Leadspace Blog

The Art of Interference in Predictive Analytics

Yaniv Eytani

Lead Data Scientist

Categories: Technology

Predictive analytics drive more effective sales and marketing efforts. When you add predictive to your sales and marketing stack, you gain the ability to rapidly identify, objectively evaluate, and confidently pursue new market opportunities. But “predictive” is more than a widget to jam into your sales machine, it’s more than just an item you either have or you lack. Predictive analytics is an entire field of knowledge, which means many different approaches to execution exist. As a data scientist, I know some approaches are more straightforward to build, and some approaches are more effective in practice. I’ll wager a marketer or salesperson who wants to improve their performance cares more about what’s effective than what’s easy.

In particular, as a marketer or salesperson, you may encounter risks with predictive analytics solutions built on black-box data models. What’s a black-box model? One where you get no knowledge of and have no ability to interact with its internal workings. I’ll admit it’s easier to build predictive algorithms in a closed, black-box system, where the machine learning just looks at historical data. However, black-box models can miss some important elements, which reduces their ability to predict your next buyer.

In particular, machine learning algorithms need a lot of data to learn from to build an accurate predictive model of your customer. I’ve seen many organizations’ customer data, and it’s common for companies to have sparse data sets. Indeed, your business may lack enough historical data to build a good predictive model of your ideal buyer. Perhaps your customer data set is thin on good example customers, or it lacks the customer personas you want to pursue in your sales and marketing efforts. What happens if you build a black-box predictive model on it anyway? That model will come out partly blind.

Chart_Three-points_Linear-trendline
With sparse data, you can get oversimplified predictions.

 

What to do in this situation? Go beyond the black box. Do you have an idea of who your ideal buyer is? Do you have experience in your field, and with your customers? Humans beat machines when it comes to making decisions on sparse data. So you want to find a predictive model that’ll listen to your insight as well as the data. The combination of your data and your intuition about your ideal buyer makes for the most robust predictive algorithm.

Chart_Six-points_Polynomial-trendline
With insight, you get better predictions.

 

As a data scientist, my role is to leverage your domain knowledge and business understanding to connect the dots between your organization’s sparse data points. This practice helps important patterns emerge from the background noise, which spur machine learning algorithms to build the most realistic predictive models. Combining hard-edged statistical models with subtle human intuition is no easy task. Indeed, despite strong market interest, I’ve heard few in our community even attempt this feat.

I’m proud to say my team at Leadspace has built a unified framework where your insights and our analysts’ expertise integrate into the statistical model. Which means our predictive analytics platform powers the most coherent, complete lead discovery, scoring, and enrichment system for B2B sales and marketing teams.