For any Australian SaaS business, recurring revenue is the lifeblood of the company. Predicting which customers will renew and which are at risk allows your Customer Success team to intervene early, protecting your bottom line and improving your valuation.
Building a renewal prediction model isn't just about complex algorithms; it’s about understanding the 'digital body language' of your users. By identifying patterns in usage, support tickets, and payment history, you can move from reactive firefighting to proactive revenue management.
What You’ll Need
- Historical Data: At least 12–24 months of subscription data (renewals and churns).
- CRM Access: Access to Salesforce, HubSpot, or similar tools.
- Product Analytics: Data from tools like Mixpanel, Pendo, or Amplitude.
- Spreadsheet or BI Tool: Excel, Google Sheets, or PowerBI/Tableau.
- Customer Feedback: Qualitative data from NPS surveys or support logs.
---
Step 1: Define Your Goal and Timeframe
Before touching any data, you must define what a 'renewal' looks like for your model. Are you looking at monthly subscriptions or annual contracts? Most Australian SaaS companies find the most success by aiming for a 90-day lead time. This gives your team enough time to reach out and resolve issues before the contract expires.Step 2: Centralise Your Data Sources
Data silos are the enemy of prediction. You need to pull data from three main areas:- Firmographics: Company size, industry, and location (e.g., is an Australian SME more likely to churn than a global enterprise?).
- Product Usage: How often do they log in? Which features are they using?
- Financials: Payment history, any credit card failures, or price increases.
Step 3: Identify Key 'Churn Indicators'
Look back at customers who didn't renew last year. What did they have in common? Common indicators include:- The 'Ghost' Factor: A 50% drop in login frequency over 30 days.
- Unresolved Support Tickets: More than two 'high priority' tickets open for over a week.
- Champion Departure: The primary contact person has left the company (check LinkedIn or CRM updates).
Step 4: Normalise Your Data
You can't compare a customer with 100 users to a startup with 2 users without 'normalising' the data. Instead of looking at total logins, look at Percentage of Active Seats. This ensures your model doesn't unfairly flag small clients who are actually highly engaged.Step 5: Assign Weighted Scores (The Health Score)
Create a 'Customer Health Score' by weighting different behaviours. For example:- Product usage (40% weight)
- Support ticket volume (20% weight)
- Invoice payment speed (15% weight)
- NPS/Survey results (25% weight)
Screenshot Description: You should see a spreadsheet or dashboard table where each customer row has columns for these metrics, ending in a calculated 'Total Score' out of 100.
Step 6: Segment Your Customers
Not all customers are equal. Segment your model by 'Tier' (e.g., Startup, Growth, Enterprise). An Enterprise customer might have a complex renewal process involving procurement, whereas a Startup might just be a credit card hit. Your prediction model needs to account for these different 'buying journeys'.Step 7: Build the Baseline Model
Start simple. Use a Logistic Regression approach (easily done in Excel or PowerBI). This looks at the relationship between your independent variables (usage, tickets) and the dependent variable (Renewed: Yes/No).Step 8: Validate Against Historical Data
Run your model against last year's data. If your model predicts a customer would have churned, did they actually churn? This is called 'Backtesting'. If your accuracy is below 70%, you need to adjust your weights in Step 5.Step 9: Integrate with Your CRM
A model is useless if it sits in a spreadsheet. Feed the 'Health Score' back into your CRM (HubSpot or Salesforce). Create a custom field called 'Renewal Probability' that updates daily.Step 10: Set Up Automated Alerts
Configure your CRM to trigger a task for the Account Manager when a score drops below a certain threshold (e.g., below 40).Tip: In an Australian context, if you are an ABN holder dealing with government contracts, ensure your model accounts for the specific fiscal year-end (June 30) as budget cycles often dictate renewals more than product usage does.
Step 11: Implement a Feedback Loop
When a customer does churn, ask the Account Manager to log the 'Primary Churn Reason'. Use this qualitative data to refine your model. If 'Price' is the main reason, usage metrics might be less predictive than financial metrics.Step 12: Review and Iterate Quarterly
Markets change and software evolves. New features might change what 'healthy usage' looks like. Review your model every quarter to ensure the variables you are tracking are still relevant to the current user experience.---
Pro Tips for Success
- Don't overcomplicate: Start with 3–5 variables. Adding 50 variables usually leads to 'noise' rather than clarity.
- Watch for 'Zombie' Accounts: These are accounts that pay but don't use the software. They are your biggest churn risk at the next renewal cycle.
- Executive Buy-in: Ensure your leadership team trusts the data, otherwise, they won't allocate resources to the 'at-risk' accounts your model identifies.
Common Mistakes to Avoid
- Ignoring Seasonality: Many Australian businesses slow down in December/January. Don't mistake a holiday break for a churn signal.
- Data Lag: Using data that is 30 days old is too late for a 90-day renewal window. Aim for real-time or weekly data syncs.
- Over-reliance on NPS: Net Promoter Scores are 'sentiment' based and can be fickle. Always prioritise 'behavioural' data (what they do) over 'sentiment' data (what they say).
Troubleshooting
- Model is predicting everyone will renew: Check if your data is 'imbalanced'. If 95% of people renew, the model might just guess 'Yes' every time. You may need to focus specifically on the 'Churn' dataset to find unique signals.
- Missing Data: If you don't have product usage data, start with 'Last Login' and 'Support Ticket Frequency' as proxies until you can implement better tracking.
- Too many false positives: Your threshold might be too high. Try lowering the 'At Risk' score trigger to avoid 'Alert Fatigue' for your sales team.
Next Steps
Now that you have a framework for your renewal prediction model, the next step is to automate the outreach. You might want to look at our guide on 'Automating Customer Success Workflows' or 'Mapping the SaaS Customer Journey'.If you need help connecting your CRM data to a predictive dashboard, our team at Local Marketing Group can help you streamline your Revenue Operations. Contact us today to discuss your data strategy.