022!
Perhaps the best way to think about this is that predictive data will fill in all the white space of a standard report. Therefore, to see more projections, you need to start with a report with more white space, which can only be done by selecting a longer time window (or by organizing cohorts by week of first order).
Because the predictive LTV model works on an individual level (projecting future sales for each individual customer), you can still choose to organize your cohorts by week, month, quarter, or year, and your projections will adjust accordingly.
Currently, the metrics that integrate with predictive data are:
- Accumulated sales per customer
- Accumulated gross margin per customer
- Cohort sales
- Accumulated sales per customer / CAC
- Accumulated gross margin per customer / CAC
We're working to integrate the rest of these metrics as soon as we can!
Because it's April 1, April sales projections for each cohort will be in the first predictive cell in each row:
To view April sales projections for earlier cohorts as well, you would just need to choose a longer time period.
To add up May sales projections, you would just need to look at the second predictive cell in each row. For June projections, you'd look at the third predictive cell in each row, and so on.
Again, because the predictive LTV model projects future sales for each individual customer, all filters can be applied as with a standard LTV report, and you can use these to analyze different customer segments at any level you choose.
CAC payback tool
CAC payback projections will be enabled as soon as the "Accumulated gross margin" metric is added to the model.
The "Predicted LTV" section
Below the report is a section called "Predicted LTV":
These figures show you average accumulated sales across all customers in the report for periods of 3 months, 6 months, 12 months, and 24 months. The numbers you see here are often a blend of historical and predictive data.
In the above example, an average customer included in the full date range of the report either generated or is expected to generate $115 in accumulated sales in the 12 months following their first order.
The calculations here are meant to zoom out of cohort analysis to give you a more aggregate view of the average values of your customers in the short-term, medium-term, and long-term.