TAGS

Predicting consumer behavior is inherently difficult, primarily because, well, there are humans involved! At the same time, being able to accurately predict what your customers are going to need or want solves a host of problems before they even happen. With something this difficult, how do you make it work and what’s even worth trying in the first place?

Where to start with predictions?

No matter what type of business, trend watching and customer behavior predictions are always going to serve a purpose. The big question isn’t should you do it, it’s how. The amount of direct vs. indirect selling involved in your sales channels is the first thing to consider. How much direct interaction does the average customer have with a representative of your brand during the major touch points of their purchase process? How much of this same interaction is occurring through your website or social media or other online channels? Whether an individual member of your sales team or your website is going to be primarily responsible for doing the prediction is going to have a significant impact on what your approach should ultimately be. If much of this prediction should be occurring on your website, then developing a database of customer information and using it to identify trends in purchase and search patterns is going to offer the largest benefit. It’s important to remember though that the ability to identify trends from a set of data is dependent on there being sufficient data for those trends to be reliable. Algorithms like Amazon and Netflix use are remarkably effective and accurate precisely because there has been such a huge volume of traffic through their services, and their predictions and suggestions are able to continually improve because they’re always acquiring more data. You don’t have to be doing the volume of business that Amazon does for your data to be sufficient for a certain amount of automation, but it’s important to have a realistic idea of how much data you really have access to.

Get the right data, and get enough of it

One way you can potentially improve your pool of data is to combine online customer activity with in-store or other offline activity, by requesting customers’ email addresses or some other form of primary identifying information during offline purchases or other interactions. It requires customer buy-in, because you’re asking them to do something that is not actually required for whatever interaction you’re involved in, so it’s important to be able to clearly articulate what the added value is going to be for the customer. Better product suggestions online? Improved customer support by allowing representatives to look up purchase histories without needing credit card numbers or receipts? Access to a complete purchase history so they can easily remember what size/brand/model number they bought last time when they need to make a repeat or related purchase? Whatever it is, be able to succinctly explain it to a customer who may be hesitant.

For companies that engage in more direct selling, especially for B2B sales, manual CRM input is going to be more useful. The biggest challenge for manual input data is remembering to make sure that every piece of data from every part of any customer interaction gets entered. The resistance here is more likely to be from employees than customers, as all that data input is of course time consuming. Yet again, it’s important to be able to articulate exactly what the benefit is for your team. Easier support interactions? Better sales conversion numbers? Whatever it is, same as with getting buy-in from customers, be able to explain it clearly, because the buy-in is critical. Whether you’re using machine learning or are manually analyzing the data, the data needs to be there, and the data has to come from the work of your team.

Demographics are not enough

For all this emphasis on getting enough data, it’s also important to remember that bare demographics are not the data we want. I don’t care about the age or gender or race of the customer nine times out of ten. What I care about is what products have they already purchased? What have they searched for? What products have they already asked questions about? Information that’s about them, not generally about millions of people who may be classified as like them in the next census. You’re looking for “customers who searched for this item you recently searched for/purchased your recent purchases were also interested in… “ and not “people from your ZIP code/people of your gender and age liked these things.” This is about affinity, the things that customers like and have chosen, not demographics, something that we’ve written about recently.

Be the signal, not the noise

Once you’ve acquired your data and have identified some trends, the next question is how to pass along that information to your customers. If it’s going to be automated via email blasts or suggestions on your website, know that it has to be accurate a significant portion of the time or it’s just noise to the customer. If you get it right, you’re making it more personal and are helpful, but if you get it wrong, it’s spam, and the customer may just stop paying attention to your direct marketing entirely. Personalized predictions coming from an actual human being, whether via email or phone, have more of a margin for error, both because your sales staff can adjust on the fly as necessary and because personal attention almost always makes people feel valued. With any kind of predictions, whether it’s with purchase or service ideas or automated phone or computer systems, it’s critical that they be suggestions and not decisions being made for a customer. Give them a way to say “nope, that’s not right” or opt out, or when the inevitable outlier comes up, they’re going to be incredibly frustrated. Even if your predictions are amazingly accurate and well-crafted, there are going to be a handful of customers who differ and you have to give them an escape valve, or risk alienating them at the moment when you’re trying to draw them closer.

Be the problem solver

Finally, know that sometimes “predicting” is really more about identifying a problem that is generally indicated but not spelled out plainly by customers and coming up with a solution, rather than pulling something that a customer wants seemingly out of thin air. Apple eventually began to offer complete device set-up at the time of purchase of their iOS based devices is a prime example of using customer comments and behaviors as a way to identify a need that was perhaps incompletely expressed by customers and finding a solution. After seeing many customers return for help with things like setting up email accounts or customizing their new device, Apple identified that the solution to the customer need that was being presented on those return trips was actually a need that could be solved at the beginning of the ownership experience. A reactive solution to some customers (helping them with set-up and configuration when they return after purchase) can become a proactive solution to the next round (including device set-up as a standard part of the purchase process). There was no Amazing Kreskin style prediction involved, just clearly identifying and addressing a need as early in the customer experience as possible.

Predicting customer behavior can make your life less stressful by seeing trends ahead of time that you can prepare for, can make customers feel attended to by having customized options or recommendations, and is most likely something you can implement to start out with using your existing data even if you’re not doing it yet. Remember that these are human beings and there are always outliers, and go forth! But maybe stop by the comments and let us know about a particularly wonderful or particularly terrible experience with attempts at predicting customer behavior first. Your best Amazon suggestion? Your worst automated phone system guessing game? Knock us out!