Picture this: you’re creating a new mobile app that suggests places for people to get dinner – they can order in from a restaurant or make a reservation through your platform.
You’ve spent a lot of time and energy collecting information and building out your user base. You’ve crowd-sourced all sorts of information about the restaurants. You’ve asked the users about their preferences upon sign up, and you’ve even done a variety of social media integrations so that your users can give you permission to collect even more data to provide context to the types of food and restaurants your users will be interested in.
I’m in town in the city you live in, in your neighborhood, for an event, and I’ve been using your app – in fact I downloaded your app through the free wifi in the hotel the last time I was in town. Since I signed up through a social media login, and I checked in through your app the last few times I ate somewhere, I’ve given you lots of data on me that allow you to ascertain the following:
Where would you say I should go to dinner? Maybe think about your top 3. In fact, go ahead and list them out:
1.
2.
3.
You probably suggested a trendy pan-asian or hip new american place not too far, but maybe an uber ride away, preferably with an extensive artisanal cocktail menu. And maybe sometimes that would be the best suggestion.
But here’s what you couldn’t know from all the previously collected data on my profile…
The right answer, and the suggestions I would have been most excited about would be the quickest in-and-out ramen shop (or noodles in general) which is probably not what you guessed (although I did get spicy tonkatsu with an aged marinated egg and pork belly).
Is there any way that you could in fact have created a better app for me in this case? It turns out, as I was browsing in this pseudo-made-up example, a lot of my behavior might have indicated that I wanted ramen: I looked at several different “soup” and “ramen” restaurants, looked at a couple restaurants that are definitely not gluten-friendly, I zoomed in and reset the boundaries to be within two blocks, and I even tried to filter suggestions by wait time. Ok so, maybe through a series of complex data analysis, you could figure out there is a “sick condition” and build out for this specific situation, but generally speaking:
People behave differently at different times and want different things than they did before.
Using somebody’s past browsing history, their purchase history, and certainly a not-so-frequently updated profile as the only means for personalizing recommendations means you are missing out on that fact. We find that the last 3-4 things we track on a user’s behavior (on-site or in-app) are by far the strongest signals to use for predictions.
When discussing this at a recent mobile marketing conference, one of the attendees said that she could see people needing to know recent behavior for things like food, but for her business with makeup and beauty regimens, people don’t seem to change their buying habits much.
While that might be true for the majority of users, there are probably conditions where that isn’t the case (seasonal changes, different color palates, etc.), or where the person is buying for somebody else. Also, with small purchases, people tend to be more willing to experiment, but not if you don’t surface different options. We’ve seen completely different browsing, discovery, and purchase habits from the same person on a fashion website when she is buying for herself then when she is buying for her husband and kids.
You are truly limiting yourself if you don’t pay attention most to the actions your user is taking right now, and then work backwards to fill out the details based on past history. It doesn’t mean that your historical data is obsolete, but someone’s behavior at that moment is much more indicative of their interests and should be weighed appropriately. Even more concerning is that with some types of personalization, you may be setting rules that completely limit the content and marketing to what someone has already expressed interest in, thus further denying your customer the option of an impulse buy.
Additionally, what if the purchases made online for your brand are substantial and in-frequent? Purchase history will probably not tell you much, so leveraging behavioral information on customers who looked at the same things or researched the same content on your site could be a great way to identify those who are more interested in making that big commitment.
So back to the ramen – how would you arrive at this fact? Well, the best way to figure this out is to:
- Leverage Permitted User Data
- Collect Recommendation Data
- Collect Behavioral Data (most important)
- Ask for Real-time Recommendations from your personalization service
If the personalization service you are using is good, it will take all this information, crunch the numbers, and come back right away with that tasty ramen place down the street.
