Predictive analytics and AI
What is predictive analytics for a small business?
The short answer
Predictive analytics uses your historical data to estimate what is likely to happen next for each customer, lead or product: who will cancel, who will buy again, which leads will close, how much stock you will need. A business needs a clear outcome to predict, a few hundred past examples of it, and a way to act on the scores.
Key takeaways
- Predictive analytics scores what is likely next: churn, repeat purchase, lead quality, demand.
- You need a defined outcome, a few hundred past examples and a way to act on the score.
- Centralising customer data is usually the first project, and it pays off on its own.
On this page 4 sections
Common predictions
| Prediction | Data it learns from | Action it drives |
|---|---|---|
| Churn risk | Past customers who cancelled and their behaviour before | Retention offers or calls to at-risk accounts |
| Lifetime value | Purchase history of past customers | Bid more for high-value customer types |
| Lead score | Leads that closed and those that did not | Sales calls the best leads first |
| Demand forecast | Past sales by product, season and promotion | Order the right stock |
Do you have enough data?
- A defined outcome: "cancelled within 90 days" is usable; "unhappy" is not.
- Enough examples of it, typically at least a few hundred.
- History that sits in one place, such as a data warehouse, or can be joined.
- A way to use the score: in your CRM, email platform or ad audiences.
If customer data lives in disconnected spreadsheets, getting it into one place is the first project, and it pays off for ordinary reporting too.
Ways to get there
Spreadsheet forecasting handles simple seasonal demand. No-code predictive platforms connect to a warehouse, build and validate the model and push scores back into business tools. Pecan AI is one: it connects to warehouses such as Snowflake, BigQuery, Redshift and Databricks and is aimed at business teams without data scientists. Hiring or contracting a data scientist suits unusual problems or very large data.
Questions
- Is predictive analytics the same as AI?
- It is one use of machine learning. The models predict a number or a probability from patterns in historical data; they do not generate text.
Next step
Compare the tools for this job
Related guides
Build or buy a predictive model?
When a no-code predictive analytics platform makes sense, when to build your own model, and what training your own model costs in people and compute.
Product analytics vs web analytics: which one do you need?
The difference between web analytics (sessions and sources) and product analytics (users and events over time), with examples of the questions each answers.