Zohar Bronfman is the CEO and co-founder of Pecan.ai, a predictive analytics platform built to solve business problems.
getty
In part one of this series, I explained the broader challenge businesses face in using years of data on past performance to inform AI-level predictions and analytics. That challenge coincides with a shortage of data scientists and the need for better approaches to reaching, engaging, acquiring and retaining customers.
Much of the business intelligence (BI) domain is stuck in the 2000s, not yet embracing the advancements and capabilities of data science. But today, companies want to progress from business logic analysis to AI-driven analytics that can identify patterns that are hard to see in a dashboard or spreadsheet.
To start leveraging their BI data and make the leap to AI predictions, companies should consider these steps toward maximizing their data and their team’s potential.
1. Start with the question in mind. Focus on which business needles you want to move and have a concrete understanding of how you’ll use the model’s predictions to make that happen. If a customer hasn’t purchased something recently, it’s important to incentivize them to do so, but finding the right combination of incentive and channel can be tricky. This is a problem that AI can help solve. For example, data analysts can employ a predictive model-based scoring system to automate the identification of customers who may respond to a bigger discount and be retained longer. That capability means you can predict and shape future customer outcomes with a deeper understanding of what makes each customer come back.
2. Don’t stress about “perfect” data. A new data project can require weeks of validation and data preprocessing. If you have business analysts on your team who employ tools like Looker and Tableau, you probably have plenty of data for them to analyze. You don’t need to make sure every data point is accounted for. You can use the data you already have to create predictive analytics. How? Use your BI-ready data, which means the data is already in a state in which you can drive classic analytics, and select a predictive analytics solution that automates time-consuming data preparation to create an AI-ready dataset. It could save you months of data preprocessing—before any feature engineering can take place and a single model is created.
3. Design A/B tests to validate the accuracy of predictions. An A/B test is one of the fastest ways to try out new changes and approaches like predictive modeling, and it can give significant results with a relatively small sample size. Once a model has been developed, the effects of using it should be tested against a control group that’s handled with your business-as-usual methods, such as business rules you may have been using to determine offers provided to customers. If you don’t test how the model integrates into your business processes and compare its impact to a control group, then you don’t know for sure if the model will produce your desired business outcomes.
4. Enrich the data you have. Your internal transaction and customer data is an ideal starting point for predictive analytics. Although it’s not necessary, some businesses also benefit from enriching their data with external data sources, such as weather, holiday and public health data. Automating that enrichment is one of the fastest ways to ensure additional data streams continually add value to your models’ accuracy and usefulness.
5. Plan for model monitoring and retraining. There’s a common misconception that machine learning models become better over time completely on their own. The opposite is actually true—models typically have a short shelf life. They’ll work great for a while, but their performance will change over time as your business enacts different strategies based on the models and as customer behaviors shift. Many companies that hire data scientists find that they also need dedicated machine learning operations (MLOps) teams who handle model implementation and ongoing model management. However, to save time and resources, automated solutions can monitor and retrain models so they continue to deliver high performance and business impact.
Predictive modeling has incredible potential for transformative business impact. One of the most powerful ways to make this shift is to make your BI data more useful. This approach will help add sophisticated capabilities without needing to spin up a completely new data science team.
If you can empower the BI team you know and trust, you can bring the impact of state-of-the-art data science methods to your business without adding more data science resources. Then, your company can build a bridge between BI and AI, automate data science processes and reap the rewards of predictive analytics.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?
