Skip to main content

What is predictive analytics? Learn how forward-looking AI can grow your revenue


What is predictive analytics?

Ever wish you had a crystal ball to guide your marketing? According to new research, 74% of consumers expect personalised experiences from B2C brands.

You don't need a psychic on staff. Predictive analytics uses your customer data and AI to estimate what individual customers are likely to do next, down to the profile level, so you can make each experience more relevant, improve engagement, reduce churn, and build loyalty.

What is predictive analytics and how does it work?

Predictive analytics is one of 4 main types of data analytics. It helps businesses recognise patterns in their data and anticipate what comes next. While diagnostic analytics looks backward to find the cause of data trends, predictive analytics uses AI and machine learning to estimate how the trends and relationships in today's data are likely to play out.

Predictive analytics isn't a crystal ball, exactly. The predictions are based on the data you already have, and they aren't guaranteed to come true. But using historical data to estimate likely scenarios supports more informed decisions.

3 key benefits of predictive analytics

Predictive analytics supports three goals most marketers share:

  • Predictive analytics can improve profit margins. Refine your discounting and promotion strategy and move away from blanket discounts. Tailor discounts and incentives to a customer's past experience and expected future behaviour.
  • Predictive analytics can reduce churn. Predict whether a customer is likely to purchase again, when, and how much they're likely to spend, so you can reach out before they drop off.
  • Predictive analytics can build customer loyalty. You already know which customers love you. Predictive analytics can also help identify who is likely to love you in the future, such as customers with a high predicted customer lifetime value (CLV), so you can invite them into loyalty programmes and ask for referrals and social proof.

Predictive analytics use cases for B2C

Predictive analytics gives you new metrics to gauge the health of your customers and your business. Each one answers a different question, so it helps to know what each can tell you. A marketing agent can act on them: describe the audience, and it builds the segment and drafts the campaign for your review.

Average time between orders

The average amount of time that passes between a customer's orders. Use cases:

  • Compare the average time between orders for specific segments of your customer base.
  • Group customers with similar buying cycles into segments, then offer incentives to customers who buy infrequently.

Average order value (AOV)

The average amount a customer spends each time they place an order. Use cases:

  • Compare the AOV of your VIP customers with the AOV of other segments.
  • Segment customers by AOV, then run targeted initiatives that raise the AOV of customers who don't spend much.
  • Branch your flows by AOV, and encourage customers who usually place small orders to add more items to their cart.

Predicted number of orders

An estimate of how many more orders a customer will place in the next 12 months, shown next to their historic order count. Because of how it's calculated, the estimate can be a non-whole number. Use cases:

  • Build your own loyalty framework from the number of orders customers have placed.
  • Build segments from the number of orders customers are likely to place to spot who is likely to purchase again and who is not. Trigger flows from those segments, or retarget customers who are likely to buy again with ads.

Expected date of next order

The date a customer will probably purchase again if nothing changes. Marketers can influence it. Use the date for personalisation, together with other properties, because the same date means different things for different customers:

  • For a customer who buys on a steady cycle, hold back messages before the expected date and invite them into your loyalty programme instead.
  • For a customer who buys rarely, send A/B tested campaigns before the expected date to try to bring the purchase forward.
  • For a customer whose expected date has passed without a purchase, send content that builds the relationship, and check whether an issue with their last order went unresolved.

Customer lifetime value (CLV)

Three related numbers describe a customer's value. Historic CLV is what a customer has spent over their lifetime, predicted CLV is what they're expected to spend in the next 12 months, and total CLV is the sum of the two. Use cases:

  • Compare the CLV of your VIP customers with the CLV of other segments.
  • Build segments by CLV, then use them to find new customers through lookalike audiences in ad platforms.
  • Retarget customers with a high predicted CLV through ads.
  • Branch flows by predicted CLV to deliver tailored offers and incentives.

Churn risk prediction

A percentage showing how likely or unlikely a customer is to make another purchase. Most customers across most industries are one- or two-time purchasers, so most of your customers will likely have a churn risk above 50%. A lower percentage means a higher likelihood that the customer will purchase again. Use cases:

  • Compare the average churn risk between groups, such as your VIPs and your entire customer base.
  • Use churn risk together with the expected date of next order to decide who needs a win-back campaign and who is a loyal customer who would respond to regular communication.

Cross-sell date

The time a customer is most likely to be ready for a cross-sell offer, based on the purchasing behaviour of similar customers. Use cases:

  • Trigger a flow from someone's cross-sell date to try to bring their next purchase forward.
  • Branch flows by cross-sell date so customers who are already likely to purchase don't get an unnecessary incentive.
  • Create on-site forms for visitors whose cross-sell date is coming up, and present a deal designed to raise AOV, such as a bundle or a buy-two-get-one-free offer.
  • Personalise emails, texts, or push notifications with dynamic content based on the cross-sell date.

Likely next product

The product a customer is most likely to buy next, based on their most recent purchase. It updates automatically as the customer places more orders. Use cases:

  • Personalise emails, texts, or push notifications with dynamic content based on the likely next product.
  • Sync audiences to ad platforms to serve retargeting ads for the likely next product.

How to avoid biased predictive analytics

Predictive analytics needs a healthy amount of data to find patterns and trends accurately. That means enough customers who have placed orders, enough order history, and some customers who have placed three or more orders.

Predictions are only as good as the data behind them. A full view of each customer's behaviour over a long period gives a model more to work from. If your data sits across point solutions and isn't consolidated in near real time, your predictions reflect only the data they can see. Connect your data sources and keep your history so predictions draw on the whole picture.

Get started with Klaviyo, the agent-first operating system for B2C, and put your predicted segments to work with a marketing agent.