As data science and AI technologies become more integrated into everyday business processes, more and more companies are seeing their tremendous benefits. But there are challenges with successfully deploying AI in a manner that drives measurable business outcomes. I wanted to share with you some of the best ways to address those challenges and help your data science program succeed.
For more than a decade, numerous companies have expanded their business intelligence resources, investing heavily in analysis of past customer behavior. That investment enabled them to develop rules-based recommendation engines, create better-performing sales strategies, segment consumers for marketing and product design purposes, and so on.
GET UNLIMITED ACCESS TO 140+ ONLINE COURSES
Choose from a wide range of on-demand Data Management courses and comprehensive training programs with our professional subscription.
These types of descriptive and retrospective analysis have become table stakes for any growth-minded company – whether public or privately owned. Many even started dipping their toes into machine learning and AI with the goal of creating competitive differentiators in their product.
To truly fuel business growth, we have an opportunity to go further by marrying data analytics and data science to predict future customer behavior. That can be accomplished without requiring companies to build massive data science teams – or possibly without building a data science team at all.
I realize this might be a little controversial, but please hear me out.
Several business-first use cases can propel a company forward and give companies that haven’t yet benefited from machine learning and AI the competitive edge they need. These use cases capitalize on historical data by using AI to create predictions of individual customer behavior using propensity modeling. These models can assess future customer lifetime value (LTV), forecast conversion likelihood, calculate cross-sell/upsell likelihood, or predict future churn. These types of business-level predictions open the door for much smarter and more efficient ways to grow, while reducing customer acquisition and retention costs, and improving the customer experience.
You probably know the stats around how many data science projects fail to deliver results. Gartner has estimated that 60-85% of data science projects either don’t make it to production or don’t deliver expected results. Most data science platforms tout the millions of models they’ve enabled building, but shy away from sharing how many of those models deliver value for their customers.
One of the biggest reasons for this lack of results is a disconnect between the data science objectives and the actual business objectives that will drive more revenue. Data science projects can uncover unique ways to use data, but applying those methods to support specific business objectives can be challenging. Models may end up being intellectually interesting but ultimately unprofitable for the business. It isn’t uncommon for some data scientists to focus months on optimizing the statistical metrics for a model (after all, that’s what many of us trained to do in our master’s and doctoral programs) – forgetting the goal of fine-tuning our models to drive business value.
And, of course, hiring data scientists – especially good ones – is itself a significant challenge! It’s no wonder that companies large and small are nervous about the requirements and investment needed to make data science work, even when they know how vital AI could be to their growth.
The good news is that companies don’t need to start from scratch to implement data science today. The field isn’t the same as a decade or even five years ago.
It’s actually possible for your business intelligence and analytics teams in sales, marketing, or customer success to harness the superpowers of data science. Analysts from these business teams are better equipped than you might expect for certain kinds of data science applications, such as generating actionable predictions for common customer churn and conversion-related use cases.
To help you get ready to implement predictive analytics in your business org, consider the following questions:
Define the business problem you’re trying to solve.
- Ask yourself, what is the business challenge that needs special attention? Is churn higher than you think it should be, even though your NPS score is high? If customers say they like your product, why are they leaving?
- Is there an opportunity to optimize acquisition and retention efforts if you could find out which customers to focus on to generate the lift you’re looking for? What if you could accurately identify VIP customers much earlier in their customer journey, say after just two or three purchases, and then identify many more just like them before they made any purchases? Or optimize sales outreach with lead scoring? Or even increase cross-sell opportunities by pinpointing the right customers to call? Once you have these answers, how could this capability change your business?
- What business actions will be taken based on the output of those models/predictions? Among other possibilities, those predictions might take the form of a numeric score for the likelihood a customer will churn or a predicted dollar amount for a customer’s lifetime value. The concept of using predictive modeling might be new to some marketing, sales, and customer experience teams, especially when it comes to feeding these predictive scores directly into CRM, CDP or marketing automation tools. However, this is really not so different from how these revenue teams already use automated lead qualification to route leads, or how they might use customer behavioral data in their CDP to orchestrate a Facebook or email campaign for a specific segment. The key difference is that the AI-based predictive scores are much more accurate, so actions orchestrated by those systems will be better targeted and more precise.
Review and define your data requirements.
- Once you’ve reviewed and scoped out the business challenge you’re trying to solve, you should have a good sense of the inputs driving the behavior and what sort of data you’ll need to uncover the answer.
- Don’t sweat the data you don’t have. Many companies spend years gathering and cleaning data before even starting on data science. But when you know the specific challenge, and the inputs feeding that challenge, you can focus on the actual data that will solve the problem, without having to gather and prepare every possible source you have access to.
- If you have transactional or event-based data you’ve captured in the past, chances are you already have the data that you need to look to the future.
Empower your data analysts to create actionable predictions.
- Look for people in-house or from the outside who understand the business conditions and rules around your current BI or other data-focused projects. The trick is to find someone familiar with the data. The ideal person would know where to locate data relevant to your business questions, how the data is structured, how different parts of the data connect to each other, and what the data’s individual components represent for the business in the “real world” (for example, whether a “customer” included in a certain database is an active customer). This person also needs to know what the team’s desired business outcomes are. You will probably have more success training members of your BI team than trying to hire outside resources, but with the right tools, you can often utilize business/marketing/sales analysts to create actionable predictions by focusing on specific questions. Just ensure that the team has a direct stake in your company’s overall success or the specific KPIs of the business team they are supporting.
- Make sure you don’t get into the trap of doing data science as a big experiment with unclear outcomes. Your projects should focus on propelling the business forward. You can build for long-term success if you arm your BI and business teams with solutions that automate many steps of the data science process, including data prep, model creation and scoring. These may not have been available just a few years ago, but they are now. Data scientists today use automation packages to streamline their projects, but automation and thoughtfully designed interfaces are now also making the data science process accessible for those without advanced training or the ability to code in Python or R.
- Ensure all teams and resources are aligned on the requirements, timelines, data sources, testing and actions for this program. Regardless of how you organize and staff your team, keep everyone focused on the business objective, the specific KPIs you’re trying to achieve. Data science can uncover all sorts of insights, but make sure you’re focused on insights that will move the business forward now with knowledge of the future.
When I look back at the evolution of business, and how business analytics have played such an important role in strategy and planning, it’s hard to imagine living today without at least one or two dashboards showing the health of your business. Those dashboards have their place. But when it comes to orchestrating the customer journey, there’s an opportunity to graduate from dashboards and rigid business rules built on a handpicked selection of data points from the past.
AI, machine learning, and data science allow us to look to the future instead of focusing on the past. We can create specific predictions about how individual customers are likely to behave. Machine learning algorithms can look at far more factors and data than humans could ever possibly analyze, finding patterns that aren’t skewed by humans’ “best guesses” and biases. AI can uncover valuable, complex patterns in your data that you might never think of.
Deploying data science can feel like an insurmountable problem. But I can tell you that it’s possible to use your analysts’ skills and the data you already have to ease your way into a data science program that can transform growth and business efficiencies in just weeks. The time is now!