When we first started helping companies by making predictions, the process for determining what to predict involved quite a bit of effort. We’d start by becoming users of their apps and websites. Next we’d take a deep look into their data, often times mapping our own explicit usage to events recorded in the data, in order to get a more thorough understanding of what events correlated with various in app actions. Finally, we’d call the companies and talk to them about their customer pain points, their business models and their goals.
Then we’d get to the actual work of making the predictions.
The process was tedious, it took a long time, and each new customer consumed a huge amount of effort from the entire team. We knew there was a better, more efficient, way to go about making predictions. The fact is that working through each new customer this way helped us learn. As we onboarded each new customer we learned a little bit faster; which feature sets made sense across verticals, which algorithms returned better results in which cases, which customer behaviors were important across business models and across industries.
In the end, goedle is a startup. The point of any startup is to learn as quickly as possible, by doing things that don’t scale.
One recurring theme that we couldn’t ignore any longer was our customers asking us to make it possible for them to set up their own predictions, to set their own goals.
With an even mix of curiosity and trepidation we quietly rolled the capability out to our customers a couple of weeks ago. Fairly quickly, one of our customers discovered the new feature and made the first self-service prediction on our platform. We were ecstatic! High fives all around! And then everything broke.
We had made the assumption that, systemically, a person would never be able to qualify for more than one prediction on any given day for the same goal. There were lots of logical and computational reasons for this, implications for marketing engagement campaigns, and so on. Regardless, we realized pretty quickly that we were wrong about this.
That issue aside, we were surprised by what our customers actually started creating. In the earliest predictions that we set up for our customers, we were predicting things that equated to core KPIs within a business, things like purchase conversions, magic number conversions, subscription cancellations, churn in freemium products, and so on.
Once our customers were able to begin creating their own predictions however, we observed that some customers were starting to create them not only for their core KPIs, but also for various aspects of their entire funnel. In effect, our customers were instrumenting their entire customer journey with goedle’s predictive analytics. This has major potential for improving marketing automation campaigns and is something we’ll touch upon in a future post.
Our customers are the experts at their businesses; they understand their KPIs, their funnels and their customer journeys. What will be really interesting to observe over time is whether what they believe is important to their business actually is. Or, will our algorithms help them discover that the customer behaviors that they believe to be important, aren’t necessarily so? We don’t have enough data for that yet, but we’ll certainly continue to look at it.
And so we keep learning.
If you’d like to set your own goals for your business, sign up now and connect your data to goedle.