May 7, 2025
At some point, shifts in your customers’ behavior become glaringly obvious. Be aware of these shifts well before then and address the shifts proactively, not reactively. Intervene before your competition does.
If you’re an e-commerce company with frequent reorders, you’ll want to focus on:
If your product catalog is not highly structured, please refer to methods from our post Use AI to Assess Competitors and Supply Chain Daily.
Even when data is extremely structured, it is still helpful to develop a substitution map, as correctly identifying substitutions may require a bit more flexibility than only using item categorizations. For example, consider disposable foot warmers, woolen socks, and electric socks; meaningful price changes to one may cause a utility company to switch to one of the others for their linemen, though they may each be catalogued under a separate product category.
The fastest way to kick off a substitution map is to run crude analyses and then have knowledgeable humans verify the output and suggest methodology improvements. At the customer level, for example: (1) identify SKUs that are being ordered in lesser quantities and SKUs that are being ordered in higher quantities (2) match them using similarity scoring methods, and (3) have humans review the matchings and also suggest improvements to approach.
The next step for enhancing a substitution map is incorporating search and browsing data — let your shoppers tell you which products are related based on which they evaluate against each other.
In some cases you may be able to detect clear shifts in purchasing patterns by simply looking at cart sku count, quantity per sku, order frequency, and other straightforward calculations. It may be helpful to initially focus on the most critical sku’s and analyze them manually from all angles.
If these simple methods are yielding clear signals, it’s likely that behavior changes are already in advanced stages and that there is a lot of opportunity to intervene earlier with customers as additional analyses are incorporated.
In certain cases, particularly industrial ones, repurchases can be described using mathematical models such as the Poisson process. When these models are applied appropriately, a handful of base parameters are derived that succinctly describe a customer’s probabilistic behaviors around reordering. Comparing these parameters across different time periods can give a much stronger signal of change relative to basic methods. Aggregate your results over many customers to find the most important trends affecting your top line.
Substitution analysis can be further improved by overlaying search, browsing, and cart modification behavior that preceded the purchase. See later sections for more.
If pricing is fluctuating, make sure to track pricing history and then derive simplified vectors (up/down, a lot/little) to describe recent changes and overall change. Price fluctuation data can enhance all of the above models. For example, if we think that a customer is experimenting with a substitute, our confidence in that theory will increase if there is a large coinciding price increase.
Hopefully you’re already tracking user activities such as page views, button clicks, searches, etc using some sort of event stream (ex: Segment).
A very simple place to start is changes in aggregate behavior prior to checkout. For example, maybe a customer that was reliably only using the reorder tool is suddenly instead using site tools to browse before ordering. Another simple case: maybe a customer is comparison shopping for longer before ordering new items.
The next step may be to break activity sessions down into subsessions, where each subsession contains all of the events related to exploring or/and purchasing a single type of item. The subsession can then be characterized by duration, number of actions, outcome, etc.
A leading indicator gives the earliest possible sign that an important event is likely to occur in the future. The classic example is for churn (a lagging indicator): sophisticated companies monitor customer behavior for early indicators that a churn is coming, and then intervene before that can happen.
As you iterate on the above approaches to identifying customer behavior changes that warrant intervention, you’ll decide which are the earliest and most accurate signals to use.
Many of the above tools and methods are advisable to implement regardless of tariff environment. Is a particular product being tried once by customers but then usually substituted with another? It would be good to understand why and then correct the problem. Is a customer suddenly exhibiting markedly different browsing behaviors or reorder patterns? Something has probably changed with personnel or process and an well-timed reachout will have high payoff.
